This is the accepted version of the article: Valencia, Enrique; de Bello, Francesco; Lepš, Jan; [et al.]. «Directional trends in species composition over time can lead to a widespread overemphasis of year-to- year asynchrony». Journal of vegetation science, Vol. 31, Issue 5 (September 2020), p. 709-802. DOI 10.1111/jvs.12916 This version is available at https://ddd.uab.cat/record/232894 under the terms of the license Journal of Vegetation Science Page 1 of 147 1 Directional trends in species composition over time can lead to a widespread 2 overemphasis of year-to-year asynchrony 3 4 Running title: Directional trends effects on synchrony 5 6 Enrique Valencia1*†, Francesco de Bello2,3†, Jan Lepš2,4, Thomas Galland2,5, Anna E- 7 Vojtkó2,5, Luisa Conti5,6, Jiří Danihelka7,8, Jürgen Dengler9,10,11, David J. Eldridge12, 8 Marc Estiarte13,14, Ricardo García-González15, Eric Garnier16, Daniel Gómez15, Susan 9 Harrison17, Tomas Herben18,19, Ricardo Ibáñez20, Anke Jentsch21, Norbert Juergens22, 10 Miklós Kertész23, Katja Klumpp24, Frédérique Louault24, Rob H. Marrs25, Gábor 11 Ónodi23, Robin J. Pakeman26, Meelis Pärtel27, Begoña Peco28, Josep Peñuelas13,14, 12 Marta Rueda29, Wolfgang Schmidt30, Ute Schmiedel22, Martin Schuetz31, Hana 13 Skalova8, Petr Šmilauer32, Marie Šmilauerová2, Christian Smit33, Ming-Hua Song34, 14 Martin Stock35, James Val12, Vigdis Vandvik36, Karsten Wesche11,37,38, Ben A. 15 Woodcock39, Truman P. Young40,41, Fei-Hai Yu42, Martin Zobel27, Lars 16 Götzenberger2,5. 17 18 1 Departamento de Biología y Geología, Física y Química Inorgánica, Escuela Superior 19 Revie de Ciencias Experimentales y Tecnología, Universidad Rey Juan Carlos, Móstoles, 20 Spain. 21 2 Department of Botany, Faculty of Sciences, University of South Bohemia, České 22 23 24 25 Budějovice, Czech Republic. w 3 Centro de Investigaciones sobre Desertificación (CSIC-UV-GV), Valencia, Spain. 4 Biology Research Centre, Institute of Entomology, Czech Academy of Sciences, České Budějovice, Czech Republic. 26 5 Institute of Botany, Czech Academy of Sciences, Třeboň, Czech Republic. 27 6 Faculty of Environmental Sciences, Czech University of Life Sciences Prague, Praha, 28 Czech Republic. 29 7 Department of Botany and Zoology, Masaryk University, Brno, Czech Republic. 30 8 Institute of Botany, Academy of Sciences of the Czech Republic, Průhonice, Czech 31 Republic. 32 9 Vegetation Ecology Group, Institute of Natural Resource Sciences (IUNR), Zurich 33 University of Applied Sciences (ZHAW), Wädenswil, Switzerland. 34 10 Plant Ecology Group, Bayreuth Center for Ecology and Environmental Research 35 (BayCEER), University of Bayreuth, Bayreuth, Germany. 36 11 German Centre for Integrative Biodiversity Research (iDiv) Halle-Jena-Leipzig, 37 Leipzig, Germany. 38 12 Biological, Earth and Environmental Sciences, University of New South Wales, 39 Sydney, Australia. 40 13 CREAF, Cerdanyola del Vallès, Catalonia, Spain. 41 14 CSIC, Global Ecology Unit CREAF-CSIC-UAB, Catalonia, Spain. 42 15 Instituto Pirenaico de Ecología (IPE-CSIC), Jaca-Zaragoza, Spain. 43 16 CEFE, CNRS, Univ Montpellier, Univ Paul Valéry Montpellier 3, EPHE, IRD, 44 Montpellier, France. 45 17 Department of Environmental Science and Policy, University of California Davis, 46 CA, USA. 47 18 Department of Botany, Faculty of Science, Charles University, Praha, Czech 48 Republic. 49 19 Institute of Botany, Academy of Sciences of the Czech Republic, Průhonice, Czech 50 Republic. 1 Page 2 of 147 Journal of Vegetation Science 51 20 Department of Environmental Biology, University of Navarra, Pamplona, Spain. 52 21 Department of Disturbance Ecology, Bayreuth Center of Ecology and Environmental 53 Research (BayCEER), University of Bayreuth, Bayreuth, Germany. 54 22 Research Unit Biodiversity, Evolution & Ecology (BEE) of Plants, Institute for Plant 55 Science and Microbiology, University of Hamburg, Germany. 56 23 MTA Centre for Ecological Research, Vácrátót, Hungary. 57 24 INRA, VetAgro Sup, UCA, UMR Ecosystem Prairial, Clermont-Ferrand, France. 58 25 University of Liverpool, Liverpool, UK. 59 26 The James Hutton Institute, Craigiebuckler, Aberdeen, UK. 60 27 Department of Botany, Institute of Ecology and Earth Sciences, University of Tartu, 61 Tartu, Estonia. 62 28 Terrestrial Ecology Group (TEG), Department of Ecology, Institute for Biodiversity 63 and Global Change, Autonomous University of Madrid, Madrid, Spain. 64 29 Department of Conservation Biology, Estación Biológica de Doñana (EBD-CSIC), 65 Sevilla, Spain. 66 30 Department of Silviculture and Forest Ecology of the Temperate Zones, University 67 of Göttingen, Germany. 68 31 Swiss Federal Institute for Forest, Snow and Landscape Research WSL, Community 69 Ecology, Switzerland. 70 32 Department of Ecosystem Biology, Faculty of Science, University of South Bohemia, 71 České Budějovice, Czech Republic. 72 33 Conservation Ecology Group, Groningen Institute for Evolutionary Life Sciences, 73 Groningen, The Netherlands. 74 34 Laboratory of Ecosystem Network Observation and Modelling, Institute of 75 Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, 76 Beijing, China. 77 35 Wadden Sea National Park of Schleswig-Holstein, Tönning, Germany. 78 36 Department of Biological Sciences, University of Bergen, Norway. 79 37 Botany Department, Senckenberg, Natural History Museum Goerlitz, Görlitz, 80 Germany. 81 38 International Institute Zittau, Technische Universität Dresden, Dresden, Germany. 82 39 NERC Centre for Ecology & Hydrology, Oxfordshire, UK. 83 40 Department of Plant Sciences, University of California, Davis, USA. 84 41 Mpala Research Centre, Nanyuki, Kenya. 85 42 Institute of Wetland Ecology & Clone Ecology / Zhejiang Provincial Key Laboratory 86 of Plant Evolutionary Ecology and Conservation, Taizhou University, Taizhou, China. 87 Correspondence 88 Departamento de Biología y Geología, Física y Química Inorgánica, Escuela Superior 89 de Ciencias Experimentales y Tecnología, Universidad Rey Juan Carlos, Móstoles, 90 Spain. 91 Email: valencia.gomez.e@gmail.com 92 †Authors contributed equally 93 94 Funding information 95 Czech Science Foundation (GAČR). Grant Number: 16- 15012S. 96 Program for attracting and retaining talent of Comunidad de Madrid (EV). Grant 97 Number: 2017-T2/AMB-5406 2 Journal of Vegetation Science Page 3 of 147 98 Web summary 99 Measures of community synchrony and stability aim at quantifying year-to-year 100 fluctuations in species abundances. However, these indices reflect also long-term 101 trends, potentially masking year-to-year signals. Using a large number of datasets with 102 permanent vegetation plots we show a frequent greater synchrony and stability in year- 103 to-year changes compared to when long-term trends are not taken into account. For Revie w 104 3 Page 4 of 147 Journal of Vegetation Science 105 Abstract 106 Questions 107 Compensatory dynamics are described as one of the main mechanisms that increase 108 community stability, e.g. where decreases of some species on a year-to-year basis are 109 offset by an increase in others. Deviations from perfect synchrony between species 110 (asynchrony) have therefore been advocated as an important mechanism underlying 111 iodiversity effects on stability. However, it is unclear to what extent existing measures 112 of synchrony actually capture the signal of year-to-year species fluctuations in the 113 presence of long-term directional trends in both species abundance and composition 114 (species directional trends hereafter). Such directional trends may lead to a 115 misinterpretation of indices commonly used to reflect year-to-year synchrony. 116 Methods 117 An approach based on three-term local quadrat variance (T3) which assess population 118 variability in a three-year moving window, was used to overcome species directional 119 trends effects. This ‘detrending’ approach was applied to common indices of synchrony 120 across a Worldwide collection of 77 temporal plant community datasets comprising 121 almost 7800 individual plots sampled for at least 6 years. Plots included were either 122 maintained under constant ‘control’ conditions over time or were subjected to different 123 management or disturbances treatments. 124 Results 125 Accounting for directional trends increased the detection of year-to-year synchronous 126 patterns in all synchrony indices considered. Specifically, synchrony values increased 127 significantly in ~40% of the datasets with the T3 detrending approach while in ~10% 128 synchrony decreased. For the 38 studies with both control and manipulated conditions, 4 Journal of Vegetation Science Page 5 of 147 129 the increase in synchrony values was stronger for longer-time series, particularly 130 following experimental manipulation. 131 Conclusions 132 Species long-term directional trends can affect synchrony and stability measures 133 potentially masking the ecological mechanism causing year-to-year fluctuations. As 134 such, previous studies on community stability might have overemphasised the role of 135 compensatory dynamic in real-world ecosystems, and particularly in manipulative 136 conditions, when not considering the possible overriding effects of long-term 137 directional trends. 138 139 Keywords: asynchrony, biodiversity, stability, synchrony, temporal dynamics, year-to- 140 year fluctuation. 5 Page 6 of 147 Journal of Vegetation Science 141 Introduction 142 Given the challenges posed by rapidly changing environments in the context of global 143 change, it is crucial to understand how biological diversity is maintained over time 144 (Cardinale et al. 2007; Tomimatsu et al. 2013; Tilman, Isbell, & Cowles 2014). There 145 is a general consensus toward the role that synchrony (or lack of) in, e.g., year-to-year 146 population fluctuations between co-existing species plays on species diversity and 147 community stability (Hautier et al. 2014; Craven et al. 2018). On the one hand, a 148 common response to environmental fluctuations (for example changes in temperature 149 or precipitation from one year to another) of most species (synchrony) will tend to 150 destabilize the community biomass or abundance. On the other hand, the opposite 151 pattern (compensatory dynamics, i.e. increases or decreases in the relative abundance 152 of some species that are offset by changes in the relative abundance of others; Hubbell 153 2001; Gonzalez & Loreau 2009) will lead to higher community stability. In this sense 154 asynchrony, i.e. the extent of the deviation from lack of perfect synchrony between 155 species, has been advocated as an important and widespread mechanism that 156 contributes to stability (Loreau & de Mazancourt 2013). 157 While there is a lively debate on the importance of compensatory dynamics on 158 the stability of communities (Houlahan et al. 2007; Blüthgen et al. 2016; Lepš et al. 159 2018) there are also important methodological aspects that can influence the detection 160 of the underlying biological patterns. Recently, Lepš et al. (2019) demonstrated that the 161 study of synchrony between species has traditionally disregarded the possible effects of 162 long-term directional compositional trends in the analysed communities (i.e. a tendency 163 of some species to increase or decrease over time, or to fluctuate cyclically, Wu et al. 164 2007). Species directional trends occur when the abundances of species respond not 165 only to short-term environmental fluctuations, but also to the presence of monotonic or 6 Journal of Vegetation Science Page 7 of 147 166 cyclical tendencies over the whole time series considered. Short term environmental 167 fluctuations (Rabotnov 1974), for example on a year-to-year basis, are expected to 168 affect species abundance but also to be largely reversible, so that species would not 169 show long-term directional trends in their abundances. In contrast, long-term 170 environmental changes, such as climate change, nutrient deposition and changes in land 171 use (e.g. abandonment or intensification of agricultural land), generally cause long- 172 term species directional trends (Stevens et al. 2011; Walter et al. 2018). Long-term 173 directional trends can also be the result of the impact of undetermined drivers 174 (Milchunas, Lauenroth, & Burkeal 1998). As repeatedly reported by many authors, long 175 term trends in species abundance are probably omnipresent, and have been 176 demonstrated even in, now, more than 160 years of the Park Grass Experiment 177 (Silvertown et al. 2006). 178 To gain a better understanding of the underlying mechanisms regulating 179 changes in species abundance, short-term fluctuations and long-term trends effects on 180 synchrony should be disentangled. Unfortunately, this differentiation has been rare in 181 studies assessing drivers of synchrony and stability (but see Vasseur & Gaedke 2007; 182 Tredennick et al. 2017; and the review by Lepš et al. 2019). Indeed, using simulations 183 and simple case studies Lepš et al. (2019) showed that species directional trends can 184 mask year-to-year fluctuations among species. This has the potential to result in a biased 185 estimation of asynchrony when using many widely used synchrony indices. Such 186 directional trends could lead to either overestimation of year-to-year synchrony when 187 the majority of species concomitantly increase or decrease over time, as well as 188 overestimation of year-to-year asynchrony when some species increase and some others 189 decrease over time. 7 Page 8 of 147 Journal of Vegetation Science 190 Multiple indices have been developed to evaluate the level of synchrony among 191 species in a community (Loreau & de Mazancourt 2008; Gross et al. 2014; Blüthgen et 192 al. 2016; Lepš et al. 2018). Further methodologies have also been developed to assess 193 directional trends, such as spectral or wavelet analyses, however, they are applicable 194 only to very long or highly resolved time series (see Lepš et al. 2019 for an overview 195 of these methods). None of the classically used synchrony indices disentangle, a priori, 196 the actual year-to-year fluctuations from the directional trends. However, such indices 197 can be ‘detrended’ using different methods (Wu et al. 2007; Lepš et al. 2019). One 198 appealing a simple solution includes computing synchrony indices over moveable 199 windows of three consecutive years (three-term local variance, ‘T3’, Hill 1973) instead 200 of over the whole sampling period (Lepš et al. 2019). This ‘detrending’ approach, which 201 we call T3 detrending approach, could allow testing the generality of the effect of 202 directional trends on synchrony indices. If the focus of the research is on year-to-year 203 fluctuations, then the minimum number of years to exclude trends and consider yearly 204 fluctuations is 3 years, hence the three-term local variance. With bigger windows the 205 computation of a common linear trend over the time window, and the focus on the 206 deviation from this trend, does recall on the other method proposed by Lepš et al. 207 (2019), using residuals of fitted linear models over a given time period. The first 208 approach has the advantage that it can be computed with any existing index of 209 synchrony and does not require the knowledge of the shape of possible linear trends in 210 species abundance. 211 A widespread assessment of the effect of species directional trends on 212 synchrony has been limited by the scarcity of available long-term data. Indeed, the 213 study of temporal dynamics requires a substantial sampling effort to obtain meaningful 214 data for temporal analyses. Although there are networks and independent groups with 8 Journal of Vegetation Science Page 9 of 147 215 long-term ecological data around the world, no major efforts have been made to compile 216 and standardize the existing data in order to achieve a worldwide perspective. 217 Consequently, a global-scale analysis would improve our understanding of both 218 directional trends and year-to-year species fluctuations among the different synchrony 219 indices and across diverse habitats, as well as how they are related with different types 220 of disturbances or stressors. To face this challenge, we compiled plant community data 221 from 77 temporal datasets with at least six sampling years, including almost 7800 222 vegetation plots distributed across the world. First, we evaluated to what extent year- 223 to-year synchrony could be masked by long-term trends, by using the T3 detrending 224 approach for temporal series proposed by Lepš et al. (2019) on commonly used indices 225 of synchrony. Second, we assessed whether synchrony patterns changed in plots in 226 which initial conditions were maintained (‘control’) vs. plots in which new conditions 227 were applied (‘manipulated’ plots, see methods), assuming that these new conditions 228 would trigger compositional changes and therefore generate a trend. Third, we 229 evaluated how detrended synchrony values are affected by the duration of the sampling. 230 Finally, we asked if relationships that are commonly assessed in the literature regarding 231 synchrony indices, i.e. the correlation between synchrony and species richness and the 232 correlation between synchrony and community stability, changed markedly depending 233 on whether the T3 detrending approach was applied. Additionally, beside the validation 234 of the T3 approach introduced by Lepš et al. (2019), we further validated (using 235 simulations) the functionality of the approach in the case of both monotonic and cyclical 236 long-term trends and depending on the time series length (Appendix S1). We expect 237 that: (1) directional trends in our datasets can overshadow either asynchrony or 238 synchrony depending on the type of trend; (2) manipulative experiments can give rise 239 to directional trends and therefore reinforce the need for detrended metrics to accurately 9 Page 10 of 147 Journal of Vegetation Science 240 evaluate and compare community dynamics; (3) longer time series would provide 241 greater chances to detect species directional trends; and (4) the presence of directional 242 trends may affect the strength of the relationship between synchrony indices and species 243 richness or community stability. For Review 10 Journal of Vegetation Science Page 11 of 147 244 Methods 245 We collected 77 worldwide datasets of aboveground dry biomass, cover percentage, or 246 frequencies of natural or semi-natural plant communities. These datasets consist of 247 7788 permanent and semi-permanent plots sampled between 6 to 53 times over periods 248 of 6 to 99 years. These datasets included plots with different treatments or 249 manipulations. The plots were thus grouped into two categories: control vs. 250 manipulated. In total 38 datasets presented both control and manipulated plots. Control 251 includes those plots where the long-term conditions prior to the establishment of the 252 sampling scheme were maintained throughout the sampling. For example, if the 253 historical conditions in a given site include periodic mowing, this represents the 254 ‘control’. The ‘manipulated’ plots were exposed to different treatments that altered the 255 long-term conditions in their respective sites. These treatments included introduction or 256 exclusion of grazing, mowing, removal of dominant species, fire, fertilization and 257 climate change treatments. These wide categories allowed us to perform broad 258 comparisons between different land-use and management conditions that are expected 259 to influence species trends. The list of datasets, their characteristics in habitat, 260 vegetation type and their available data on location and main manipulations is provided 261 in Appendix S2. 262 263 Synchrony measures 264 For each of the 7788 plots, we computed the most common indices of community-level 265 synchrony from existing literature. The main indices fall into two families. The first 266 one is based on correlations between species’ abundances and includes two indices: the 267 one proposed by Gross et al. (2014) and then this modified by Blüthgen et al. (2016), 268 which weighs the contribution of species to community synchrony in terms of their 11 Page 12 of 147 Journal of Vegetation Science 269 abundance. We call these indices ‘Gross’ and ‘GrossW’, respectively. The second 270 family of indices is based on variance ratios, i.e. the variance in species fluctuations is 271 compared against the null model of independent fluctuations of individual populations, 272 and includes two indices: log variance ratio (‘Logvar’, Lepš et al. 2018) and 𝜑 (‘Phi’, 273 Loreau & de Mazancourt 2008). 274 The Gross and GrossW indices range from -1 to +1 and Logvar from -Inf to 275 +ln(nsp), with nsp being the number of species in a community. High values indicate a 276 common response of the species (synchrony), while any deviation from perfect 277 synchrony indicates asynchrony; the lowest and negative values indicate that the 278 increases or decreases in some species are compensated by opposite changes in others. 279 For all, Gross, GrossW and Logvar, zero corresponds to a situation where the species 280 fluctuate completely independently of each other. Finally, Phi ranges from 0 to 1, 1 281 being perfect synchrony and any deviation from this value means asynchrony. 282 For each plot we also computed the average number of species in the plots 283 across years, as well as the coefficient of variation (CV) of species abundances (standard 284 deviation of the total sum of abundances or biomass across years divided by the mean 285 of abundances or biomass across years). CV of total community abundance is a 286 common measure of community (in)stability, where high values of CV indicate low 287 stability in the community. 288 All measures of synchrony (and the CV) can be computed using the three-term 289 local variance (T3; see Lepš et al. 2019 for an explanation of how to apply this method 290 to the synchrony measures), originally introduced by Hill (1973) in the context of 291 spatial pattern analysis. T3 is then calculated as: 292 12 Journal of Vegetation Science Page 13 of 147 ∑𝑛 ― 2(𝑥𝑖 ― 2𝑥𝑖 + 1 + 𝑥𝑖 + 2)2 𝑖 293 𝑇3 = 6(𝑛 ― 2) 294 295 where n is the number of years in the time-series, i is the year index, and xi is the 296 abundance recorded in year i. Consequently, T3 computes the variance by averaging 297 variance estimates within a moving window of three consecutive years over the data. 298 Any eventual increase in window size needs to be considered with respect to the limits 299 imposed by total length of the series (Lepš 1990). In this context that the minimum 300 length of the time series in our collection of datasets was 6 years, a movable window 301 of 3 years seemed as a reasonable solution. 302 For the three-year window used in the calculations, the variance (which is 303 needed in all existing index of synchrony) is estimated from the squared difference of 304 the middle year and average of the years before and after. Therefore, if there is a perfect 305 linear trend within these three years, the difference is zero. If there is no temporal trend 306 in the time series analysed, then T3 is an estimate of classic variance (i.e. for long-time 307 series without a trend the values of T3 and classical variance will converge; see below; 308 Lepš et al. 2019). For each plot, each synchrony index (Gross, GrossW, Logvar and 309 Phi) as well as the CV were calculated both with and without the T3 detrending method. 310 311 Data analysis 312 To assess to what extent the synchrony indices were affected by directional trends we 313 followed different approaches. First, we correlated (across plots within each dataset) 314 synchrony values with and without the T3 detrending approach. Specifically, for each 315 dataset we retained a Rho coefficient from the Spearman correlation between indices 316 calculated using the T3 detrending approach and their respective indices calculated 13 Page 14 of 147 Journal of Vegetation Science 317 without the T3 approach. Then, to test consistency across datasets another Spearman 318 test was run on the average of each synchrony index per dataset to test if the ranking in 319 synchrony between datasets was maintained. 320 Second, we determined in how many datasets the T3 detrending approach 321 significantly increased, or decreased, the synchrony values. For this we ran a series of 322 paired t-tests, with a correction of the resulting p-values using the Benjamini–Hochberg 323 approach (Benjamini & Hochberg 1995) for false discovery rates (n = 77 tests for each 324 index). To assess how the T3 detrending approach affected overall community stability, 325 this test was also applied to the CV. For each of the assessed synchrony indices, we also 326 retained for each dataset the t-statistic of the paired t-test, which indicates the strength 327 and the direction of the effect (positive values implying T3 increased synchrony, 328 negative ones when T3 decreased synchrony). Additionally, we evaluated how globally 329 the synchrony values responded to the T3 detrending approach using Linear Mixed 330 Models (LMM). In one approach, we computed for each plot two separate synchrony 331 values (synchrony with and without the T3 detrending approach). The LMM contained 332 one categorical variable (TraT3) as explanatory variable, specifying if the index was 333 calculated with the T3 detrending approach or not. Plots nested in each dataset were 334 considered as a random factor. Also, we computed for each plot the difference between 335 the synchrony values with the T3 detrending approach and the values without it. Then, 336 we evaluated how the effect of detrending (i.e. the difference between synchrony with 337 and without T3) varied across habitat types and the biomes by fitting a LMM in which 338 the dataset identity was considered as a random factor. 339 Third, we assessed whether synchrony values were affected by directional 340 trends depending on the presence of an experimental manipulation changing abruptly 341 the ecological conditions in a plot. To do this, we evaluated the effect of T3 using the 14 Journal of Vegetation Science Page 15 of 147 342 t-statistic of the paired t-test within dataset (see above), separately in control and 343 manipulated plots within datasets. This analysis was restricted to those 38 datasets (out 344 of 77) in which both control and manipulated plots were present and with at least three 345 plots in each category. The same approach was used to test the effect of the duration 346 (number of years) of the sampling period. This was undertaken using a linear model to 347 test the relationship between the t-statistic (resulting from the paired-test) and number 348 of years sampled in each dataset. We also used a similar LMM as described above to 349 jointly evaluate the effects of the duration of the sampling period and experimental 350 manipulation on the difference between the synchrony values with and without the T3 351 detrending approach in these 38 datasets. In this model, we used the number of years 352 of sampling, the experimental manipulation (manipulated vs. control plots) and their 353 interaction as fixed factor, while each dataset was considered as a random factor. When 354 a significant interaction was found, we split the database in control and manipulated 355 plots and evaluated the effects of duration of the sampling period on both groups of 356 plots. 357 Finally, to assess changes in strength of the commonly found ecological 358 relationships involving synchrony with or without the use of the T3 detrending 359 approach, we tested for each dataset using paired t-tests how strong were the (Pearson) 360 correlations between synchrony and (i) species richness and (ii) community stability. 361 For each of these two correlations, we considered the Pearson r and tested through a 362 paired t-test if this r value (one for each dataset) was greater or smaller when using the 363 T3 approach compared to when not using the T3 approach. 364 For simplicity, we mostly present the results of one index (GrossW) in the main 365 text because it is widely applied in the literature. However, most of the results for the 366 other indices considered are shown in Appendix (S3 and S4). Similarly, all results 15 Page 16 of 147 Journal of Vegetation Science 367 concerning simulations are also included as Supporting Information material (Appendix 368 S1). All the analysis were run in R (R Development Core Team 2018). 16 Journal of Vegetation Science Page 17 of 147 369 Results 370 The ranking of synchrony values with and without the T3 detrending approach was 371 relatively consistent, both within and across datasets (Fig. 1). The Spearman Rho values 372 computed within each of the 77 datasets were mostly positive and significant (Fig. 1a, 373 for GrossW as an example; similar patterns were obtained for the other indices, 374 Appendix S3). For example, in 44 out of the 77 datasets, the Spearman Rho was above 375 0.5. This indicates a moderate correspondence in the ranking in synchronicity values 376 across plots within datasets. Nevertheless, notable exceptions were present, for example 377 in six datasets (~8% of the cases) Rho was below 0.1. However, in five out of these six 378 datasets, either the number of manipulated plots was greater than the control plots, or 379 the control plots were entirely absent. Overall, the Spearman ranking test done on the 380 mean synchrony values indicated that greater synchrony without the T3 approach also 381 provided greater synchrony with the T3 approach (Fig. 1b: Rho = 0.81 and p < 0.001). 382 Most importantly, synchrony mean values were frequently greater where the T3 383 detrending approach was applied than without its use (paired t-test p < 0.001; Fig. 1b 384 and Appendix S3). 385 We generally found a greater synchrony when accounting for long-terms trends 386 with the T3 methods than without. A significant increase in synchrony values was found 387 for over 1/3 of the datasets (~30 datasets of 77, i.e. in ~40% of datasets synchrony 388 significantly increase, p < 0.05, after correcting p-values for multiple tests with the 389 Benjamini & Hochberg correction for false discovery rate within each synchrony index, 390 Fig. 2; all significant tests reported in this section account for this p-value correction). 391 Conversely, in around 10 datasets (13%, depending on the indices) synchrony values 392 decreased using the T3 approach. In total around 50% of the datasets showed a 393 significant change in synchrony values when using or not using the T3 detrending 17 Page 18 of 147 Journal of Vegetation Science 394 approach. The pattern described for GrossW index was similar for all other synchrony 395 indices. The number of datasets showing greater synchrony with the T3 approach was 396 lower using Phi, which also showed a higher number of datasets showing lower 397 synchrony with the T3 approach. In the majority of datasets (around 60) the CV 398 computed using the T3 approach was significantly lower compared to the one computed 399 without the T3 approach. 400 The LMM on the whole dataset showed a significant difference between the use 401 of synchrony with and without the T3 detrending approach (p < 0.001) with an overall 402 increase in synchrony with T3, meaning that the T3 detrending approach generally led 403 to increased synchrony values among all the plots (other synchrony indices yielded 404 similar results). This result (which is similar to the significant deviation from the 1:1 405 line in Fig. 1b mentioned above) further confirms that across the whole dataset long- 406 term trends generally blur the importance of synchrony between species. 407 The results of the LMM evaluating the effects of habitat type and biomes on the 408 T3 difference (i.e. on the difference between indices of synchrony with and without T3 409 within a plot) showed a significant effect of the habitat type (χ² = 47.21; p < 0.001), but 410 no effect of the biomes. Grassland and savanna had in average positive values, meaning 411 that a difference between T3 synchrony and synchrony without T3 were greater in these 412 two habitats. 413 As expected, detrending had greater impacts on measures of synchrony in 414 experimental plots than controls. Specifically evaluating ‘control’ vs. ‘manipulated’ 415 plots (using 38 datasets in which there were both types of plots), showed a greater 416 number of cases in which the T3 approach produced significant changes in synchrony 417 in the manipulated than in the control plots (Fig. 3 for the GrossW and Appendix S4 418 for the other synchrony indices): 21 significant datasets (60%) in the manipulated plots 18 Journal of Vegetation Science Page 19 of 147 419 but only 10 (27%) in the control plots. Moreover, the effect of the sampling period 420 length (number of years plots were sampled) was significantly related to the change in 421 mean synchrony with the T3 approach only in the case of the manipulated plots (Fig. 3, 422 using, as dependent variable, the t-values resulting by comparing synchrony with and 423 without T3 approach using the paired t-tests within plot described above). Specifically, 424 in the manipulated plots a longer sampling period improved the predictive ability of the 425 effect of T3 approach on synchrony (increased detection of synchrony over long-term 426 periods and increased detection of asynchrony in short-time periods). We confirmed 427 these results using an LMM in which the difference of synchrony with and without T3 428 were computed for each plot. This analyses showed a significant interaction between 429 sampling period length and experimental manipulation. Sampling period length 430 significantly increased the difference between synchrony values with and without the 431 T3 approach only in manipulated plots (χ² = 10.37; p = 0.001, n = 3414). 432 Finally, we found that overall the relationships between synchrony and both 433 species richness and community stability were similar (Appendix S5). Nevertheless 434 there were slightly more frequent significant cases after detrending for Gross and 435 GrossW (Appendix S5). For instance, the relationship between species richness and 436 synchrony (i.e. when considering GrossW) was found significant in 15 and 11 datasets 437 (out of 77) respectively when using or not using the T3 detrending approach (in both 438 cases correcting for false discovery rates). However, this relationship, with LogVar, 439 was found significant in 4 datasets less when using the T3. Further, with GrossW the 440 expected positive relationship between synchrony and community CV was significant 441 in 58 and 54 datasets while using or not using the T3 detrending, respectively (we did 442 not detect significant negative relationship between CV and synchrony). The strength 443 of these relationships, however, was not affected by the detrending approach. In neither 19 Page 20 of 147 Journal of Vegetation Science 444 the (i) species richness and synchrony correlations, nor the (ii) community CV and 445 synchrony correlations, did we detect significant differences when using or not using 446 the T3 detrending approach (in both cases p > 0.2). This implies that the use of the T3 447 detrending approach did not systematically produce greater or weaker correlations 448 when analyzing these common relationships. 20 Journal of Vegetation Science Page 21 of 147 449 Discussion 450 In this study we show that the synchrony patterns usually attributed to compensatory 451 dynamics could be actually caused by trends in species composition. Without 452 accounting for these trends effectively, it is possible that compensatory effects could be 453 generally overemphasized (in 30% of our datasets) or even underemphasized (in 10% 454 of our datasets). Previous studies of synchrony and compensatory dynamics have often 455 overlooked the possible effects of directional trends on the studied communities. Only 456 few studies, such as Vasseur and Gaedke (2007), Loreau & de Mazancourt (2008) and 457 Tredennick et al. (2017), have effectively filtered out species trends (using wavelet 458 based methods or considering growth rates of species in time, instead of raw 459 abundances). Long-term trends in abundances, either directional or cyclical, indeed 460 have the potential to bias the interpretation of synchrony with the most commonly used 461 indices. The T3 detrending approach can account for this bias (see simulation in Lepš 462 et al. 2019 and in Appendix S1). The advantages of the T3 approach, compared to other 463 approaches, are its lower data requirement and consideration of all species in a 464 community, not just the most frequent ones (Lepš et al. 2019). 465 In ~40% of the datasets, and in the overall model across all plots, synchrony 466 using the T3 detrending approach was significantly greater than synchrony without 467 using it (Fig. 2). The ~40% estimate is, furthermore, a conservative one as we account 468 for Type I errors. Overall, the mean values of synchrony computed with the T3 469 detrending approach were higher than without it in the majority of cases, both within 470 and across datasets (Fig. 1b, and LMM). This is an important finding because it suggests 471 that our appreciation of the importance of asynchrony, and therefore compensatory 472 dynamics, may have been possibly overestimated, leading to wrong conclusions about 473 synchrony-asynchrony in communities. These findings highlight the necessity of 21 Page 22 of 147 Journal of Vegetation Science 474 evaluating the effects of possible directional trends on synchrony to accurately estimate 475 the importance of ecological mechanisms regulating compensatory dynamics. The 476 difference between the indices calculated using T3 detrending approach and without it 477 were higher in grasslands and meadows, possibly because in the absence of slow- 478 growing, less dynamic, woody species. In these communities temporal trends can thus 479 be more easily detected compared to other types of vegetation. The increase in 480 synchrony after detrending also suggests the presence of opposite trends of species 481 abundances in time, such as when one species is decreasing steadily and another 482 increasing. For example, trends could be the result of species responding differently to 483 disturbance or to an increase in nutrient availability. Such opposite trends could be 484 monotonic or following waves in time (Wu et al. 2007), e.g. resulting from periodic 485 climate events such as “El Niño”, or intrinsic cycling of particular functional groups 486 such as legumes (Herben et al. 2017). These results are partially expected because our 487 datasets comprised natural or semi-natural well-established plant communities but 488 included experimental conditions in which changes in abundance or composition of 489 species are common. 490 When considering datasets with both control and manipulated plots (~50% of 491 the datasets) the effect of the T3 approach was more frequently significant in 492 manipulated plots than in control plots (Fig. 3). These plots were more prone to be 493 affected by a directional trend promoted by the specific manipulation imposed. This 494 result agrees with our hypothesis that events like soil-nutrient alteration (e.g. by 495 fertilization) and recovery from disturbance might promote directional trends. This 496 result was expected as some of the experimental manipulations were designed to 497 directly alter species composition, in order to test their effects on community 498 synchrony. However, such prompted changes, often due to colonization-competition 22 Journal of Vegetation Science Page 23 of 147 499 trade-offs in species composition, can mask year-to-year fluctuations, and hence these 500 experiments should disentangle these biologically different effects on synchrony. For 501 these reasons, we recommend that any index of synchrony should be computed with 502 and without the T3 approach to properly evaluate the corresponding effects of long- 503 term experimental treatments and year-to-year fluctuations. Our result reinforces the 504 assumption that the effect of the T3 approach could be stronger in changing 505 environments/communities and the combination of indices with and without the T3 506 approach can be important to distinguish the mechanisms causing differential long-term 507 species responses to changes in environmental conditions from the differential species 508 responses to short-term species fluctuations on synchrony/asynchrony relationships. 509 The effect of detrending on synchrony values was particularly pronounced in 510 the case of succession. During succession the majority of species will increase their 511 abundance, which will cause them to be ultimately positively correlated in time. 512 However, these same species can compensate each other or vary independently on a 513 year-by-year basis, even if they all generally increase in time, so the existing synchrony 514 indices would tend to overestimate their actual year-to-year synchrony between species 515 within such communities. In fact, among the seven datasets with a Rho below 0.1 (Fig. 516 1a), the majority were characterised by being exposed to intense disturbance regimes 517 that triggered some type of successional process. For instance, plots of four datasets 518 had been exposed to a fire before or during the experiment, and two evaluated the effect 519 of herbivory exclusion (where the reduction in grazing intensity allowed the 520 development of higher vegetation like shrubs and trees). Both treatments are good 521 examples of environmental conditions promoting species directional trends (Pardo et 522 al. 2015) and thus affect synchrony values. 23 Page 24 of 147 Journal of Vegetation Science 523 Interestingly, the effect of the T3 approach on the synchrony measured in 524 manipulated plots depended on the period length of the sampling scheme. Manipulated 525 plots sampled over longer time periods revealed higher synchrony values when using 526 the T3 detrending approach (Fig. 3). In other words, the longer is the sampling period 527 the greatest chance that there is a difference between T3 synchrony and synchrony 528 without T3 in manipulated plots. Longer time series likely increased the chances that 529 some species will have opposite trends in response to manipulation, with some 530 increasing over time and others decreasing. In a shorter time series, on the contrary, the 531 time lag in species responses (particularly extinction debt, Helm, Hanski, & Partel 532 2006; Lepš 2014) could cause that some species increase quickly in response to 533 manipulation, while others might respond more slowly. The T3 detrending approach, 534 therefore, will affect those species with a similar temporal trend in response to short- 535 term manipulations. Consequently, the duration of the sampling period stands out as a 536 key factor in the evaluation of temporal dynamics. We showed that, in the case of 537 manipulated communities, classical methods tended to overestimate year-to-year 538 synchrony when the sampling period was shorter, and underestimate it when the 539 sampling period was longer. This highlights the importance of T3 approach for a correct 540 evaluation of year-to-year synchrony between species. However, further research is 541 required to find the causes and consequences of these results. 542 Finally, we generally found that the T3 detrending approach did not cause strong 543 changes in the correlation between synchrony and both species richness and community 544 stability, two of the most iconic relationships in temporal dynamics studies (Hautier et 545 al. 2014; Blüthgen et al. 2016). However, there were more cases of significant 546 correlations with the T3 approach and strength of the correlations could vary 547 considerably (i.e. R < 0.6) across datasets. In summary, this suggests that while the 24 Journal of Vegetation Science Page 25 of 147 548 applications of the T3 detrending approach did not produce systematically greater or 549 weaker correlations on commonly used tests in ecology, the strength of the relationships 550 could differ. These results confirm that the use of T3 approach to detrend the synchrony 551 indices is far from trivial. As such, the conclusions obtained previously from studies 552 that did not apply the method are not necessarily incorrect. Therefore, applying the 553 detrended and non-detrended methods in a complementary way might bring us closer 554 to understanding the directional changes in community dynamics. For instance, 555 divergent trends, e.g. due to differential response to global warming with some species 556 increasing and other decreasing, might stabilize communities and could maintain 557 ecosystem functions unaltered in response to global warming, even if there are no short- 558 term compensatory mechanisms between species. Hence, it is important to consider 559 both the synchrony with and without detrending approach for teasing apart different 560 causes of stability, or instability, in response to global change drivers. 561 The evaluation of synchrony with the T3 detrending method provides a feasible 562 measure to reveal year-to-year fluctuations of species by removing the effect of 563 directional trends. In comparison to methods using species growth rates, the T3 564 approach can be important because it enables the evaluation of the indices with and 565 without the approach and also accounts for species which are not dominant and/or less 566 frequent (in the case of the growth rates, log-transformation is needed, which might not 567 be advisable in the case of zero abundances in specific years). This method has the 568 advantage of evaluating both monotonic and non-monotonic directional trends, and can 569 thus be used to detect year-to-year fluctuations in the face of cyclical periods, such as 570 alternation between drought-wet periods (e.g. Riginos et al. 2018). 25 Page 26 of 147 Journal of Vegetation Science 571 Acknowledgements 572 We thank multiple entities for the financial support necessary to obtain the different 573 databases: the U.S. National Science Foundation under grant numbers DEB-8114302, 574 DEB-8811884, DEB-9411972, DEB-0080382, DEB-0620652, DEB-1234162, DEB- 575 9707477, DEB-0316402, DEB-08-16453, and DEB-12-56034, DEB-0618210, the 576 Nutrient Network (http://www.nutnet.org) experiment from the National Science 577 Foundation Research Coordination Network (NSF-DEB-1042132), the New Zealand 578 National Vegetation Survey Databank, the Spanish MINECO (Project CGL2014- 579 53789-R), the Madrid Regional Government (Projects REMEDINAL-3 and 580 REMEDINAL-TE), the European Research Council Synergy grant 610028 581 (IMBALANCE-P), the Institute on the Environment (DG-0001-13), the SOERE- 582 ACBB financed through French National Agency for Research (ANAEE, ANR-11- 583 INBS-0001), the Estonian Research Council (IUT 20-28, IUT 20-29), Czech Science 584 Foundation (GAČR 17-05506S), the European Regional Development Fund (Centre of 585 Excellence EcolChange), the German Federal Environmental Foundation (DBU) for a 586 grant to the NABU Hamburg (management experiment Calamagrostis epigejos), and 587 the German Federal Ministry of Education and Research within the framework of the 588 project BIOTA Southern Africa (promotion numbers 01LC0024, 01LC0024A and 589 01LC0624A2), Task 159 of SASSCAL (promotion number 01LG1201) and the 590 Scottish Government’s Rural and Environmental Science and Analytical Services 591 division. Acknowledgement Data owned by NERC© Database Right/Copyright NERC. 592 Further support was provided by the Jornada Basin Long-Term Ecological Research 593 (LTER) project, Cedar Creek Ecosystem Science Reserve and the University of 594 Minnesota. We also thank the Lawes Agricultural Trust and Rothamsted Research for 595 data from the e-RA database. The Rothamsted Long-term Experiments National 596 Capability (LTE-NCG) is supported by the UK Biotechnology and Biological Sciences 597 Research Council (Grant BBS/E/C/000J0300) and the Lawes Agricultural Trust. 598 599 600 Author contributions: F.B., L.G. and J.L. conceived the project. All authors but E.V., 601 F.B., T.G and L.G. collected the data used in this analysis. E.V. and T.G. assembled 602 data. F.B. performed the analyses. E.V. and F.B. wrote the first draft of the manuscript 603 and all the authors (especially L.G. and J. L.) contributed substantially to the revisions. 604 605 Data accessibility 606 The data that support the findings of this study are available at Figshare (Valencia et al. 607 2019). 608 26 Journal of Vegetation Science Page 27 of 147 609 References 610 611 Benjamini, Y., & Hochberg, Y. 1995. Controlling the False Discovery Rate: A 612 Practical and Powerful Approach to Multiple Testing. Journal of the Royal 613 Statistical Society: Series B (Methodological) 57: 289–300. 614 Blüthgen, N., Simons, N.K., Jung, K., Prati, D., Renner, S.C., Boch, S., Fischer, M., 615 Hölzel, N., Klaus, V.H., Kleinebecker, T., Tschapka, M., Weisser, W.W., & 616 Gossner, M.M. 2016. Land use imperils plant and animal community stability 617 through changes in asynchrony rather than diversity. Nature Communications 7: 618 10697. 619 Cardinale, B., Wright, J., Cadotte, M.W., Carroll, I.T., Hector, A., Srivastava, D.S., 620 Loreau, M., & Weis, J.J. 2007. Impacts of plant diversity on biomass production 621 increase through time because of species complementarity. Proceedings of the 622 National Academy of Sciences of the United States of America 104: 18123– 623 18128. 624 Craven, D., Eisenhauer, N., Pearse, W.D., Hautier, Y., Isbell, F., Roscher, C., Bahn, 625 M., Beierkuhnlein, C., Bönisch, G., Buchmann, N., Byun, C., Catford, J.A., 626 Cerabolini, B.E.L., Cornelissen, J.H.C., Craine, J.M., De Luca, E., Ebeling, A., 627 Griffin, J.N., Hector, A., Hines, J., Jentsch, A., Kattge, J., Kreyling, J., Lanta, V., 628 Lemoine, N., Meyer, S.T., Minden, V., Onipchenko, V., Polley, H.W., Reich, 629 P.B., van Ruijven, J., Schamp, B., Smith, M.D., Soudzilovskaia, N.A., Tilman, 630 D., Weigelt, A., Wilsey, B., & Manning, P. 2018. Multiple facets of biodiversity 631 drive the diversity–stability relationship. Nature Ecology and Evolution 2: 1579– 632 1587. 633 Gonzalez, A., & Loreau, M. 2009. The causes and consequences of compensatory 634 dynamics in ecological communities. Annual Review of Ecology, Evolution, and 635 Systematics 40: 393–414. 636 Gross, K., Cardinale, B.J., Fox, J.W., Gonzalez, A., Loreau, M., Wayne Polley, H., 637 Reich, P.B., & van Ruijven, J. 2014. Species richness and the temporal stability 638 of biomass production: a new analysis of recent biodiversity experiments. The 639 American Naturalist 183: 1–12. 640 Hautier, Y., Seabloom, E.W., Borer, E.T., Adler, P.B., Harpole, W.S., Hillebrand, H., 641 Lind, E.M., MacDougall, A.S., Stevens, C.J., Bakker, J.D., Buckley, Y.M., Chu, 642 C., Collins, S.L., Daleo, P., Damschen, E.I., Davies, K.F., Fay, P.A., Firn, J., 643 Gruner, D.S., Jin, V.L., Klein, J.A., Knops, J.M.H., La Pierre, K.J., Li, W., 644 McCulley, R.L., Melbourne, B.A., Moore, J.L., O’Halloran, L.R., Prober, S.M., 645 Risch, A.C., Sankaran, M., Schuetz, M., & Hector, A. 2014. Eutrophication 646 weakens stabilizing effects of diversity in natural grasslands. Nature 508: 521– 647 525. 648 Helm, A., Hanski, I., & Partel, M. 2006. Slow response of plant species richness to 649 habitat loss and fragmentation. Ecology Letters 9: 72-77. 650 Herben, T., Mayerová, H., Skálová, H., Hadincová, V., Pecháčková, S., & Krahulec, 651 F. 2017. Long-term time series of legume cycles in a semi-natural montane 652 grassland: evidence for nitrogen-driven grass dynamics? (R. Ostertag, Ed.). 653 Functional Ecology 31: 1430–1440. 654 Hill, M.O. 1973. The Intensity of spatial pattern in plant communities. The Journal of 655 Ecology 61: 225. 27 Page 28 of 147 Journal of Vegetation Science 656 Houlahan, J.E., Currie, D.J., Cottenie, K., Cumming, G.S., Ernest, S.K.M., Findlay, 657 C.S., Fuhlendorf, S.D., Gaedke, U., Legendre, P., Magnuson, J.J., McArdle, 658 B.H., Muldavin, E.H., Noble, D., Russell, R., Stevens, R.D., Willis, T.J., 659 Woiwod, I.P., & Wondzell, S.M. 2007. Compensatory dynamics are rare in 660 natural ecological communities. Proceedings of the National Academy of 661 Sciences 104: 3273–3277. 662 Hubbell, S.P. 2001. The unified neutral theory of biodiversity and biogeography. 663 Princeton University Press. 664 Lepš, J. 1990. Comparison of transect methods for the analysis of spatial pattern. 665 Spatial processes in plant communities 71-82. 666 Lepš, J. 2014. Scale- and time-dependent effects of fertilization, mowing and 667 dominant removal on a grassland community during a 15-year experiment (S. 668 Wan, Ed.). Journal of Applied Ecology 51: 978–987. 669 Lepš, J., Götzenberger, L., Valencia, E., & de Bello, F. 2019. Accounting for 670 long‐term directional trends on year‐to‐year synchrony in species fluctuations. 671 Ecography 42: 1728-1741. 672 Lepš, J., Májeková, M., Vítová, A., Doležal, J., & de Bello, F. 2018. Stabilizing 673 effects in temporal fluctuations: management, traits, and species richness in high- 674 diversity communities. Ecology 99: 360–371. 675 Loreau, M., & de Mazancourt, C. 2013. Biodiversity and ecosystem stability: a 676 synthesis of underlying mechanisms (E. Duffy, Ed.). Ecology Letters 16: 106– 677 115. 678 Loreau, M., & de Mazancourt, C. 2008. Species synchrony and its drivers: neutral and 679 nonneutral community dynamics in fluctuating environments. The American 680 Naturalist 172: E48–E66. 681 Milchunas, D., Lauenroth, W., & Burke, I. 1998. Livestock grazing: animal and plant 682 biodiversity of shortgrass steppe and the relationship to ecosystem function. 683 Oikos 83: 65–74. 684 Pardo, I., Doak, D.F., García-González, R., Gómez, D., & García, M.B. 2015. Long- 685 term response of plant communities to herbivore exclusion at high elevation 686 grasslands. Biodiversity and Conservation 24: 3033–3047. 687 R Development Core Team. 2018. R: A language and environment for statistical 688 computing (A. R Foundation for Statistical Computing, Vienna, Ed.). R 689 Foundation for Statistical Computing, Vienna, Austria. 690 Rabotnov, T.A. 1974. Differences Between Fluctuations and Successions. In 691 Vegetation Dynamics, pp. 19–24. Springer Netherlands, Dordrecht. 692 Riginos, C., Porensky, L.M., Veblen, K.E., & Young, T.P. 2018. Herbivory and 693 drought generate short-term stochasticity and long-term stability in a savanna 694 understory community. Ecological Applications 28: 323–335. 695 Silvertown, J., Poulton, P., Johnston, E., Edwards, G., Heard, M., & Biss, P.M. 2006. 696 The Park Grass Experiment 1856–2006: its contribution to ecology. Journal of 697 Ecology 94: 801–814. 698 Stevens, C., Duprè, C., Gaudnik, C., Dorland, E., Dise, N., Gowing, D., Bleeker, A., 699 Alard, D., Bobbink, R., Fowler, D., Vandvik, V., Corcket, E., Mountford, J.O., 700 Aarrestad, P.A., Muller, S., & Diekmann, M. 2011. Changes in species 701 composition of European acid grasslands observed along a gradient of nitrogen 28 Journal of Vegetation Science Page 29 of 147 702 deposition. Journal of Vegetation Science 22: 207–215. 703 Tilman, D., Isbell, F., & Cowles, J.M. 2014. Biodiversity and ecosystem functioning. 704 Annual Review of Ecology, Evolution, and Systematics 45: 471–493. 705 Tomimatsu, H., Sasaki, T., Kurokawa, H., Bridle, J.R., Fontaine, C., Kitano, J., 706 Stouffer, D.B., Vellend, M., Martijn Bezemer, T., Fukami, T., Hadly, E.A., Van 707 Der Heijden, M.G.A., Kawata, M., Efi, S.K., Kraft, N.J.B., Mccann, K.S., 708 Mumby, P.J., Nakashizuka, T., Petchey, O.L., Romanuk, T.N., Suding, K.N., 709 Takimoto, G., Urabe, J., & Yachi, S. 2013. Sustaining ecosystem functions in a 710 changing world: a call for an integrated approach. Journal of Applied Ecology 711 50: 1124–1130. 712 Tredennick, A.T., de Mazancourt, C., Loreau, M., & Adler, P.B. 2017. Environmental 713 responses, not species interactions, determine synchrony of dominant species in 714 semiarid grasslands. Ecology 98: 971–981. 715 Vasseur, D.A., & Gaedke, U. 2007. Spectral analysis unmasks synchronous and 716 compensatory dynamics in plankton communities. Ecology 88: 2058–2071. 717 Walter, J.A., Ives, A.R., Tooker, J.F., & Johnson, D.M. 2018. Life history and habitat 718 explain variation among insect pest populations subject to global change. 719 Ecosphere 9: e02274. 720 Wu, Z., Huang, N.E., Long, S.R., & Peng, C.-K. 2007. On the trend, detrending, and 721 variability of nonlinear and nonstationary time series. Proceedings of the 722 National Academy of Sciences of the United States of America 104: 14889– 723 14894. 724 725 29 Page 30 of 147 Journal of Vegetation Science 726 Figure 1. Effects of the T3 detrending approach on synchrony, using the GrossW index 727 (Blüthgen et al. 2016) as an example. In panel (a), a ranking correlation between 728 synchrony values with and without detrending was computed for each of the 77 datasets 729 considered. The histogram reports the 77 Rho values of the Spearman ranking 730 correlations. Panel (b) reports, for each of the 77 datasets, the mean (+/- standard error) 731 of the synchrony values with and without the T3 detrending approach. Vertical and 732 horizontal dashed lines indicate zero synchrony (i.e. absence of synchrony). The solid 733 line represents the 1:1 line above which, for example T3 synchrony was greater than 734 synchrony without T3. 735 736 30 Journal of Vegetation Science Page 31 of 147 737 Figure 2. Summary of the directional effects of the T3 detrending approach on various 738 synchrony indices and on CV. The bar plots indicate the numbers of datasets (n=77) in 739 which the T3 approach significantly increased (red bars) or decreased (blue bars) 740 synchrony values using a paired t-test after correction for false discovery rates. Grey 741 bars indicate the number of datasets with non-significant paired t-tests. 742 For 743 31 Page 32 of 147 Journal of Vegetation Science 744 Figure 3. Effects of the T3 detrending approach in manipulated vs. control plots. The 745 plots report results of t-tests on 38 datasets in which there were both manipulated and 746 control plots. For each dataset we used a pairwise t-test to compare synchrony values 747 (using the GrossW synchrony index, Blüthgen et al. 2016) with and without the T3 748 approach (a: manipulated plots, and b: control plots). Positive values of the t-statistic 749 indicate that the T3 approach increased synchrony and negative ones indicate that the 750 T3 decreased synchrony. Values outside the grey area in each plot indicate significant 751 t-tests after correction for false discovery rates (‘ns’ indicates p > 0.05). For each panel 752 an R2 for the relationship between t-statistic and number of years sampled in each 753 dataset is provided together with the p-value of the regression model (the corresponding 754 regression line is shown when significant). 755 For Revie w 756 32 Journal of Vegetation Science Page 33 of 147 757 Supporting Information 758 Additional Supporting Information may be found in the online version of this article: 759 760 Appendix S1. Simulating long term trends in artificial communities to validate 761 effectiveness of the T3 approach 762 Appendix S2. Descriptions of each dataset, highlighting the treatments of the datasets 763 with ‘control’ and ‘manipulated’ plots. 764 Appendix S3. Application of the analyses shown in Fig. 1 of the main text to the 765 three remaining indices of synchrony. 766 Appendix S4. Application of the analyses shown in Fig. 3 of the main text to the 767 three remaining indices of synchrony. 768 Appendix S5. Results of the correlation between synchrony indices with species 769 richness or with the CV of total abundance. 33 Page 1 of 147 Journal of Vegetation Science 1 Supporting Information to the paper 2 Valencia et al. Directional trends in species composition over time can lead to a 3 widespread overestimation of asynchrony. Journal of Vegetation Science. 4 5 Appendix S1. Simulating long term trends in artificial communities to validate 6 effectiveness of the T3 approach 7 We created artificial temporal community data with desired patterns of temporal 8 fluctuations (prevailing synchrony or asynchrony) using the “syngenr” R function 9 (Lepš et al. 2019). This function offers the possibility to build simulated communities, 10 fixing some parameters, such as the years of the time series (100 years) and the 11 number of species (8 species). Once the communities were established, communities 12 fluctuating in time were created according to the following scenarios: prevailing 13 synchrony or prevailing asynchrony. A synchronous pattern was simulated by having 14 a common response for all species to a hypothetical environmental cue. Accordingly, 15 an asynchronous pattern was created by having half of the species responding 16 positively and the other half negatively to the environmental cue. Furthermore, we 17 simulated directional (monotonic) and cyclical long term trends for these artificial 18 communities. First, we simulated a case where most species had a common long-term 19 positive trend (monotonic) leading to a steady increase of individual species over 20 time. This would lead to detect synchrony with the synchrony indices (unmodified), 21 even if the species are actually behaving asynchronous. Second, we simulated the 22 opposite case, where species either increase or decrease in time, with the 23 increase/decrease for each species defined by a combined bimodal distribution from 24 two normal distributions with -1 and 1 as means, and random subset from half of the 25 species more probably have a positive long-term trend and the other half of the 26 species more probably a negative long-term trend. Finally, we simulated a case were 27 the directional long-term had cyclical tendencies. The cyclical long term trends were 1 Journal of Vegetation Science Page 2 of 147 28 evaluated with different cycle length: 3, 6, 18 and 40 years. These cycle lengths 29 reflect some known potential long term cycles that drive communities across the 30 world, such as the El Niño Oscillation or Pacific Decadal Oscillation, which have 31 intervals of 3-6 years and 10-20 year, respectively. In summary, we simulated two 32 scenarios of year-to-year species fluctuations (prevailing synchrony or prevailing 33 asynchrony) and three types of long-term directional trends (i.e. monotonic with a 34 common or contrasted trend, and cyclical trends), resulting in six possible 35 combinations of trend–fluctuation scenarios. In all these simulated communities, we 36 calculated the different synchrony indices (Gross, GrossW, Logvar and Phi), with or 37 without the use of the T3 detrending approach, using the “calc_sync” R function 38 (Lepš et al. 2019). 39 We assessed the effectiveness of the T3 detrending approach when long-term 40 monotonic or cyclical trends are present in the data across the most common 41 synchrony indices, using a paired t-test. Fluctuations simulated under scenarios of 42 long-term trends in species abundances showed biased index estimates, i.e. the 43 simulated synchrony or asynchrony patterns were overshadowed by the patterns 44 caused by long-term trends. In the case of species having long-term directional or 45 cyclical trends, asynchrony was masked by the synchrony (Figure Appendix S1a and 46 S1c). Then, the synchrony indices without the T3 detrending approach were not able 47 to detect asynchrony, even if the species were actually behaving asynchronously. 48 These synchrony indices values were significantly higher than those with the T3 49 detrending approach. These biases were found across all indices but the application of 50 the T3 detrending approach was correctly able to remove them, in all indices (Figure 51 Appendix S1). In the opposite case, simulation of synchrony together with long-term 52 monotonic or cyclical trends, the difference still prevailed among the synchrony with 2 Page 3 of 147 Journal of Vegetation Science 53 and without the T3 detrending approach, but with a less pronounced effect (Figure 54 Appendix S1b and S1d). 55 56 Figure Appendix S1. Results of synchrony indices (Gross, GrossW Logvar, and 57 Phi)(‘Logvar’, Lepš et al. 2018)(‘Logvar’, Lepš et al. 2018), with or without the use 58 of the T3 detrending approach, in artificial temporal communities where long term 59 trends were simulated. The panels report results for a common long-term directional 60 trend (a) (i.e. creating synchrony; all species increasing in time), a contrasted long- 61 term trend (b) (i.e. half species increasing, the other half decreasing, creating 62 asynchrony) and a cyclical trend (c and d). Within each of these scenarios we 63 considered two scenarios: year to year asynchrony (a and c) and synchrony (b and d). 64 The cyclic trends also included different cycle length (3, 6, 18 and 40 years). The 65 created communities had a total of 8 species. Asterisks above and between boxes 66 depict significant differences among the synchrony indices with or without the T3 67 approach as assessed with a paired t-test. *: P < 0.05; **: P < 0.01; ***: P < 0.001; 68 ****: P < 0.0001. 69 70 Gross 71 72 3 Journal of Vegetation Science Page 4 of 147 73 GrossW 74 75 76 77 Logvar 78 79 80 81 4 Page 5 of 147 Journal of Vegetation Science 82 Phi 83 84 85 5 Journal of Vegetation Science Page 6 of 147 86 Supporting Information to the paper 87 Valencia et al. Directional trends in species composition over time can lead to a 88 widespread overestimation of asynchrony. Journal of Vegetation Science. 89 90 Appendix S2. Descriptions of each dataset, highlighting the treatments of the datasets 91 with ‘control’ and ‘manipulated’ plots. LAT: latitude (WGS84 datum), and LON: 92 longitude (WGS84 datum). 93 94 1. The dataset is issued from an experiment in a northern mixed prairie at a field 95 station in Miles City, Montana, USA (LAT: 46.32, and LON: -105.80). This dataset 96 consists of 42 plots, where each plot was sampled an average of 12.5 times. In each 97 plot, individual plants were quantified and mapped annually. More information: 98 http://esapubs.org/archive/ecol/E092/143/#data 99 100 2. The dataset is issued from an experiment located on a mixed grass prairie in Hays, 101 Kansas, USA (LAT: 38.80, and LON: -99.30). This dataset consists of 51 plots, where 102 each plot was sampled an average of 34.5 times. In each plot, individual plants were 103 quantified and mapped. Thirty-six permanent quadrats were located inside livestock 104 exclosures and 15 in grazed areas. More information: 105 https://web.archive.org/web/20150128015820/http://esapubs.org:80/archive/ecol/E08 106 8/161/default.htm 107 108 3. The dataset is issued from an experiment located on a shortgrass steppe of North 109 America in Nunn, Colorado, USA (LAT: 40.85, and LON: -104.71). This dataset 110 consists of 24 plots, where each plot was sampled an average of 13.5 times. In each 111 plot, individual plants were quantified and mapped. The quadrats were established in 112 six grazed and ungrazed study sites on the Central Plains Experimental Range. There 113 were four treatments combining past and present grazing status: ungrazed in the past 114 and at present (ungrazed/ungrazed), grazed by livestock in the past and present 115 (grazed/grazed), grazed in the past and ungrazed during the experiment 116 (grazed/ungrazed), and ungrazed in the past and grazed during the experiment 117 (ungrazed/grazed). More information: 118 https://web.archive.org/web/20150502183659/http:/www.esapubs.org/archive/ecol/E0 119 94/128/ 120 6 Page 7 of 147 Journal of Vegetation Science 121 4. The dataset is issued from an experiment located on semi-desert grasslands at the 122 Santa Rita Experimental Range, Arizona, USA (LAT: 31.83, and LON: -110.88). This 123 dataset consists of 160 plots, where each plot was sampled an average of 11.2 times. 124 In each plot, individual plants were quantified and mapped. Quadrats were located in 125 exclosures (ungrazed) and in pastures grazed by livestock (grazed). More information: 126 https://web.archive.org/web/20150502183207/http://esapubs.org:80/archive/ecol/E09 127 3/132/default.htm 128 129 5. The dataset is issued from an experiment located in sagebrush steppe in eastern 130 Idaho, USA (LAT: 44.20, and LON: -112.20). This dataset consists of 23 plots, where 131 132 orF each plot was sampled an average of 21.5 times. In each plot, individual plants were quantified and mapped. These permanent quadrats were located in both grazed (4 133 quadrats) and ungrazed units (18 quadrats), and one quadrat was grazed in the past 134 and ungrazed during the experiment. More information: 135 https://web.archive.org/web/20150128015825/http:/esapubs.org/archive/ecol/E091/24 136 3/default.htm. 137 138 6. The dataset is issued from an experiment on the Jornada Long-Term Ecological 139 Research site in southern New Mexico, USA (LAT: 32.83, and LON: -107.33). This 140 dataset consists of 222 plots, where each plot was sampled an average of 8.0 times. 141 Previously grazing domestic livestock was excluded from the area where three 142 permanent transects (2.7 km) were established. One of the transects received 143 fertilization of 10 g/m2 of nitrogen. One of the two control transects (not fertilized), 144 was sampled at 40 stations, the other two transects had 91 stations each. At each 145 station abundance of each species was estimated by point-intercept method along a 30 146 m transect perpendicular to each of the three permanent transects. More information: 147 https://portal.lternet.edu/nis/mapbrowse?packageid=knb-lter-jrn.2100119001.50. 148 149 7. The dataset is issued from an experiment on the Jornada Basin Long-Term 150 Ecological Research Program (LTER) site in the Chihuahuan desert, New Mexico, 151 USA (LAT: 32.93, and LON: -107.36). This dataset consists of 1001 plots, where 152 each plot was sampled an average of 11.5 times. On the grassland site, three exclusion 153 treatment levels were set in addition to the control treatment left open to all grazers. 154 The first level excluded only domestic animals (cattle), the second excluded 7 Journal of Vegetation Science Page 8 of 147 155 lagomorphs, and the third excluded rodents. In the shrubland site, only lagomorph- 156 and rodent-exclusion treatments were set in addition to the control. In each treatment 157 of each site, 4 grids of 36 permanent plots (1 m2) were sampled (visual estimated 158 cover). More information: 159 https://portal.lternet.edu/nis/metadataviewer?packageid=knb-lter-jrn.2100086002.39. 160 161 8. The dataset is issued from an experiment in an open grassland of the South African 162 Kalahari near Askham, South Africa (LAT: -26.76, and LON: 20.61). This dataset 163 consists of 20 plots, where each plot was sampled an average of 10.7 times. The cover 164 values (%) of all individual plant species were estimated annually. More information: 165 Jürgens et al. (2010). 166 167 9. The dataset is issued from an experiment located in the Succulent Karoo in 168 Soebatsfontein, South Africa (LAT: -30.19, and LON: 17.54). This dataset consists of 169 24 plots, where each plot was sampled an average of 15.8 times. The cover values (%) 170 of all individual plant species were estimated annually. More information: Jürgens et 171 al. (2010). 172 173 10. The dataset is issued from an experiment located in the Succulent Karoo, near 174 Leliefontein, South Africa (LAT: 18.28, and LON: -30.40). This dataset consists of 42 175 plots, where each plot was sampled an average of 14.7 times. The cover values (%) of 176 all individual plant species were estimated annually. More information: Jürgens et al. 177 (2010). 178 179 11. The dataset is issued from an experiment located in the Succulent Karoo, 180 Knersvlakte near Vanrhynsdorp, South Africa (LAT: -31.28, and LON: 18.59). This 181 dataset consists of 40 plots, where each plot was sampled an average of 16.0 times. 182 The cover values (%) of all individual plant species were estimated annually. More 183 information: Jürgens et al. (2010). 184 185 12. The dataset is issued from an experiment on the Kiskun LTER located in Bugac 186 and Orgovány sites of Kiskunság National Park, Hungary (LAT: 46.73, and LON: 187 19.54). This dataset consists of 380 plots, where each plot was sampled an average of 188 14.5 times. Half of the plots were fenced to control grazing pressure. In each plot, the 8 Page 9 of 147 Journal of Vegetation Science 189 cover values (%) were visually estimated annually. More information: Kertész et al. 190 (2017). 191 192 13. The dataset is issued from an experiment on a grassland in Cedar Creek LTER 193 Ecosystem Science Reserve, Minnesota, USA (LAT: 45.41, and LON: -93.16). This 194 dataset consists of 50 plots, where each plot was sampled an average of 7.0 times. The 195 plots were divided in 10 treatments of fertilization and grazing exclusion (Control=no 196 treatment, K=potassium, P=phosphate, N=nitrogen, PK=phosphate and potassium, 197 NK=nitrogen and potassium, NP=nitrogen and phosphate, NPK=nitrogen, phosphate 198 199 200 oF and potassium, Fence=Fence, NPK+Fence=nitrogen, phosphate and potassium + fence). In each plot, the cover values (%) were visually estimated annually. This dataset was provided from Cedar Creek LTER. More information: 201 http://cedarcreek.umn.edu/research/data/dataset?acze247. 202 203 14. The dataset is issued from an experiment located in the Cedar Creek LTER 204 Ecosystem Science Reserve, Minnesota, USA (LAT: 45.41, and LON: -93.19). This 205 dataset consists of 184 plots, where each plot was sampled an average of 6.2 times. 206 Plots were distributed across 6 treatments with increasing burning frequency: i) no 207 burning – control – (48 plots), ii) 1 per 10 years (16 plots), iii) 1 per 3 years (32 plots), 208 iv) 1 per 2 years (32 plots), v) 2 per 3 years (8 plots) and vi) 4 per 5 years (48 plots). 209 Plots are located on 12 management areas ranging in size from 2.4 to 30 ha. In each 210 plot, the cover values (%) were visually estimated. More information: 211 http://cedarcreek.umn.edu/research/data/dataset?herbe133. 212 213 15. The dataset is issued from an experiment located in the Cedar Creek LTER 214 Ecosystem Science Reserve, Minnesota, USA (LAT: 45.41, and LON: -93.19). This 215 dataset consists of 60 plots, where each plot was sampled an average of 24.8 times. In 216 each plot, the biomass of individual plants was recorded from 4 plots (0.3 m2) per 217 field until 2013. More information: 218 http://cedarcreek.umn.edu/research/data/dataset?ple054. 219 220 16. The dataset is issued from an experiment located in the Cedar Creek LTER 221 Ecosystem Science Reserve, Minnesota, USA (LAT: 45.40, and LON: -93.20). This 222 dataset consists of 234 plots, where each plot was sampled an average of 22.0 times. 9 Journal of Vegetation Science Page 10 of 147 223 The experiment combines different levels of fertilization on 4 fields that were 224 abandoned for different periods (14, 25, 48 years and never ploughed before the 225 experiment started in 1982) and where mammal grazers were excluded. In each plot, 226 individual plant biomass was recorded on 5 to 6 replicate plots of different 227 fertilization treatments (from 0 to 40 g/m2 of nitrogen) per field every year. More 228 information: http://cedarcreek.umn.edu/research/data/dataset?ple001. 229 230 17. The dataset is issued from an experiment located in the Cedar Creek LTER 231 Ecosystem Science Reserve, Minnesota, USA (LAT: 45.40, and LON: -93.20). This 232 dataset consists of 237 plots, where each plot was sampled an average of 14.8 times. 233 The experiment combines 9 levels of fertilization (from 0 to 40 g/m2 of nitrogen) and 234 prescribed burning on three fields that were abandoned since 14, 25 and 48 years, and 235 where mammal grazers were excluded. All 3 fields had 6 replicate plots of the 9 236 fertility treatments from 1982. From 1992 half of the plots in field B were burned 237 every spring, and half of the plots in field A and C stopped receiving the fertilization 238 treatment. To maintain continuity of the treatments within plots the 1992-2011 period 239 of those plots were entered in the database as separate plots of the same data set. 240 Individual plant biomass was measured. More information: 241 http://cedarcreek.umn.edu/research/data/dataset?ple002. 242 243 18. The dataset is issued from an experiment located in the Shortgrass Steppe LTER 244 in the Central Plains Experimental Range, Colorado, USA (LAT: 40.85, and LON: - 245 104.77). This dataset consists of 795 plots, where each plot was sampled an average 246 of 13.5 times. Plots were distributed across four combinations of past/current 247 management: grazed/grazed, ungrazed/ungrazed, grazed/ungrazed and 248 ungrazed/grazed. In 1998, additional plots were added in a fifth treatment with fences 249 excluding both large and small grazers (rodent exclusion). More information: 250 https://portal.lternet.edu/nis/metadataviewer?packageid=knb-lter-sgs.527.1. 251 252 19. The dataset is issued from an experiment located in sandy semi-natural grasslands 253 of the Elbe valley in Höhbeck, Germany (LAT: 53.05, and LON: 11.41). This dataset 254 consists of 96 plots, where each plot was sampled an average of 6.0 times. The 255 vegetation was surveyed once a year in 1 m2 plots using the Londo scale (Londo 256 1976). More information: Schuhmacher & Dengler (2013). 10 Page 11 of 147 Journal of Vegetation Science 257 258 20. The dataset is issued from an experiment located near Dufftown, Morayshire, 259 United Kingdom (LAT: 57.73, and LON: -3.10). This dataset consists of 12 plots, 260 where each plot was sampled an average of 6.0 times. Each species was measured in a 261 transect, using the inclined-point quadrat method (Tinney et al. 1937) (32·5° to the 262 horizontal). All contacts with 5 pins were recorded in 20 quadrat positions per plot. 263 More information: Pakeman et al. (2003). 264 265 21. The dataset is issued from an experiment located in Andrew Experimental forest 266 267 268 For Program (AND-LTER), Oregon, USA (LAT: 44.35, and LON: -122.41). This dataset consists of 193 plots, where each plot was sampled an average of 21.4 times. Plots were established in i) undisturbed, ii) logged, iii) logged and lightly burned, and iv) 269 logged and severely burned areas. In each plot, the cover values (%) were estimated. 270 More information: https://portal.lternet.edu/nis/metadataviewer?packageid=knb-lter- 271 and.3217.11. 272 273 22. The dataset is issued from an experiment located on woodlands, grasslands, and 274 shrublands in eastern Australia (LAT: -30.12, and LON: 147.17). This dataset consists 275 of 47 plots, where each plot was sampled an average of 10.2 times. In each plot, the 276 biomass of the vegetation was measured annually, from 1991 to 2002, in four 300 m 277 long transects each containing 13 quadrats of 0.72 m x 0.72 m. Dataset owners: James 278 Val and David Eldridge (Office of Environment & Heritage, University of New South 279 Wales). 280 281 23. The dataset is issued from an experiment located on a pasture in Fasque, United 282 Kingdom (LAT: 56.87, and LON: -2.60). This dataset consists of eight plots, where 283 each plot was sampled an average of 8.0 times. Inclined-point quadrat method (32·5° 284 to the horizontal) was used to record each species in a transect, with a minimum of 20 285 point contacts at 18 locations per plot (i.e. a minimum of 360 contacts per plot). More 286 information: Marriott et al. (2002). 287 288 24. The dataset is issued from an experiment located on La Fage French National 289 Institute for Agricultural Research (INRA) experimental station, close to Millau, 290 France (LAT: 43.92, and LON: 3.10). This dataset consists of 16 plots, where each 11 Journal of Vegetation Science Page 12 of 147 291 plot was sampled an average of 28.0 times. Individual plants were identified using the 292 point intercept method on 5 m permanent lines (1 point/10 cm, i.e. 50 points/line). 293 More information: Chollet et al. (2014) and Garnier et al. (2018). 294 295 25. The data sourced from BioTIME (Dornelas et al. 2018), Study_ID 483 and 497- 296 ITEX Dataset 5 - Teberda (Malaya Alpine-Snowbed and Geranium Hedysarum 297 Meadow) and ITEX Dataset 19 - Teberda (Festuca Varia Grassland, Malaya Alpine 298 Lichen-Heath). The dataset is issued from an experiment located in Teberda State 299 Reserve, a part of the Karachaevo-Cherkessian Republic in the northwestern 300 Caucasus, Russia (LAT: 43.45, and LON: 41.69). This dataset consists of 145 plots, 301 where each plot was sampled an average of 24.3 times. In each plot, the cover of each 302 plant species was recorded as number of shoots per m2. More information: 303 Onipchenko et al. (1998). 304 305 26. The dataset is issued from an experiment located a moorland in the Clocaenog 306 Forest, United Kingdom (LAT: 53.06, and LON: -3.47). This dataset consists of 9 307 plots, where each plot was sampled an average of 12.0 times. The experiment was 308 designed with three treatments: control, drought (~20% reduction in total annual 309 rainfall) and warming (~64% reduction in heat loss during night and 14% reduction in 310 total annual rainfall). Three quadrats per plot were chosen, and in each quadrat 311 vegetation was quantified using a grid of 100 pins (pin-point methodology). Pin hits 312 were then converted to biomass (g m-2) using a biomass calibration-conversion. More 313 information: https://catalogue.ceh.ac.uk/documents/5b39a644-d614-4f2b-8df6- 314 202ed440b4ab. Doi: https://doi.org/10.5285/5b39a644-d614-4f2b-8df6- 315 202ed440b4ab. 316 317 27. The dataset is issued from an experiment located on serpentine and non-serpentine 318 meadows in California, USA (LAT: 38.85, and LON: -123.50). This dataset consists 319 of 400 plots, where each plot was sampled an average of 10.0 times. In each plot, the 320 species cover (%) was visually estimated annually. More information: Fernandez- 321 Going et al. (2012) and Harrison (1999). 322 323 28. The dataset is issued from an experiment located on the Jornada Basin 324 Experimental Range JRN-LTER in the Chihuahuan desert, New Mexico, USA (LAT: 12 Page 13 of 147 Journal of Vegetation Science 325 32.62, and LON: -106.67). This dataset consists of 68 plots, where each plot was 326 sampled an average of 27.8 times. Density of individuals per species and per plot was 327 recorded. More information: 328 https://portal.lternet.edu/nis/metadataviewer?packageid=knb-lter-jrn.210351002.75. 329 330 29. The dataset is issued from an experiment located on a grassland in Krkonose 331 Mountains, Czech Republic (LAT: 50.69, and LON: 15.71). This dataset consists of 332 four plots, where each plot was sampled an average of 16.8 times. Standing biomass 333 was sampled annually. More information: Herben et al. (1997). 334 335 30. The dataset is issued from an experiment located on a grassland in Krkonose 336 Mountains, Czech Republic (LAT: 50.69, and LON: 15.79). This dataset consists of 337 four plots, where each plot was sampled an average of 29.8 times. Standing biomass 338 was sampled annually. More information: Herben et al. (2017). 339 340 31. The data sourced from BioTIME (Dornelas et al. 2018), Study_ID 243 - Virginia 341 Coast Reserve Long-Term Ecological Research. The dataset is issued from an 342 experiment located in the coastal sand dunes of Hog island, Virginia, USA (LAT: 343 37.67, and LON: -75.67). This dataset consists of 28 plots, where each plot was 344 sampled an average of 18.9 times. Half of the plots received nitrogen fertilization each 345 year in the form of urea nitrogen (30% uncoated (46-0-0) and 70% (40-0-0) coated for 346 slow release). The fertilizer was applied evenly in a dry form (15 g/m2 of nitrogen). In 347 each plot, species cover (%) was visually estimated in five 0.25 m2 plots. More 348 information: Day et al. (2016). 349 350 32. The dataset is issued from an experiment located on a grassland near Napal, Spain 351 (LAT: 42.72, and LON: -1.22). This dataset consists of 12 plots, where each plot was 352 sampled an average of 12.0 times. The experimental area was fenced and shrubs were 353 removed. Six plots were fertilized (sewage sludge to the soil surface with 5000 g/m2) 354 and six plots were used as controls. All vascular plant species were measured annually 355 using frequencies. To do so, each plot was divided into 100 subplots, and the 356 presence/absence of each species was recorded. More information: Gazol et al. 357 (2016). 358 13 Journal of Vegetation Science Page 14 of 147 359 33. The data were sourced from BioTIME (Dornelas et al. 2018), Study_ID 491 - 360 ITEX Dataset 13 - Toolik (Dry, Moist). The dataset is issued from an experiment 361 located on tundra vegetation near Toolik, Alaska, USA (LAT: 68.62, and LON: - 362 149.61). This dataset consists of eight plots, where each plot was sampled an average 363 of 6.0 times. The plots are divided between dry tundra with control and warming 364 treatments and moist tundra with only control treatment. Biomass estimates were 365 obtained using a fixed 75 cm2 point frame, with 100 measurements spaced 7 cm apart. 366 367 34. The data was sourced from BioTIME (Dornelas et al. 2018), Study_ID 492 - 368 ITEX Dataset 14 - Toolik (LTER Heath, LTER Moist acidic tussock, LTER non- 369 acidic tussock, LTER wet sedge, SAG wet sedge2, Tussock 1981 plots). The dataset 370 is issued from an experiment located in Toolik, Alaska, USA (LAT: 68.63, and LON: 371 -149.58). This dataset consists of four plots, where each plot was sampled an average 372 of 6.0 times. In each plot, species biomass was assessed by clipping of four or five 373 0.25 m x 0.25 m plots, and sorting to species level. 374 375 35. The dataset is issued from an experiment located on a grassland in Bayreuth, 376 Germany (LAT: 49.92, and LON: 11.59). This dataset consists of 15 plots, where 377 each plot was sampled an average of 7.7 times. Three treatments were applied: 1) 378 control (ambient condition), 2) winter warming (October–March), and 3) summer 379 warming (April–September). In each plot, species cover (%) was visually estimated 380 annually. More information: Grant et al. (2017). 381 382 36. The dataset is issued from an experiment located on a grassland in the Czech 383 Republic (LAT: 48.87, and LON: 16.64). This dataset consists of seven plots, where 384 each plot was sampled an average of 8.0 times. In each plot (1 m2), the species cover 385 (%) was visually estimated annually from 1993 to 2001. Dataset owner: Jiří Danihelka 386 (Department of Botany and Zoology, Masaryk University and Department of 387 Vegetation Ecology, Institute of Botany, The Czech Academy of Sciences). 388 389 37. The dataset is issued from an experiment located on a grassland in Laqueuille, 390 France (LAT: 45.64, and LON: 2.73). This dataset consists of 10 plots, where each 391 plot was sampled an average of 13.0 times. Half of the plots were located in an 392 intensively managed grassland (10-15 animals ha-1 yr-1 and 20 g/m2 of nitrogen), and 14 Page 15 of 147 Journal of Vegetation Science 393 the other half were located in a neighbouring grassland under extensive management 394 (5-8 animals ha-1 yr-1 and no fertilization). In each plot, presence/absence of each 395 species was recorded in 40 pin-points regularly spaced (pin-point methodology. 396 Dataset owner: Katja Klumpp (INRA, Grassland Ecosystem Research Unit). 397 398 38. The dataset is issued from an experiment located on Shortgrass Steppe (SGS- 399 LTER) in the Central Plains Experimental Range Nunn, Colorado, USA (LAT: 40.85, 400 and LON: -104.71). This dataset consists of 48 plots, where each plot was sampled an 401 average of 9.0 times. The experiment evaluated four treatments: control inside 402 exclosure, control outside exclosures, Bouteloua gracilis removal inside exclosure and 403 404 orF Bouteloua gracilis removal outside exclosure. Species density was measured in a quadrat (1 m2) using vegetation point intercept method (40 points of contact was 405 recorded for each quadrat). More information: 406 https://portal.lternet.edu/nis/mapbrowse?packageid=knb-lter-sgs.703.1. 407 408 39. The dataset is issued from an experiment located on a wet meadow in Ohrazeni, 409 Czech Republic (LAT: 48.95, and LON: 14.59). This dataset consists of 12 plots, 410 where each plot was sampled an average of 16.0 times. The experiment evaluated four 411 treatments: control, mowing (annually in the second half of June), fertilization (65 412 g/m2 of commercial NPK fertilizer) and dominant removal (Molinia caerulea plants 413 were manually removed annually). In each plot, the biomass of each species was 414 measured annually. More information: Lepš (2014). 415 416 40. The dataset is issued from an experiment (Long Term Experiment SOERE- 417 ACBB) located on a grassland in Theix, France (LAT: 45.72, and LON: 3.02). This 418 dataset consists of eight plots, where each plot was sampled an average of 8.0 times. 419 The experiment evaluated, on one hand, the effect of the intensity of grazing with two 420 treatments with cattle rotational grazing at high (Ca+) or low (Ca-) level of herbage 421 utilisation; these two treatments did not receive any mineral fertilisation. On the other 422 hand, it also evaluated the effect of nutrient availability, comparing two treatments 423 conducted under fixed cutting regime (three cuts/per year), one with fertilization 424 (NPK fertilizer) and the other without fertilization. The presence/absence of each 425 plant species was measured using 40 pin-points regularly spaced along fixed transects. 426 Complementarily, at each pin-point, 6 points are distributed to species according to 15 Journal of Vegetation Science Page 16 of 147 427 visual estimation of their volume. Dataset owner: Frédérique Louault (INRA-UREP). 428 More information: Louault et al. (2017). 429 430 41. The dataset is issued from an experiment belonging to the Sevilleta LTER and 431 located on Chihuahuan desert in Sevilleta National Wildlife Refuge, New Mexico, 432 USA (LAT: 34.27, and LON: -106.68). This dataset consists of six plots, where each 433 plot was sampled an average of 14.3 times. More information: 434 https://portal.lternet.edu/nis/mapbrowse?packageid=knb-lter-sev.200.174699. 435 436 42. The dataset is issued from an experiment located on hyper-oceanic coastal 437 grasslands in United Kingdom (LAT: 57.27, and LON: -7.40). This dataset consists of 438 48 plots, where each plot was sampled an average of 6.8 times. The experiment 439 evaluated six treatments: 1) vertebrate grazing exclusion, 2) burial box with no sand 440 added, 3) buried to 10 cm, 4) buried to 20 cm, 5) windbreak - shelter from prevailing 441 SW winds, 6) no treatment. The cover values (%) of all individual plant species were 442 estimated annually from 2004 to 2010. Data owners: Robin Pakeman (James Hutton 443 Institute. Aberdeen) and Jack J. Lennon (School of Biological Sciences, Queen's 444 University Belfast). 445 446 43. The dataset is issued from an experiment located on a grassland in Cleish and 447 Kirkton, United Kingdom (LAT: 56.29, and LON: -4.07). This dataset consists of 16 448 plots, where each plot was sampled an average of 6.0 times. The experiment evaluated 449 ungrazed and sheep-grazed plots to maintain three different levels of sward height. In 450 each plot, the inclined-point quadrat method (32·5° to the horizontal) at 20 locations 451 (with a minimum of 25 contacts per location) was used to measure each species. More 452 information: Hulme et al. (1999). 453 454 44. The dataset is issued from an experiment located on a grassland in Bell Hill and 455 Cleish, United Kingdom (LAT: 55.80, and LON: -2.84). This dataset consists of eight 456 plots, where each plot was sampled an average of 7.0 times. In each plot, the inclined- 457 point quadrat method (32·5° to the horizontal) at 20 locations (with a minimum of 25 458 contacts per location) was used to measure each species. More information: Grant et 459 al. (1996a). 460 16 Page 17 of 147 Journal of Vegetation Science 461 45. The dataset is issued from an experiment located on a grassland in Cleish and 462 Sourhope, United Kingdom (LAT: 55.81, and LON: -2.86). This dataset consists of 463 seven plots, where each plot was sampled an average of 6.0 times. There were 464 different treatments where cattle or sheep density was adjusted twice a week to 465 maintain the vegetation height between tussocks. In each plot, the inclined-point 466 quadrat method (32·5° to the horizontal) at 20 locations (with a minimum of 25 467 contacts per location) was used to measure each species. More information: Grant et 468 al. (1996) and Common et al. (1998). 469 470 46. The dataset is issued from an experiment located on a moorland previously on the 471 Burnhead heft at the Redesdale Experimental Farm in Northumberland, United 472 Kingdom (LAT: 55.37, and LON: -2.45). This dataset consists of 12 plots, where each 473 plot was sampled an average of 6.0 times. The 12 plots were divided in three areas 474 with different grazing treatments: ungrazed, sheep-grazed (three levels: 0.4, 0.8 and 475 1.2 ha-1 yr-1). In each plot, the inclined-point quadrat method (32·5° to the horizontal) 476 at 20 locations (with a minimum of 25 contacts per location) was used to measure 477 each species. More information: Pakeman & Nolan (2009). 478 479 47. The dataset is issued from an experiment located on a heather moorland at 480 Dundonnell near Ullapool and at Claonaig, near Tarbert Loch Fyne, Argyll and Bute, 481 United Kingdom (LAT: 57.35, and LON: -5.55). This dataset consists of 17 plots, 482 where each plot was sampled an average of 6.0 times. The experiment had different 483 sheep grazing and exclusion treatments: 1) low at 0.4 sheep ha-1 yr-1, 2) moderate at 484 0.8 sheep ha-1 yr-1, 3) high at 1.2 sheep ha-1 yr-1, 4) fenced against both cattle and 485 sheep, and 5) fenced against cattle, also 6) sheep and cattle recorded from the open 486 hill. In each plot, the inclined-point quadrat method (32·5° to the horizontal) at 20 487 locations was used to measure each species. More information: Pakeman & Nolan 488 (2009). 489 490 48. The dataset is issued from an experiment located on a grassland in the Ordesa- 491 Monte Perdido National Park, Spain (LAT: 42.67, and LON: -0.06). This dataset 492 consists of four plots, where each plot was sampled an average of 19.0 times. The 493 point intercept method at 20 locations was used to measure each species. 17 Journal of Vegetation Science Page 18 of 147 494 In each plot, the point intercept method was used annually to measure vegetation 495 along two perpendicular transects (a total of 400 sample points). More information: 496 Pardo et al. (2015). 497 498 49. The dataset is issued from an experiment located in Soto de Viñuelas, Spain 499 (LAT: 40.60, and LON: -3.63). This dataset consists of 68 plots, where each plot was 500 sampled an average of 11.5 times. In each plot, all plant species was recorded using 501 presence/absence data in five quadrats of 400 cm2 each from 1980 to 1995. Dataset 502 owner: Begoña Peco (Ecology Department Autonomous, University of Madrid). 503 504 50. The dataset is issued from an experiment located on a shrubland in Garraf, Spain 505 (LAT: 41.30, and LON: 1.82). This dataset consists of nine plots, where each plot was 506 sampled an average of 17.0 times. Three experiment evaluated three treatments: 1) 507 control, 2) warming (metallic curtain covering the plots during the night), and 3) 508 drought (transparent curtain covering the plots during rainfall). Number of contacts 509 per plot was used to quantify each species. Dataset owners: Josep Penuelas, Marc 510 Estiarte and Romà Ogaya (Global Ecology Unit CREAF-CSIC-UAB). 511 512 51. The dataset is issued from an experiment belong to the Jornada LTER (JRN- 513 LTER) and located in Chihuahuan desert, Jornada Basin Experimental Range, New 514 Mexico, USA (LAT: 32.00, and LON: -106.00). This dataset consists of 734 plots, 515 where each plot was sampled an average of 24.0 times. In each plot, the biomass of 516 each species was calculated from field measurement of individual species cover and 517 height. More information: 518 https://portal.lternet.edu/nis/metadataviewer?packageid=knb-lter-jrn.2100011001.49. 519 520 52. The dataset is issued from an experiment located on a moorland on the Burnhead 521 heft at the Redesdale Experimental Farm in Northumberland, United Kingdom (LAT: 522 55.37, and LON: -2.45). This dataset consists of 10 plots, where each plot was 523 sampled an average of 6.0 times. The experiment had different grazing treatments: 524 summer grazing, winter grazing or year-round grazing (0.7 sheep ha-1 yr-1), year- 525 round grazing (1.4 sheep ha-1 yr-1), and no grazing. In each plot, the inclined-point 526 quadrat method (32·5° to the horizontal) at 20 locations (with a minimum of 25 18 Page 19 of 147 Journal of Vegetation Science 527 contacts per location) was used to measure each species. More information: Hulme et 528 al. (2002) and Pakeman & Nolan (2009). 529 530 53. The dataset is issued from an experiment located on moorlands in Derbyshire, 531 United Kingdom (LAT: 54.69, and LON: -2.41). This dataset consists of 216 plots, 532 where each plot was sampled an average of 10.0 times. The experiment evaluated 36 533 treatments: no treatment; cut once per year; cut twice per year; herbicide sprayed; 534 herbicide sprayed in first year, cut in second; and cut in first year, sprayed in second. 535 Within each of these main plot treatments there were two sub-plot grazing treatments 536 - sheep grazing and no sheep grazing. Finally, there were three restoration treatments 537 applied at the sub-sub-plot level: untreated, Calluna moorland litter applied as litter, 538 and Calluna vegetation applied as cut brash. All these 36 treatments had 6 replicates. 539 In each plot, the species composition was recorded using point-quadrats (1 m-long 540 frame with 10 pin positions at 10 cm intervals, pin diameter = 2 mm). Dataset owner: 541 Rob Marrs (University of Liverpool). 542 543 54. The dataset is issued from an experiment belonging to the Environmental Change 544 Network (ECN) and located in the United Kingdom (LAT: 53.95, and LON: -3.23). 545 This dataset consists of 198 plots, where each plot was sampled an average of 11.1 546 times. In each plot (ten quadrats of 0.16 m2), the inclined-point quadrat method was 547 used to evaluate the vegetation annually. More information: Rennie et al. (2016) and 548 https://catalogue.ceh.ac.uk/documents/b98efec8-6de0-4e0c-85dc-fe4cdf01f086 and 549 https://catalogue.ceh.ac.uk/documents/d349babc-329a-4d6e-9eca-92e630e1be3f. 550 551 55. The dataset is issued from an experiment belonging to the Andrews Forest LTER 552 (AND-LTER) and located in a forest in the Oregon Cascade Range, USA (LAT: 553 44.22, and LON: -122.25). This dataset consists of 5 plots, where each plot was 554 sampled an average of 10.0 times. The vegetation cover (%) was visually estimated 10 555 times in a quadrat of 4 m2 for trees (vegetation > 60 cm tall) and 9 quadrats (0.1 m2) 556 for herb and low shrub (< 60 cm tall). More information: Rothacher (Rothacher 2013) 557 and https://portal.lternet.edu/nis/mapbrowse?packageid=knb-lter-and.3190.7. 558 559 56. The dataset is issued from an experiment belonging to the Park Grass permanent 560 grassland and located in Rothamsted, United Kingdom (LAT: 51.81, and LON: - 19 Journal of Vegetation Science Page 20 of 147 561 0.37). This dataset consists of 74 plots, where each plot was sampled an average of 562 9.9 times. The purpose of the experiment was to evaluate different fertility and lime 563 treatments. Herbage was taken from six randomly located quadrats measuring 0.5 m x 564 0.25 m within each plot, resulting in a total sampling area of 0.75 m2 within each plot. 565 In each plot, the biomass of each species was measured annually in quadrats 566 (sampling area: 0.75 m2). More information: Crawley et al. (2005) and 567 http://www.era.rothamsted.ac.uk/Park. 568 569 57. The dataset is issued from an experiment located on a savannah in central Spain 570 (LAT: 40.38, and LON: -4.20). This dataset consists of 210 plots, where each plot was 571 sampled an average of 6.0 times. The experiment evaluated two types of pastures 572 (higher-productivity pastures and low-productivity pastures) and three treatments 573 (ungrazed, grazed by small herbivores, and grazed by large and small herbivores). In 574 each plot, the species cover (%) was visually estimated. More information: Rueda et 575 al. (2013). 576 577 58. The dataset is issued from an experiment located in Central Germany (LAT: 578 51.55, and LON: 10.07). This dataset consists of 14 plots, where each plot was 579 sampled an average of 14.9 times. In each plot, species vegetation cover (%) was 580 visually estimated. More information: Schmidt (2007). 581 582 59. The dataset is issued from an experiment located on a former arable field in the 583 Experimental Botanical Garden of the University of Göttingen, Germany (LAT: 584 51.56, and LON: 9.96). This dataset consists of six plots, where each plot was 585 sampled an average of 38.0 times. In each plot, species vegetation cover (%) was 586 visually estimated. More information: Schmidt (Schmidt 2006) and Bernhardt- 587 Römermann et al. (2011). 588 589 60. The dataset is issued from an experiment located in the Swiss National Park 590 (IUCN Ia reserve, LAT: 46.68, and LON: 10.22). This dataset consists of 41 plots, 591 where each plot was sampled an average of 12.2 times. In each plot, plant species 592 cover (%) was visually estimated. More information: Braun-Blanquet et al. (1931), 593 Schütz et al. (2000). 594 20 Page 21 of 147 Journal of Vegetation Science 595 61. The dataset is issued from an experiment belonging to the Sevilleta LTER (SEV- 596 LTER) and located in Sevilleta National Wildlife Refuge, New Mexico, USA (LAT: 597 34.31, and LON: -106.49). This dataset consists of 95 plots, where each plot was 598 sampled an average of 9.8 times. The experiment was designed to evaluate the effect 599 of prescribed burning (two areas were left unburned as control treatments, and the 600 other plots were burned in different dates) and grazing exclusion (fenced and 601 unfenced). In each plot, the individuals present in 36 quadrats (0.1 m2) were recorded. 602 More information: https://portal.lternet.edu/nis/mapbrowse?packageid=knb-lter- 603 sev.148.131885. 604 605 62. The dataset is issued from an experiment belonging to the Sevilleta LTER (SEV- 606 LTER) and located in Sevilleta National Wildlife Refuge, New Mexico, USA (LAT: 607 34.33, and LON: -106.74). This dataset consists of 81 plots, where each plot was 608 sampled an average of 9.2 times. The experiment had three treatments: 1) control 609 plots (natural rainfall regime) 2) drought was induced by rainfall shelters, and 3) 610 watering was applied by redirecting the water from the nearby rainfall shelters. In 611 each plot, the plant cover (%) was estimated every spring. More information: 612 https://portal.lternet.edu/nis/mapbrowse?packageid=knb-lter-sev.147.167839. 613 614 63. The dataset is issued from an experiment belonging to the Sevilleta LTER (SEV- 615 LTER) and located on a grassland in Sevilleta National Wildlife Refuge, New 616 Mexico, USA (LAT: 34.33, and LON: -106.63). This dataset consists of 216 plots, 617 where each plot was sampled an average of 7.7 times. The experiment evaluated the 618 impact of prairie dog reintroduction (grazed and ungrazed areas) on vegetation. In 619 each plot, the plant cover (%) was estimated annually. More information: 620 https://portal.lternet.edu/nis/metadataviewer?packageid=knb-lter-sev.212.4. 621 622 64. The dataset is issued from an experiment belonging to the Sevilleta LTER (SEV- 623 LTER) and located on a woodland in Sevilleta National Wildlife Refuge, New 624 Mexico, USA (LAT: 34.37, and LON: -106.54). This dataset consists of 100 plots, 625 where each plot was sampled an average of 13.0 times. In each plot, the plant cover 626 (%) was visually estimated annually. More information: 627 https://portal.lternet.edu/nis/mapbrowse?packageid=knb-lter-sev.278.245672. 628 21 Journal of Vegetation Science Page 22 of 147 629 65. The dataset is issued from an experiment belonging to the Sevilleta LTER (SEV- 630 LTER) and located in Sevilleta National Wildlife Refuge, New Mexico, USA (LAT: 631 34.37, and LON: -106.58). This dataset consists of 100 plots, where each plot was 632 sampled an average of 16.4 times. The experiment evaluated three treatments: 1) 633 control plots (untouched vegetation), 2) removal of all three dominant species (Larrea 634 tridentata, Bouteloua eriopoda, Bouteloua gracilis), and 3) removal of one dominant 635 species. In each plot, the plant cover (%) was visually estimated annually. More 636 information: https://portal.lternet.edu/nis/mapbrowse?packageid=knb-lter- 637 sev.168.192543. 638 639 66. The dataset is issued from an experiment belonging to the Shortgrass Steppe 640 LTER (SGS-LTER) and located on grasslands and shrublands in Central Plains 641 Experimental Range, Colorado, USA (LAT: 40.85, and LON: -104.77). This dataset 642 consists of 18 plots, where each plot was sampled an average of 8.2 times. In each 643 plot, the plant cover was recorded on three permanent transects (1 m2: sum of plots 644 along the transect). More information: Stapp (Stapp 2013) and 645 https://portal.lternet.edu/nis/mapbrowse?packageid=knb-lter-sgs.140.17. 646 647 67. The dataset is issued from an experiment located in a beech forest near Göttingen, 648 Central Germany (LAT: 51.57, and LON: 10.32). This dataset consists of seven plots, 649 where each plot was sampled an average of 18.0 times. Four plots had a fertilization 650 treatment (NP addition) and three were the control plots. In each plot, the species 651 cover (%) was visually estimated. More information: Schmidt (2009). 652 653 68. The dataset is issued from an experiment located on a meadow near Zvíkov, 654 Czech Republic (LAT: 48.99, and LON: 14.61). This dataset consists of 40 plots, 655 where each plot was sampled an average of 10.3 times. The experiment evaluated four 656 treatments: 1) control (intact vegetation), 2) mycorrhizal grasses and forbs left, non- 657 mycorrhizal species weeded out, 3) mycorrhizal forbs remaining, everything else 658 weeded out, and 4) mycorrhizal grasses remaining, everything else weeded out. In 659 each plot, the species cover (%) was visually estimated annually. More information: 660 Šmilauer & Šmilauerová (2013). 661 22 Page 23 of 147 Journal of Vegetation Science 662 69. The dataset is issued from an experiment located on a floodplain grassland in 663 Anloo and Taarlo, The Netherlands (LAT: 53.05, and LON: 6.66). This dataset 664 consists of 80 plots, where each plot was sampled an average of 28.9 times. In each 665 plot, the species cover (%) was estimated almost every year from 1973 to 2008. 666 Dataset owners: Christian Smit and Jan P. Bakker (Conservation Ecology Group, 667 Groningen Institute for Evolutionary Life Sciences). 668 669 70. The dataset is issued from an experiment located on a meadow in the north-eastern 670 Tibetan Plateau in Qinghai Province, China (LAT: 37.62, and LON: 101.20). This 671 dataset consists of 30 plots, where each plot was sampled an average of 9.0 times. The 672 experiment was designed to evaluate 10 nitrogen treatments (no N added and 9 673 combinations of three N forms and three N rates). In each plot, the species cover (%) 674 was visually estimated annually. More information: Song et al. (2012). 675 676 71. The dataset is issued from an experiment located on salt marshes of the 677 Schleswig-Holstein Wadden Sea National Park in Hamburger Hallig and 678 Westerhever, Germany (LAT: 54.49, and LON: 8.75). This dataset consists of 212 679 plots, where each plot was sampled an average of 18.7 times. There were two 680 treatments in Westerhever: natural condition and intensive grazing, and only natural 681 conditions in Hamburger Hallig. In each plot, the species cover was measured 682 annually using the Londo scale (percentage of vegetation cover) from 1997 to 2015 in 683 Hamburger Hallig and from 1995 to 2012 in Westerhever. Dataset owner: Martin 684 Stock (Wadden Sea National Park of Schleswig-Holstein). 685 686 72. The dataset is issued from an experiment located on a wooded savanna in 687 Laikipia, Kenya (LAT: 0.28, and LON: 36.87). This dataset consists of 18 plots, 688 where each plot was sampled an average of 14.7 times. The treatments were six 689 combinations (3 replicates) of cattle, wildlife, and mega-herbivore grazing. These 690 either allowed (1) the entry of all large mammalian herbivores, (2) all large 691 mammalian herbivores except mega-herbivores (elephants Loxodonta africana and 692 giraffe Giraffa camelopardis) to enter, or (3) excluded all large herbivores. In each 693 plot, vegetation was assessed annually by counting the number of pins hit by each 694 species over a ten-point pin frame at each station. More information: Veblen et al. 695 (2016). 23 Journal of Vegetation Science Page 24 of 147 696 697 73. The dataset is issued from an experiment located on a coastal heathland in 698 Lurekalven, Norway (LAT: 60.70, and LON: 5.08). This dataset consists of 42 plots, 699 where each plot was sampled an average of 6.0 times. In each plot, all vascular plants, 700 bryophytes and lichens were recorded annually using frequencies (1 m x 1 m metal 701 frame divided into 16 subplots). More information: Vandvik et al. (2005). 702 703 74. The dataset is issued from an experiment located in Bonanza Creek LTER, 704 Alaska, USA (LAT: 65.00, and LON: -148.00). This dataset consists of 59 plots, 705 where each plot was sampled an average of 12.0 times. In each plot, the species cover 706 (%) was visually estimated. More information: Viereck et al. (2010) and 707 https://portal.lternet.edu/nis/mapbrowse?packageid=knb-lter-bnz.174.19. 708 709 75. The dataset is issued from an experiment located on desert steppes in Gobi Gurvan 710 Saykhan National Park, Mongolia (LAT: 43.61, and LON: 104.13). This dataset 711 consists of 18 plots, where each plot was sampled an average of 7.1 times. The 712 experiment evaluated two treatments: 1) exclusion of large ungulates, and 2) no 713 exclusion of large ungulates. In each plot, the species cover (%) was visually 714 estimated annually. More information: Wesche et al. (2010). 715 716 76. The 9dataset is issued from an experiment located on a floodplain grassland on 717 formerly arable land (LAT: 51.78, and LON: -1.31). From 1989 the site was divided 718 into nine plots of c. 0.4 ha over which three contrasting grazing management practices 719 (control, cattle and sheep) were randomly superimposed. These nine plots were 720 monitored in June of each year from 1991-2009. More information: Woodcock et al. 721 (2011). 722 723 77. The dataset is issued from an experiment located on a grassland in southeast 724 Estonia (LAT: 58.11, and LON: 27.07). This dataset consists of 55 plots, where each 725 plot was sampled an average of 8.9 times. The treatments were: fertilizer, sucrose and 726 control. In each plot, the species cover (%) was visually estimated annually. More 727 information: Liira et al. (2012). 728 24 Page 25 of 147 Journal of Vegetation Science 729 Figure Table S2. Characteristics of the study sites. 730 ID Country Biome Habitats Duration CP MP 1 USA TGD Grassland 12.5 NO YES 2 USA WS Grassland 34.5 YES YES 3 USA TGD Grassland 13.5 YES YES 4 USA SD Grassland 11.2 YES YES 5 USA TGD Savanna 21.5 YES YES 6 USA TGD Grassland 8.0 YES YES 7 USA TGD Grassland and Shrubland 11.5 YES YES 8 South Africa SD Grassland 10.7 YES NO 9 South Africa SD Savanna 15.8 YES NO 10 South Africa TGD Savanna 14.7 YES NO 11 South Africa SD Savanna 16.0 YES NO 12 Hungary WS Savanna 14.5 YES YES 13 USA TF Grassland 7.0 YES YES 14 USA TF Savanna 6.2 YES YES 15 USA TF Grassland 24.8 YES NO 16 USA TF Grassland 22.0 YES YES 17 USA TF Grassland 14.8 YES YES 18 USA TGD Grassland 13.5 YES YES 19 Germany WS Grassland 6.0 YES NO 20 United Kingdom WS Shrubland 6.0 NO YES 21 USA TRF Forest 21.4 YES YES 22 Australia SD and WS Savanna 10.2 YES NO 23 United Kingdom WS Grassland 8.0 YES YES 24 France WS Grassland 28.0 NO YES 25 Russia BF Grassland 24.3 YES NO 26 United Kingdom TF Shrubland 12.0 YES YES 27 USA TF Grassland 10.0 YES NO 28 USA TGD Grassland 27.8 YES NO 29 Czech Republic TF Grassland 16.8 YES NO 30 Czech Republic TF Grassland 29.8 YES NO 31 USA WS Grassland 18.9 YES YES 32 Spain WS Grassland 12.0 YES YES 33 USA Tu Grassland 6.0 YES YES 34 USA Tu Grassland 6.0 YES NO 35 Germany WS Grassland 7.7 YES YES 36 Czech Republic WS Grassland 8.0 YES NO 37 France TF Grassland 13.0 NO YES 38 USA TGD Grassland 9.0 YES YES 39 Czech Republic WS Grassland 16.0 YES YES 40 France WS Grassland 8.0 YES YES 41 USA TGD Grassland, Shrubland and Savanna 14.3 YES NO 42 United Kingdom TF Grassland 6.8 YES YES 43 United Kingdom TF and TRF Grassland 6.0 YES YES 25 Journal of Vegetation Science Page 26 of 147 44 United Kingdom TF Grassland 7.0 YES YES 45 United Kingdom TF Grassland 6.0 YES YES 46 United Kingdom TF Shrubland 6.0 NO YES 47 United Kingdom TF Savanna 6.0 YES YES 48 Spain BF Grassland 19.0 YES YES 49 Spain TGD Grassland 11.5 YES NO 50 Spain WS Shrubland 17.0 YES YES 51 USA TGD Grassland, Shrubland and Savanna 24.0 YES NO 52 United Kingdom TF Savanna 6.0 NO YES 53 United Kingdom TF Shrubland 10.0 YES YES 54 United Kingdom TF and WS Grassland, Savanna and Forest 11.1 YES NO 55 USA TF Forest 10.0 YES NO 56 United Kingdom WS Grassland 9.9 YES YES 57 Spain TGD Savanna 6.0 YES YES 58 Germany WS Grassland 14.9 YES NO 59 Germany WS Grassland 38.0 NO YES 60 Switzerland BF Grassland and Forest 12.2 NO YES 61 USA TGD Savanna 9.8 YES YES 62 USA TGD Grassland, Shrubland and Savanna 9.2 YES YES 63 USA TGD Grassland 7.7 YES YES 64 USA TGD Forest 13.0 YES NO 65 USA TGD Grassland and Savanna 16.4 YES YES 66 USA TGD Grassland and Shrubland 8.2 YES NO 67 Germany WS Forest 18.0 YES YES 68 Czech Republic WS Grassland 10.3 YES YES 69 Netherlands WS Grassland 28.9 NO YES 70 China WS Grassland 9.0 YES YES 71 Germany WS Salt marsh 18.7 YES YES 72 Kenya WS Savanna 14.7 NO YES 73 Norway TRF Grassland 6.0 NO YES 74 USA BF Grassland and Savanna 12.0 YES NO 75 Mongolia TGD Grassland 7.1 YES YES 76 United Kingdom WS Grassland 18.0 NO YES 77 Estonia WS Grassland 8.9 YES YES 731 732 ID: Identification of the data set, biomes (TGD: temperate grassland desert, SD: 733 subtropical desert, WS: woodland shrubland, TF: temperate forest, BF: boreal forest, 734 Tu: Tundra, and TRF: temperate rain forest), Duration: Average number of years of 735 the dataset, CP: presence of plots where the long-term conditions prior to the 736 establishment of the sampling scheme were maintained throughout the sampling, MP: 737 presence of plots exposed to different treatments that altered the long-term conditions. 26 Page 27 of 147 Journal of Vegetation Science 738 Supporting Information to the paper 739 Valencia et al. Directional trends in species composition over time can lead to a 740 widespread overestimation of asynchrony. Journal of Vegetation Science. 741 742 Appendix S3. Application of the analyses shown in Fig. 1 of the main text to the 743 three remaining indices of synchrony. 744 745 Gross 746 Note: on panel (b) the mean synchrony values with the T3 approach per datasets are 747 significantly higher than without the T3 approach (p < 0.001, paired t-test) 748 749 750 751 Logvar 752 Note: on panel (b) the mean synchrony values with the T3 approach per datasets are 753 significantly higher than without the T3 approach (p < 0.06, paired t-test) 754 755 756 757 27 Journal of Vegetation Science Page 28 of 147 758 Phi 759 Note: on panel (b) the mean synchrony values with the T3 approach per datasets are 760 significantly higher than without the T3 approach (p < 0.22, paired t-test) 761 762 28 Page 29 of 147 Journal of Vegetation Science 763 Supporting Information to the paper 764 Valencia et al. Directional trends in species composition over time can lead to a 765 widespread overestimation of asynchrony. Journal of Vegetation Science. 766 767 Appendix S4. Application of the analyses shown in Fig. 3 of the main text to the 768 three remaining indices of synchrony. For each index, also, a table of number of 769 datasets with either positive or negative significant t-statitstic values is reported for 770 both manipulated and control plots (positive means that the T3 approach increased 771 synchrony; negative means that the T3 approach decreased synchrony). The grey area 772 in each panel reports and approximate are where t-statistic values were not significant 773 (‘ns’). 774 775 Gross For manipulated 'control' 25 25 (a) (b) 20 20 Review 15 15 T−statistic (synT3 − syn) 10 10 T3>synch 5 5 ns 0 0 T3synch 5 5 ns 0 0 For T3synch 5 5 ns 0 0 T3 0.2). This implies that the use of the T3 448 detrending approach did not systematically produce greater or weaker correlations 449 when analyzing these common relationships. 20 Page 21 of 147 Journal of Vegetation Science 450 Discussion 451 In this study we show that the synchrony patterns usually attributed to compensatory 452 dynamics could be actually caused by trends in species composition. Without 453 accounting for these trends effectively, it is possible that compensatory effects could be 454 generally overemphasized (in 30% of our datasets) or even underemphasized (in 10% 455 of our datasets). Previous studies of synchrony and compensatory dynamics have often 456 overlooked the possible effects of directional trends on the studied communities. Only 457 few studies, such as Vasseur and Gaedke (2007), Loreau & de Mazancourt (2008) and 458 Tredennick et al. (2017), have effectively filtered out species trends (using wavelet 459 based methods or considering growth rates of species in time, instead of raw 460 abundances). Long-term trends in abundances, either directional or cyclical, indeed 461 have the potential to bias the interpretation of synchrony with the most commonly used 462 indices. The T3 detrending approach can account for this bias (see simulation in Lepš 463 et al. 2019 and in Appendix S1). The advantages of the T3 approach, compared to other 464 approaches, are its lower data requirement and consideration of all species in a 465 community, not just the most frequent ones (Lepš et al. 2019). Using the T3 approach 466 across a large global dataset, shows that species’ directional trends in time can often 467 result in overemphasising year-to-year asynchrony, especially in rather long time series, 468 and when analysing manipulative experiments. 469 In ~40% of the datasets, and in the overall model across all plots, synchrony 470 using the T3 detrending approach was significantly greater than synchrony without 471 using it (Fig. 2). The ~40% estimate is, furthermore, a conservative one as we account 472 for Type I errors. Overall, the mean values of synchrony computed with the T3 473 detrending approach were higher than without it in the majority of cases, both within 474 and across datasets (Fig. 1b, and LMM). This is an important finding because it suggests 21 Journal of Vegetation Science Page 22 of 147 475 that our appreciation of the importance of asynchrony, and therefore compensatory 476 dynamics, may have been possibly overestimated, leading to wrong conclusions about 477 synchrony-asynchrony in communities. These findings highlight the necessity of 478 evaluating the effects of possible directional trends on synchrony to accurately estimate 479 the importance of ecological mechanisms regulating compensatory dynamics. The 480 difference between the indices calculated using T3 detrending approach and without it 481 were higher in grasslands and meadows, possibly because in the absence of slow- 482 growing, less dynamic, woody species. In these communities temporal trends can thus 483 be more easily detected compared to other types of vegetation. The increase in 484 synchrony after detrending also suggests the presence of opposite trends of species 485 abundances in time, such as when one species is decreasing steadily and another 486 increasing. For example, trends could be the result of species responding differently to 487 disturbance or to an increase in nutrient availability. Such opposite trends could be 488 monotonic or following waves in time (Wu et al. 2007), e.g. resulting from periodic 489 climate events such as “El Niño”, or intrinsic cycling of particular functional groups 490 such as legumes (Herben et al. 2017). These results are partially expected because our 491 datasets comprised natural or semi-natural well-established plant communities but 492 included experimental conditions in which changes in abundance or composition of 493 species are common. 494 When considering datasets with both control and manipulated plots (~50% of 495 the datasets) the effect of the T3 approach was more frequently significant in 496 manipulated plots than in control plots (Fig. 3). These plots were more prone to be 497 affected by a directional trend promoted by the specific manipulation imposed. This 498 result agrees with our hypothesis that events like soil-nutrient alteration (e.g. by 499 fertilization) and recovery from disturbance might promote directional trends. This 22 Page 23 of 147 Journal of Vegetation Science 500 result was expected as some of the experimental manipulations were designed to 501 directly alter species composition, in order to test their effects on community 502 synchrony. However, such prompted changes, often due to colonization-competition 503 trade-offs in species composition, can mask year-to-year fluctuations, and hence these 504 experiments should disentangle these biologically different effects on synchrony. For 505 these reasons, we recommend that any index of synchrony should be computed with 506 and without the T3 approach to properly evaluate the corresponding effects of long- 507 term experimental treatments and year-to-year fluctuations. Our result reinforces the 508 assumption that the effect of the T3 approach could be stronger in changing 509 environments/communities and the combination of indices with and without the T3 510 approach can be important to distinguish the mechanisms causing differential long-term 511 species responses to changes in environmental conditions from the differential species 512 responses to short-term species fluctuations on synchrony/asynchrony relationships. 513 The effect of detrending on synchrony values was particularly pronounced in 514 the case of succession. During succession the majority of species will increase their 515 abundance, which will cause them to be ultimately positively correlated in time. 516 However, these same species can compensate each other or vary independently on a 517 year-by-year basis, even if they all generally increase in time, so the existing synchrony 518 indices would tend to overestimate their actual year-to-year synchrony between species 519 within such communities. In fact, among the seven datasets with a Rho below 0.1 (Fig. 520 1a), the majority were characterised by being exposed to intense disturbance regimes 521 that triggered some type of successional process. For instance, plots of four datasets 522 had been exposed to a fire before or during the experiment, and two evaluated the effect 523 of herbivory exclusion (where the reduction in grazing intensity allowed the 524 development of higher vegetation like shrubs and trees). Both treatments are good 23 Journal of Vegetation Science Page 24 of 147 525 examples of environmental conditions promoting species directional trends (Pardo et 526 al. 2015) and thus affect synchrony values. 527 Interestingly, the effect of the T3 approach on the synchrony measured in 528 manipulated plots depended on the period length of the sampling scheme. Manipulated 529 plots sampled over longer time periods revealed higher synchrony values when using 530 the T3 detrending approach (Fig. 3). In other words, the longer is the sampling period 531 the greatest chance that there is a difference between T3 synchrony and synchrony 532 without T3 in manipulated plots. Longer time series likely increased the chances that 533 some species will have opposite trends in response to manipulation, with some 534 increasing over time and others decreasing. In a shorter time series, on the contrary, the 535 time lag in species responses (particularly extinction debt, Helm, Hanski, & Partel 536 2006; Lepš 2014) could cause that some species increase quickly in response to 537 manipulation, while others might respond more slowly. The T3 detrending approach, 538 therefore, will affect those species with a similar temporal trend in response to short- 539 term manipulations. Consequently, the duration of the sampling period stands out as a 540 key factor in the evaluation of temporal dynamics. We showed that, in the case of 541 manipulated communities, classical methods tended to overestimate year-to-year 542 synchrony when the sampling period was shorter, and underestimate it when the 543 sampling period was longer. This highlights the importance of T3 approach for a correct 544 evaluation of year-to-year synchrony between species. However, further research is 545 required to find the causes and consequences of these results. 546 Finally, we generally found that the T3 detrending approach did not cause strong 547 changes in the correlation between synchrony and both species richness and community 548 stability, two of the most iconic relationships in temporal dynamics studies (Hautier et 549 al. 2014; Blüthgen et al. 2016). However, there were more cases of significant 24 Page 25 of 147 Journal of Vegetation Science 550 correlations with the T3 approach and strength of the correlations could vary 551 considerably (i.e. R < 0.6) across datasets. In summary, this suggests that while the 552 applications of the T3 detrending approach did not produce systematically greater or 553 weaker correlations on commonly used tests in ecology, the strength of the relationships 554 could differ. These results confirm that the use of T3 approach to detrend the synchrony 555 indices is far from trivial. As such, the conclusions obtained previously from studies 556 that did not apply the method are not necessarily incorrect. Therefore, applying the 557 detrended and non-detrended methods in a complementary way might bring us closer 558 to understanding the directional changes in community dynamics. For instance, 559 divergent trends, e.g. due to differential response to global warming with some species 560 increasing and other decreasing, might stabilize communities and could maintain 561 ecosystem functions unaltered in response to global warming, even if there are no short- 562 term compensatory mechanisms between species. Hence, it is important to consider 563 both the synchrony with and without detrending approach for teasing apart different 564 causes of stability, or instability, in response to global change drivers. 565 The evaluation of synchrony with the T3 detrending method provides a feasible 566 measure to reveal year-to-year fluctuations of species by removing the effect of 567 directional trends. In comparison to methods using species growth rates, the T3 568 approach can be important because it enables the evaluation of the indices with and 569 without the approach and also accounts for species which are not dominant and/or less 570 frequent (in the case of the growth rates, log-transformation is needed, which might not 571 be advisable in the case of zero abundances in specific years). This method has the 572 advantage of evaluating both monotonic and non-monotonic directional trends, and can 573 thus be used to detect year-to-year fluctuations in the face of cyclical periods, such as 574 alternation between drought-wet periods (e.g. Riginos et al. 2018). 25 Journal of Vegetation Science Page 26 of 147 575 Acknowledgements 576 We thank multiple entities for the financial support necessary to obtain the different 577 databases: the U.S. National Science Foundation under grant numbers DEB-8114302, 578 DEB-8811884, DEB-9411972, DEB-0080382, DEB-0620652, DEB-1234162, DEB- 579 9707477, DEB-0316402, DEB-08-16453, and DEB-12-56034, DEB-0618210, the 580 Nutrient Network (http://www.nutnet.org) experiment from the National Science 581 Foundation Research Coordination Network (NSF-DEB-1042132), the New Zealand 582 National Vegetation Survey Databank, the Spanish MINECO (Project CGL2014- 583 53789-R), the Madrid Regional Government (Projects REMEDINAL-3 and 584 REMEDINAL-TE), the European Research Council Synergy grant 610028 585 (IMBALANCE-P), the Institute on the Environment (DG-0001-13), the SOERE- 586 ACBB financed through French National Agency for Research (ANAEE, ANR-11- 587 INBS-0001), the Estonian Research Council (IUT 20-28, IUT 20-29), Czech Science 588 Foundation (GAČR 17-05506S), the European Regional Development Fund (Centre of 589 Excellence EcolChange), the German Federal Environmental Foundation (DBU) for a 590 grant to the NABU Hamburg (management experiment Calamagrostis epigejos), and 591 the German Federal Ministry of Education and Research within the framework of the 592 project BIOTA Southern Africa (promotion numbers 01LC0024, 01LC0024A and 593 01LC0624A2), Task 159 of SASSCAL (promotion number 01LG1201) and the 594 Scottish Government’s Rural and Environmental Science and Analytical Services 595 division. Acknowledgement Data owned by NERC© Database Right/Copyright NERC. 596 Further support was provided by the Jornada Basin Long-Term Ecological Research 597 (LTER) project, Cedar Creek Ecosystem Science Reserve and the University of 598 Minnesota. We also thank the Lawes Agricultural Trust and Rothamsted Research for 599 data from the e-RA database. The Rothamsted Long-term Experiments National 600 Capability (LTE-NCG) is supported by the UK Biotechnology and Biological Sciences 601 Research Council (Grant BBS/E/C/000J0300) and the Lawes Agricultural Trust. 602 603 604 Author contributions: F.B., L.G. and J.L. conceived the project. All authors but E.V., 605 F.B., T.G and L.G. collected the data used in this analysis. E.V. and T.G. assembled 606 data. F.B. performed the analyses. E.V. and F.B. wrote the first draft of the manuscript 607 and all the authors (especially L.G. and J. L.) contributed substantially to the revisions. 608 609 Data accessibility 610 The data that support the findings of this study are available at Figshare (Valencia et al. 611 2019). 612 26 Page 27 of 147 Journal of Vegetation Science 613 References 614 615 Benjamini, Y., & Hochberg, Y. 1995. Controlling the False Discovery Rate: A 616 Practical and Powerful Approach to Multiple Testing. Journal of the Royal 617 Statistical Society: Series B (Methodological) 57: 289–300. 618 Blüthgen, N., Simons, N.K., Jung, K., Prati, D., Renner, S.C., Boch, S., Fischer, M., 619 Hölzel, N., Klaus, V.H., Kleinebecker, T., Tschapka, M., Weisser, W.W., & 620 Gossner, M.M. 2016. Land use imperils plant and animal community stability 621 through changes in asynchrony rather than diversity. Nature Communications 7: 622 10697. 623 Cardinale, B., Wright, J., Cadotte, M.W., Carroll, I.T., Hector, A., Srivastava, D.S., 624 Loreau, M., & Weis, J.J. 2007. Impacts of plant diversity on biomass production 625 increase through time because of species complementarity. Proceedings of the 626 National Academy of Sciences of the United States of America 104: 18123– 627 18128. 628 Craven, D., Eisenhauer, N., Pearse, W.D., Hautier, Y., Isbell, F., Roscher, C., Bahn, 629 M., Beierkuhnlein, C., Bönisch, G., Buchmann, N., Byun, C., Catford, J.A., 630 Cerabolini, B.E.L., Cornelissen, J.H.C., Craine, J.M., De Luca, E., Ebeling, A., 631 Griffin, J.N., Hector, A., Hines, J., Jentsch, A., Kattge, J., Kreyling, J., Lanta, V., 632 Lemoine, N., Meyer, S.T., Minden, V., Onipchenko, V., Polley, H.W., Reich, 633 P.B., van Ruijven, J., Schamp, B., Smith, M.D., Soudzilovskaia, N.A., Tilman, 634 D., Weigelt, A., Wilsey, B., & Manning, P. 2018. Multiple facets of biodiversity 635 drive the diversity–stability relationship. Nature Ecology and Evolution 2: 1579– 636 1587. 637 Gonzalez, A., & Loreau, M. 2009. The causes and consequences of compensatory 638 dynamics in ecological communities. Annual Review of Ecology, Evolution, and 639 Systematics 40: 393–414. 640 Gross, K., Cardinale, B.J., Fox, J.W., Gonzalez, A., Loreau, M., Wayne Polley, H., 641 Reich, P.B., & van Ruijven, J. 2014. Species richness and the temporal stability 642 of biomass production: a new analysis of recent biodiversity experiments. The 643 American Naturalist 183: 1–12. 644 Hautier, Y., Seabloom, E.W., Borer, E.T., Adler, P.B., Harpole, W.S., Hillebrand, H., 645 Lind, E.M., MacDougall, A.S., Stevens, C.J., Bakker, J.D., Buckley, Y.M., Chu, 646 C., Collins, S.L., Daleo, P., Damschen, E.I., Davies, K.F., Fay, P.A., Firn, J., 647 Gruner, D.S., Jin, V.L., Klein, J.A., Knops, J.M.H., La Pierre, K.J., Li, W., 648 McCulley, R.L., Melbourne, B.A., Moore, J.L., O’Halloran, L.R., Prober, S.M., 649 Risch, A.C., Sankaran, M., Schuetz, M., & Hector, A. 2014. Eutrophication 650 weakens stabilizing effects of diversity in natural grasslands. Nature 508: 521– 651 525. 652 Helm, A., Hanski, I., & Partel, M. 2006. Slow response of plant species richness to 653 habitat loss and fragmentation. Ecology Letters 9: 72-77. 654 Herben, T., Mayerová, H., Skálová, H., Hadincová, V., Pecháčková, S., & Krahulec, 655 F. 2017. Long-term time series of legume cycles in a semi-natural montane 656 grassland: evidence for nitrogen-driven grass dynamics? (R. Ostertag, Ed.). 657 Functional Ecology 31: 1430–1440. 658 Hill, M.O. 1973. The Intensity of spatial pattern in plant communities. The Journal of 659 Ecology 61: 225. 27 Journal of Vegetation Science Page 28 of 147 660 Houlahan, J.E., Currie, D.J., Cottenie, K., Cumming, G.S., Ernest, S.K.M., Findlay, 661 C.S., Fuhlendorf, S.D., Gaedke, U., Legendre, P., Magnuson, J.J., McArdle, 662 B.H., Muldavin, E.H., Noble, D., Russell, R., Stevens, R.D., Willis, T.J., 663 Woiwod, I.P., & Wondzell, S.M. 2007. Compensatory dynamics are rare in 664 natural ecological communities. Proceedings of the National Academy of 665 Sciences 104: 3273–3277. 666 Hubbell, S.P. 2001. The unified neutral theory of biodiversity and biogeography. 667 Princeton University Press. 668 Lepš, J. 1990. Comparison of transect methods for the analysis of spatial pattern. 669 Spatial processes in plant communities 71-82. 670 Lepš, J. 2014. Scale- and time-dependent effects of fertilization, mowing and 671 dominant removal on a grassland community during a 15-year experiment (S. 672 Wan, Ed.). Journal of Applied Ecology 51: 978–987. 673 Lepš, J., Götzenberger, L., Valencia, E., & de Bello, F. 2019. Accounting for 674 long‐term directional trends on year‐to‐year synchrony in species fluctuations. 675 Ecography 42: 1728-1741. 676 Lepš, J., Májeková, M., Vítová, A., Doležal, J., & de Bello, F. 2018. Stabilizing 677 effects in temporal fluctuations: management, traits, and species richness in high- 678 diversity communities. Ecology 99: 360–371. 679 Loreau, M., & de Mazancourt, C. 2013. Biodiversity and ecosystem stability: a 680 synthesis of underlying mechanisms (E. Duffy, Ed.). Ecology Letters 16: 106– 681 115. 682 Loreau, M., & de Mazancourt, C. 2008. Species synchrony and its drivers: neutral and 683 nonneutral community dynamics in fluctuating environments. The American 684 Naturalist 172: E48–E66. 685 Milchunas, D., Lauenroth, W., & Burke, I. 1998. Livestock grazing: animal and plant 686 biodiversity of shortgrass steppe and the relationship to ecosystem function. 687 Oikos 83: 65–74. 688 Pardo, I., Doak, D.F., García-González, R., Gómez, D., & García, M.B. 2015. Long- 689 term response of plant communities to herbivore exclusion at high elevation 690 grasslands. Biodiversity and Conservation 24: 3033–3047. 691 R Development Core Team. 2018. R: A language and environment for statistical 692 computing (A. R Foundation for Statistical Computing, Vienna, Ed.). R 693 Foundation for Statistical Computing, Vienna, Austria. 694 Rabotnov, T.A. 1974. Differences Between Fluctuations and Successions. In 695 Vegetation Dynamics, pp. 19–24. Springer Netherlands, Dordrecht. 696 Riginos, C., Porensky, L.M., Veblen, K.E., & Young, T.P. 2018. Herbivory and 697 drought generate short-term stochasticity and long-term stability in a savanna 698 understory community. Ecological Applications 28: 323–335. 699 Silvertown, J., Poulton, P., Johnston, E., Edwards, G., Heard, M., & Biss, P.M. 2006. 700 The Park Grass Experiment 1856–2006: its contribution to ecology. Journal of 701 Ecology 94: 801–814. 702 Stevens, C., Duprè, C., Gaudnik, C., Dorland, E., Dise, N., Gowing, D., Bleeker, A., 703 Alard, D., Bobbink, R., Fowler, D., Vandvik, V., Corcket, E., Mountford, J.O., 704 Aarrestad, P.A., Muller, S., & Diekmann, M. 2011. Changes in species 705 composition of European acid grasslands observed along a gradient of nitrogen 28 Page 29 of 147 Journal of Vegetation Science 706 deposition. Journal of Vegetation Science 22: 207–215. 707 Tilman, D., Isbell, F., & Cowles, J.M. 2014. Biodiversity and ecosystem functioning. 708 Annual Review of Ecology, Evolution, and Systematics 45: 471–493. 709 Tomimatsu, H., Sasaki, T., Kurokawa, H., Bridle, J.R., Fontaine, C., Kitano, J., 710 Stouffer, D.B., Vellend, M., Martijn Bezemer, T., Fukami, T., Hadly, E.A., Van 711 Der Heijden, M.G.A., Kawata, M., Efi, S.K., Kraft, N.J.B., Mccann, K.S., 712 Mumby, P.J., Nakashizuka, T., Petchey, O.L., Romanuk, T.N., Suding, K.N., 713 Takimoto, G., Urabe, J., & Yachi, S. 2013. Sustaining ecosystem functions in a 714 changing world: a call for an integrated approach. Journal of Applied Ecology 715 50: 1124–1130. 716 Tredennick, A.T., de Mazancourt, C., Loreau, M., & Adler, P.B. 2017. Environmental 717 responses, not species interactions, determine synchrony of dominant species in 718 semiarid grasslands. Ecology 98: 971–981. 719 Vasseur, D.A., & Gaedke, U. 2007. Spectral analysis unmasks synchronous and 720 compensatory dynamics in plankton communities. Ecology 88: 2058–2071. 721 Walter, J.A., Ives, A.R., Tooker, J.F., & Johnson, D.M. 2018. Life history and habitat 722 explain variation among insect pest populations subject to global change. 723 Ecosphere 9: e02274. 724 Wu, Z., Huang, N.E., Long, S.R., & Peng, C.-K. 2007. On the trend, detrending, and 725 variability of nonlinear and nonstationary time series. Proceedings of the 726 National Academy of Sciences of the United States of America 104: 14889– 727 14894. 728 729 29 Journal of Vegetation Science Page 30 of 147 730 Figure 1. Effects of the T3 detrending approach on synchrony, using the GrossW index 731 (Blüthgen et al. 2016) as an example. In panel (a), a ranking correlation between 732 synchrony values with and without detrending was computed for each of the 77 datasets 733 considered. The histogram reports the 77 Rho values of the Spearman ranking 734 correlations. Panel (b) reports, for each of the 77 datasets, the mean (+/- standard error) 735 of the synchrony values with and without the T3 detrending approach. Vertical and 736 horizontal dashed lines indicate zero synchrony (i.e. absence of synchrony). The solid 737 line represents the 1:1 line above which, for example T3 synchrony was greater than 738 synchrony without T3. 739 740 30 Page 31 of 147 Journal of Vegetation Science 741 Figure 2. Summary of the directional effects of the T3 detrending approach on various 742 synchrony indices and on CV. The bar plots indicate the numbers of datasets (n=77) in 743 which the T3 approach significantly increased (red bars) or decreased (blue bars) 744 synchrony values using a paired t-test after correction for false discovery rates. Grey 745 bars indicate the number of datasets with non-significant paired t-tests. 746 For 747 31 Journal of Vegetation Science Page 32 of 147 748 Figure 3. Effects of the T3 detrending approach in manipulated vs. control plots. The 749 plots report results of t-tests on 36 38 datasets in which there were both manipulated 750 and ‘control’ plots. For each dataset we used a pairwise t-test to compare synchrony 751 values (using the GrossW synchrony index, Blüthgen et al. 2016) with and without the 752 T3 approach (a: manipulated plots, and b: control plots). Positive values of the t-statistic 753 indicate that the T3 approach increased synchrony and negative ones indicate that the 754 T3 decreased synchrony. Values outside the grey area in each plot indicate significant 755 t-tests after correction for false discovery rates (‘ns’ indicates p > 0.05). For each panel 756 an R2 for the relationship between t-statistic and number of years sampled in each 757 dataset is provided together with the p-value of the regression model (the corresponding 758 regression line is shown when significant). 759 For Revie w 760 Only 761 32 Page 33 of 147 Journal of Vegetation Science 762 Supporting Information 763 Additional Supporting Information may be found in the online version of this article: 764 765 Appendix S1. Simulating long term trends in artificial communities to validate 766 effectiveness of the T3 approach 767 Appendix S2. Descriptions of each dataset, highlighting the treatments of the datasets 768 with ‘control’ and ‘manipulated’ plots. 769 Appendix S3. Application of the analyses shown in Fig. 1 of the main text to the 770 three remaining indices of synchrony. 771 Appendix S4. Application of the analyses shown in Fig. 3 of the main text to the 772 three remaining indices of synchrony. 773 Appendix S5. Results of the correlation between synchrony indices with species 774 richness or with the CV of total abundance. 33 Journal of Vegetation Science Page 1 of 147 1 Supporting Information to the paper 2 Valencia et al. Directional trends in species composition over time can lead to a 3 widespread overestimation of asynchrony. Journal of Vegetation Science. 4 5 Appendix S1. Simulating long term trends in artificial communities to validate 6 effectiveness of the T3 approach 7 We created artificial temporal community data with desired patterns of temporal 8 fluctuations (prevailing synchrony or asynchrony) using the “syngenr” R function 9 (Lepš et al. 2019). This function offers the possibility to build simulated communities, 10 fixing some parameters, such as the years of the time series (100 years) and the 11 number of species (8 species). Once the communities were established, communities 12 fluctuating in time were created according to the following scenarios: prevailing 13 synchrony or prevailing asynchrony. A synchronous pattern was simulated by having 14 a common response for all species to a hypothetical environmental cue. Accordingly, 15 an asynchronous pattern was created by having half of the species responding 16 positively and the other half negatively to the environmental cue. Furthermore, we 17 simulated directional (monotonic) and cyclical long term trends for these artificial 18 communities. First, we simulated a case where most species had a common long-term 19 positive trend (monotonic) leading to a steady increase of individual species over 20 time. This would lead to detect synchrony with the synchrony indices (unmodified), 21 even if the species are actually behaving asynchronous. Second, we simulated the 22 opposite case, where species either increase or decrease in time, with the 23 increase/decrease for each species defined by a combined bimodal distribution from 24 two normal distributions with -1 and 1 as means, and random subset from half of the 25 species more probably have a positive long-term trend and the other half of the 26 species more probably a negative long-term trend. Finally, we simulated a case were 27 the directional long-term had cyclical tendencies. The cyclical long term trends were 1 Page 2 of 147 Journal of Vegetation Science 28 evaluated with different cycle length: 3, 6, 18 and 40 years. These cycle lengths 29 reflect some known potential long term cycles that drive communities across the 30 world, such as the El Niño Oscillation or Pacific Decadal Oscillation, which have 31 intervals of 3-6 years and 10-20 year, respectively. In summary, we simulated two 32 scenarios of year-to-year species fluctuations (prevailing synchrony or prevailing 33 asynchrony) and three types of long-term directional trends (i.e. monotonic with a 34 common or contrasted trend, and cyclical trends), resulting in six possible 35 combinations of trend–fluctuation scenarios. In all these simulated communities, we 36 calculated the different synchrony indices (Gross, GrossW, Logvar and Phi), with or 37 without the use of the T3 detrending approach, using the “calc_sync” R function 38 (Lepš et al. 2019). 39 We assessed the effectiveness of the T3 detrending approach when long-term 40 monotonic or cyclical trends are present in the data across the most common 41 synchrony indices, using a paired t-test. Fluctuations simulated under scenarios of 42 long-term trends in species abundances showed biased index estimates, i.e. the 43 simulated synchrony or asynchrony patterns were overshadowed by the patterns 44 caused by long-term trends. In the case of species having long-term directional or 45 cyclical trends, asynchrony was masked by the synchrony (Figure Appendix S1a and 46 S1c). Then, the synchrony indices without the T3 detrending approach were not able 47 to detect asynchrony, even if the species were actually behaving asynchronously. 48 These synchrony indices values were significantly higher than those with the T3 49 detrending approach. These biases were found across all indices but the application of 50 the T3 detrending approach was correctly able to remove them, in all indices (Figure 51 Appendix S1). In the opposite case, simulation of synchrony together with long-term 52 monotonic or cyclical trends, the difference still prevailed among the synchrony with 2 Journal of Vegetation Science Page 3 of 147 53 and without the T3 detrending approach, but with a less pronounced effect (Figure 54 Appendix S1b and S1d). 55 56 Figure Appendix S1. Results of synchrony indices (Gross, GrossW Logvar, and 57 Phi)(‘Logvar’, Lepš et al. 2018)(‘Logvar’, Lepš et al. 2018), with or without the use 58 of the T3 detrending approach, in artificial temporal communities where long term 59 trends were simulated. The panels report results for a common long-term directional 60 trend (a) (i.e. creating synchrony; all species increasing in time), a contrasted long- 61 term trend (b) (i.e. half species increasing, the other half decreasing, creating 62 asynchrony) and a cyclical trend (c and d). Within each of these scenarios we 63 considered two scenarios: year to year asynchrony (a and c) and synchrony (b and d). 64 The cyclic trends also included different cycle length (3, 6, 18 and 40 years). The 65 created communities had a total of 8 species. Asterisks above and between boxes 66 depict significant differences among the synchrony indices with or without the T3 67 approach as assessed with a paired t-test. *: P < 0.05; **: P < 0.01; ***: P < 0.001; 68 ****: P < 0.0001. 69 70 Gross 71 72 3 Page 4 of 147 Journal of Vegetation Science 73 GrossW 74 75 76 77 Logvar 78 79 80 81 4 Journal of Vegetation Science Page 5 of 147 82 Phi 83 84 85 5 Page 6 of 147 Journal of Vegetation Science 86 Supporting Information to the paper 87 Valencia et al. Directional trends in species composition over time can lead to a 88 widespread overestimation of asynchrony. Journal of Vegetation Science. 89 90 Appendix S2. Descriptions of each dataset, highlighting the treatments of the datasets 91 with ‘control’ and ‘manipulated’ plots. LAT: latitude (WGS84 datum), and LON: 92 longitude (WGS84 datum). 93 94 1. The dataset is issued from an experiment in a northern mixed prairie at a field 95 station in Miles City, Montana, USA (LAT: 46.32, and LON: -105.80). This dataset 96 consists of 42 plots, where each plot was sampled an average of 12.5 times. In each 97 plot, individual plants were quantified and mapped annually. More information: 98 http://esapubs.org/archive/ecol/E092/143/#data 99 100 2. The dataset is issued from an experiment located on a mixed grass prairie in Hays, 101 Kansas, USA (LAT: 38.80, and LON: -99.30). This dataset consists of 51 plots, where 102 each plot was sampled an average of 34.5 times. In each plot, individual plants were 103 quantified and mapped. Thirty-six permanent quadrats were located inside livestock 104 exclosures and 15 in grazed areas. More information: 105 https://web.archive.org/web/20150128015820/http://esapubs.org:80/archive/ecol/E08 106 8/161/default.htm 107 108 3. The dataset is issued from an experiment located on a shortgrass steppe of North 109 America in Nunn, Colorado, USA (LAT: 40.85, and LON: -104.71). This dataset 110 consists of 24 plots, where each plot was sampled an average of 13.5 times. In each 111 plot, individual plants were quantified and mapped. The quadrats were established in 112 six grazed and ungrazed study sites on the Central Plains Experimental Range. There 113 were four treatments combining past and present grazing status: ungrazed in the past 114 and at present (ungrazed/ungrazed), grazed by livestock in the past and present 115 (grazed/grazed), grazed in the past and ungrazed during the experiment 116 (grazed/ungrazed), and ungrazed in the past and grazed during the experiment 117 (ungrazed/grazed). More information: 118 https://web.archive.org/web/20150502183659/http:/www.esapubs.org/archive/ecol/E0 119 94/128/ 120 6 Journal of Vegetation Science Page 7 of 147 121 4. The dataset is issued from an experiment located on semi-desert grasslands at the 122 Santa Rita Experimental Range, Arizona, USA (LAT: 31.83, and LON: -110.88). This 123 dataset consists of 160 plots, where each plot was sampled an average of 11.2 times. 124 In each plot, individual plants were quantified and mapped. Quadrats were located in 125 exclosures (ungrazed) and in pastures grazed by livestock (grazed). More information: 126 https://web.archive.org/web/20150502183207/http://esapubs.org:80/archive/ecol/E09 127 3/132/default.htm 128 129 5. The dataset is issued from an experiment located in sagebrush steppe in eastern 130 Idaho, USA (LAT: 44.20, and LON: -112.20). This dataset consists of 23 plots, where 131 132 orF each plot was sampled an average of 21.5 times. In each plot, individual plants were quantified and mapped. These permanent quadrats were located in both grazed (4 133 quadrats) and ungrazed units (18 quadrats), and one quadrat was grazed in the past 134 and ungrazed during the experiment. More information: 135 https://web.archive.org/web/20150128015825/http:/esapubs.org/archive/ecol/E091/24 136 3/default.htm. 137 138 6. The dataset is issued from an experiment on the Jornada Long-Term Ecological 139 Research site in southern New Mexico, USA (LAT: 32.83, and LON: -107.33). This 140 dataset consists of 222 plots, where each plot was sampled an average of 8.0 times. 141 Previously grazing domestic livestock was excluded from the area where three 142 permanent transects (2.7 km) were established. One of the transects received 143 fertilization of 10 g/m2 of nitrogen. One of the two control transects (not fertilized), 144 was sampled at 40 stations, the other two transects had 91 stations each. At each 145 station abundance of each species was estimated by point-intercept method along a 30 146 m transect perpendicular to each of the three permanent transects. More information: 147 https://portal.lternet.edu/nis/mapbrowse?packageid=knb-lter-jrn.2100119001.50. 148 149 7. The dataset is issued from an experiment on the Jornada Basin Long-Term 150 Ecological Research Program (LTER) site in the Chihuahuan desert, New Mexico, 151 USA (LAT: 32.93, and LON: -107.36). This dataset consists of 1001 plots, where 152 each plot was sampled an average of 11.5 times. On the grassland site, three exclusion 153 treatment levels were set in addition to the control treatment left open to all grazers. 154 The first level excluded only domestic animals (cattle), the second excluded 7 Page 8 of 147 Journal of Vegetation Science 155 lagomorphs, and the third excluded rodents. In the shrubland site, only lagomorph- 156 and rodent-exclusion treatments were set in addition to the control. In each treatment 157 of each site, 4 grids of 36 permanent plots (1 m2) were sampled (visual estimated 158 cover). More information: 159 https://portal.lternet.edu/nis/metadataviewer?packageid=knb-lter-jrn.2100086002.39. 160 161 8. The dataset is issued from an experiment in an open grassland of the South African 162 Kalahari near Askham, South Africa (LAT: -26.76, and LON: 20.61). This dataset 163 consists of 20 plots, where each plot was sampled an average of 10.7 times. The cover 164 values (%) of all individual plant species were estimated annually. More information: 165 Jürgens et al. (2010). 166 167 9. The dataset is issued from an experiment located in the Succulent Karoo in 168 Soebatsfontein, South Africa (LAT: -30.19, and LON: 17.54). This dataset consists of 169 24 plots, where each plot was sampled an average of 15.8 times. The cover values (%) 170 of all individual plant species were estimated annually. More information: Jürgens et 171 al. (2010). 172 173 10. The dataset is issued from an experiment located in the Succulent Karoo, near 174 Leliefontein, South Africa (LAT: 18.28, and LON: -30.40). This dataset consists of 42 175 plots, where each plot was sampled an average of 14.7 times. The cover values (%) of 176 all individual plant species were estimated annually. More information: Jürgens et al. 177 (2010). 178 179 11. The dataset is issued from an experiment located in the Succulent Karoo, 180 Knersvlakte near Vanrhynsdorp, South Africa (LAT: -31.28, and LON: 18.59). This 181 dataset consists of 40 plots, where each plot was sampled an average of 16.0 times. 182 The cover values (%) of all individual plant species were estimated annually. More 183 information: Jürgens et al. (2010). 184 185 12. The dataset is issued from an experiment on the Kiskun LTER located in Bugac 186 and Orgovány sites of Kiskunság National Park, Hungary (LAT: 46.73, and LON: 187 19.54). This dataset consists of 380 plots, where each plot was sampled an average of 188 14.5 times. Half of the plots were fenced to control grazing pressure. In each plot, the 8 Journal of Vegetation Science Page 9 of 147 189 cover values (%) were visually estimated annually. More information: Kertész et al. 190 (2017). 191 192 13. The dataset is issued from an experiment on a grassland in Cedar Creek LTER 193 Ecosystem Science Reserve, Minnesota, USA (LAT: 45.41, and LON: -93.16). This 194 dataset consists of 50 plots, where each plot was sampled an average of 7.0 times. The 195 plots were divided in 10 treatments of fertilization and grazing exclusion (Control=no 196 treatment, K=potassium, P=phosphate, N=nitrogen, PK=phosphate and potassium, 197 NK=nitrogen and potassium, NP=nitrogen and phosphate, NPK=nitrogen, phosphate 198 199 200 oF and potassium, Fence=Fence, NPK+Fence=nitrogen, phosphate and potassium + fence). In each plot, the cover values (%) were visually estimated annually. This dataset was provided from Cedar Creek LTER. More information: 201 http://cedarcreek.umn.edu/research/data/dataset?acze247. 202 203 14. The dataset is issued from an experiment located in the Cedar Creek LTER 204 Ecosystem Science Reserve, Minnesota, USA (LAT: 45.41, and LON: -93.19). This 205 dataset consists of 184 plots, where each plot was sampled an average of 6.2 times. 206 Plots were distributed across 6 treatments with increasing burning frequency: i) no 207 burning – control – (48 plots), ii) 1 per 10 years (16 plots), iii) 1 per 3 years (32 plots), 208 iv) 1 per 2 years (32 plots), v) 2 per 3 years (8 plots) and vi) 4 per 5 years (48 plots). 209 Plots are located on 12 management areas ranging in size from 2.4 to 30 ha. In each 210 plot, the cover values (%) were visually estimated. More information: 211 http://cedarcreek.umn.edu/research/data/dataset?herbe133. 212 213 15. The dataset is issued from an experiment located in the Cedar Creek LTER 214 Ecosystem Science Reserve, Minnesota, USA (LAT: 45.41, and LON: -93.19). This 215 dataset consists of 60 plots, where each plot was sampled an average of 24.8 times. In 216 each plot, the biomass of individual plants was recorded from 4 plots (0.3 m2) per 217 field until 2013. More information: 218 http://cedarcreek.umn.edu/research/data/dataset?ple054. 219 220 16. The dataset is issued from an experiment located in the Cedar Creek LTER 221 Ecosystem Science Reserve, Minnesota, USA (LAT: 45.40, and LON: -93.20). This 222 dataset consists of 234 plots, where each plot was sampled an average of 22.0 times. 9 Page 10 of 147 Journal of Vegetation Science 223 The experiment combines different levels of fertilization on 4 fields that were 224 abandoned for different periods (14, 25, 48 years and never ploughed before the 225 experiment started in 1982) and where mammal grazers were excluded. In each plot, 226 individual plant biomass was recorded on 5 to 6 replicate plots of different 227 fertilization treatments (from 0 to 40 g/m2 of nitrogen) per field every year. More 228 information: http://cedarcreek.umn.edu/research/data/dataset?ple001. 229 230 17. The dataset is issued from an experiment located in the Cedar Creek LTER 231 Ecosystem Science Reserve, Minnesota, USA (LAT: 45.40, and LON: -93.20). This 232 dataset consists of 237 plots, where each plot was sampled an average of 14.8 times. 233 The experiment combines 9 levels of fertilization (from 0 to 40 g/m2 of nitrogen) and 234 prescribed burning on three fields that were abandoned since 14, 25 and 48 years, and 235 where mammal grazers were excluded. All 3 fields had 6 replicate plots of the 9 236 fertility treatments from 1982. From 1992 half of the plots in field B were burned 237 every spring, and half of the plots in field A and C stopped receiving the fertilization 238 treatment. To maintain continuity of the treatments within plots the 1992-2011 period 239 of those plots were entered in the database as separate plots of the same data set. 240 Individual plant biomass was measured. More information: 241 http://cedarcreek.umn.edu/research/data/dataset?ple002. 242 243 18. The dataset is issued from an experiment located in the Shortgrass Steppe LTER 244 in the Central Plains Experimental Range, Colorado, USA (LAT: 40.85, and LON: - 245 104.77). This dataset consists of 795 plots, where each plot was sampled an average 246 of 13.5 times. Plots were distributed across four combinations of past/current 247 management: grazed/grazed, ungrazed/ungrazed, grazed/ungrazed and 248 ungrazed/grazed. In 1998, additional plots were added in a fifth treatment with fences 249 excluding both large and small grazers (rodent exclusion). More information: 250 https://portal.lternet.edu/nis/metadataviewer?packageid=knb-lter-sgs.527.1. 251 252 19. The dataset is issued from an experiment located in sandy semi-natural grasslands 253 of the Elbe valley in Höhbeck, Germany (LAT: 53.05, and LON: 11.41). This dataset 254 consists of 96 plots, where each plot was sampled an average of 6.0 times. The 255 vegetation was surveyed once a year in 1 m2 plots using the Londo scale (Londo 256 1976). More information: Schuhmacher & Dengler (2013). 10 Journal of Vegetation Science Page 11 of 147 257 258 20. The dataset is issued from an experiment located near Dufftown, Morayshire, 259 United Kingdom (LAT: 57.73, and LON: -3.10). This dataset consists of 12 plots, 260 where each plot was sampled an average of 6.0 times. Each species was measured in a 261 transect, using the inclined-point quadrat method (Tinney et al. 1937) (32·5° to the 262 horizontal). All contacts with 5 pins were recorded in 20 quadrat positions per plot. 263 More information: Pakeman et al. (2003). 264 265 21. The dataset is issued from an experiment located in Andrew Experimental forest 266 267 268 For Program (AND-LTER), Oregon, USA (LAT: 44.35, and LON: -122.41). This dataset consists of 193 plots, where each plot was sampled an average of 21.4 times. Plots were established in i) undisturbed, ii) logged, iii) logged and lightly burned, and iv) 269 logged and severely burned areas. In each plot, the cover values (%) were estimated. 270 More information: https://portal.lternet.edu/nis/metadataviewer?packageid=knb-lter- 271 and.3217.11. 272 273 22. The dataset is issued from an experiment located on woodlands, grasslands, and 274 shrublands in eastern Australia (LAT: -30.12, and LON: 147.17). This dataset consists 275 of 47 plots, where each plot was sampled an average of 10.2 times. In each plot, the 276 biomass of the vegetation was measured annually, from 1991 to 2002, in four 300 m 277 long transects each containing 13 quadrats of 0.72 m x 0.72 m. Dataset owners: James 278 Val and David Eldridge (Office of Environment & Heritage, University of New South 279 Wales). 280 281 23. The dataset is issued from an experiment located on a pasture in Fasque, United 282 Kingdom (LAT: 56.87, and LON: -2.60). This dataset consists of eight plots, where 283 each plot was sampled an average of 8.0 times. Inclined-point quadrat method (32·5° 284 to the horizontal) was used to record each species in a transect, with a minimum of 20 285 point contacts at 18 locations per plot (i.e. a minimum of 360 contacts per plot). More 286 information: Marriott et al. (2002). 287 288 24. The dataset is issued from an experiment located on La Fage French National 289 Institute for Agricultural Research (INRA) experimental station, close to Millau, 290 France (LAT: 43.92, and LON: 3.10). This dataset consists of 16 plots, where each 11 Page 12 of 147 Journal of Vegetation Science 291 plot was sampled an average of 28.0 times. Individual plants were identified using the 292 point intercept method on 5 m permanent lines (1 point/10 cm, i.e. 50 points/line). 293 More information: Chollet et al. (2014) and Garnier et al. (2018). 294 295 25. The data sourced from BioTIME (Dornelas et al. 2018), Study_ID 483 and 497- 296 ITEX Dataset 5 - Teberda (Malaya Alpine-Snowbed and Geranium Hedysarum 297 Meadow) and ITEX Dataset 19 - Teberda (Festuca Varia Grassland, Malaya Alpine 298 Lichen-Heath). The dataset is issued from an experiment located in Teberda State 299 Reserve, a part of the Karachaevo-Cherkessian Republic in the northwestern 300 Caucasus, Russia (LAT: 43.45, and LON: 41.69). This dataset consists of 145 plots, 301 where each plot was sampled an average of 24.3 times. In each plot, the cover of each 302 plant species was recorded as number of shoots per m2. More information: 303 Onipchenko et al. (1998). 304 305 26. The dataset is issued from an experiment located a moorland in the Clocaenog 306 Forest, United Kingdom (LAT: 53.06, and LON: -3.47). This dataset consists of 9 307 plots, where each plot was sampled an average of 12.0 times. The experiment was 308 designed with three treatments: control, drought (~20% reduction in total annual 309 rainfall) and warming (~64% reduction in heat loss during night and 14% reduction in 310 total annual rainfall). Three quadrats per plot were chosen, and in each quadrat 311 vegetation was quantified using a grid of 100 pins (pin-point methodology). Pin hits 312 were then converted to biomass (g m-2) using a biomass calibration-conversion. More 313 information: https://catalogue.ceh.ac.uk/documents/5b39a644-d614-4f2b-8df6- 314 202ed440b4ab. Doi: https://doi.org/10.5285/5b39a644-d614-4f2b-8df6- 315 202ed440b4ab. 316 317 27. The dataset is issued from an experiment located on serpentine and non-serpentine 318 meadows in California, USA (LAT: 38.85, and LON: -123.50). This dataset consists 319 of 400 plots, where each plot was sampled an average of 10.0 times. In each plot, the 320 species cover (%) was visually estimated annually. More information: Fernandez- 321 Going et al. (2012) and Harrison (1999). 322 323 28. The dataset is issued from an experiment located on the Jornada Basin 324 Experimental Range JRN-LTER in the Chihuahuan desert, New Mexico, USA (LAT: 12 Journal of Vegetation Science Page 13 of 147 325 32.62, and LON: -106.67). This dataset consists of 68 plots, where each plot was 326 sampled an average of 27.8 times. Density of individuals per species and per plot was 327 recorded. More information: 328 https://portal.lternet.edu/nis/metadataviewer?packageid=knb-lter-jrn.210351002.75. 329 330 29. The dataset is issued from an experiment located on a grassland in Krkonose 331 Mountains, Czech Republic (LAT: 50.69, and LON: 15.71). This dataset consists of 332 four plots, where each plot was sampled an average of 16.8 times. Standing biomass 333 was sampled annually. More information: Herben et al. (1997). 334 335 30. The dataset is issued from an experiment located on a grassland in Krkonose 336 Mountains, Czech Republic (LAT: 50.69, and LON: 15.79). This dataset consists of 337 four plots, where each plot was sampled an average of 29.8 times. Standing biomass 338 was sampled annually. More information: Herben et al. (2017). 339 340 31. The data sourced from BioTIME (Dornelas et al. 2018), Study_ID 243 - Virginia 341 Coast Reserve Long-Term Ecological Research. The dataset is issued from an 342 experiment located in the coastal sand dunes of Hog island, Virginia, USA (LAT: 343 37.67, and LON: -75.67). This dataset consists of 28 plots, where each plot was 344 sampled an average of 18.9 times. Half of the plots received nitrogen fertilization each 345 year in the form of urea nitrogen (30% uncoated (46-0-0) and 70% (40-0-0) coated for 346 slow release). The fertilizer was applied evenly in a dry form (15 g/m2 of nitrogen). In 347 each plot, species cover (%) was visually estimated in five 0.25 m2 plots. More 348 information: Day et al. (2016). 349 350 32. The dataset is issued from an experiment located on a grassland near Napal, Spain 351 (LAT: 42.72, and LON: -1.22). This dataset consists of 12 plots, where each plot was 352 sampled an average of 12.0 times. The experimental area was fenced and shrubs were 353 removed. Six plots were fertilized (sewage sludge to the soil surface with 5000 g/m2) 354 and six plots were used as controls. All vascular plant species were measured annually 355 using frequencies. To do so, each plot was divided into 100 subplots, and the 356 presence/absence of each species was recorded. More information: Gazol et al. 357 (2016). 358 13 Page 14 of 147 Journal of Vegetation Science 359 33. The data were sourced from BioTIME (Dornelas et al. 2018), Study_ID 491 - 360 ITEX Dataset 13 - Toolik (Dry, Moist). The dataset is issued from an experiment 361 located on tundra vegetation near Toolik, Alaska, USA (LAT: 68.62, and LON: - 362 149.61). This dataset consists of eight plots, where each plot was sampled an average 363 of 6.0 times. The plots are divided between dry tundra with control and warming 364 treatments and moist tundra with only control treatment. Biomass estimates were 365 obtained using a fixed 75 cm2 point frame, with 100 measurements spaced 7 cm apart. 366 367 34. The data was sourced from BioTIME (Dornelas et al. 2018), Study_ID 492 - 368 ITEX Dataset 14 - Toolik (LTER Heath, LTER Moist acidic tussock, LTER non- 369 acidic tussock, LTER wet sedge, SAG wet sedge2, Tussock 1981 plots). The dataset 370 is issued from an experiment located in Toolik, Alaska, USA (LAT: 68.63, and LON: 371 -149.58). This dataset consists of four plots, where each plot was sampled an average 372 of 6.0 times. In each plot, species biomass was assessed by clipping of four or five 373 0.25 m x 0.25 m plots, and sorting to species level. 374 375 35. The dataset is issued from an experiment located on a grassland in Bayreuth, 376 Germany (LAT: 49.92, and LON: 11.59). This dataset consists of 15 plots, where 377 each plot was sampled an average of 7.7 times. Three treatments were applied: 1) 378 control (ambient condition), 2) winter warming (October–March), and 3) summer 379 warming (April–September). In each plot, species cover (%) was visually estimated 380 annually. More information: Grant et al. (2017). 381 382 36. The dataset is issued from an experiment located on a grassland in the Czech 383 Republic (LAT: 48.87, and LON: 16.64). This dataset consists of seven plots, where 384 each plot was sampled an average of 8.0 times. In each plot (1 m2), the species cover 385 (%) was visually estimated annually from 1993 to 2001. Dataset owner: Jiří Danihelka 386 (Department of Botany and Zoology, Masaryk University and Department of 387 Vegetation Ecology, Institute of Botany, The Czech Academy of Sciences). 388 389 37. The dataset is issued from an experiment located on a grassland in Laqueuille, 390 France (LAT: 45.64, and LON: 2.73). This dataset consists of 10 plots, where each 391 plot was sampled an average of 13.0 times. Half of the plots were located in an 392 intensively managed grassland (10-15 animals ha-1 yr-1 and 20 g/m2 of nitrogen), and 14 Journal of Vegetation Science Page 15 of 147 393 the other half were located in a neighbouring grassland under extensive management 394 (5-8 animals ha-1 yr-1 and no fertilization). In each plot, presence/absence of each 395 species was recorded in 40 pin-points regularly spaced (pin-point methodology. 396 Dataset owner: Katja Klumpp (INRA, Grassland Ecosystem Research Unit). 397 398 38. The dataset is issued from an experiment located on Shortgrass Steppe (SGS- 399 LTER) in the Central Plains Experimental Range Nunn, Colorado, USA (LAT: 40.85, 400 and LON: -104.71). This dataset consists of 48 plots, where each plot was sampled an 401 average of 9.0 times. The experiment evaluated four treatments: control inside 402 exclosure, control outside exclosures, Bouteloua gracilis removal inside exclosure and 403 404 orF Bouteloua gracilis removal outside exclosure. Species density was measured in a quadrat (1 m2) using vegetation point intercept method (40 points of contact was 405 recorded for each quadrat). More information: 406 https://portal.lternet.edu/nis/mapbrowse?packageid=knb-lter-sgs.703.1. 407 408 39. The dataset is issued from an experiment located on a wet meadow in Ohrazeni, 409 Czech Republic (LAT: 48.95, and LON: 14.59). This dataset consists of 12 plots, 410 where each plot was sampled an average of 16.0 times. The experiment evaluated four 411 treatments: control, mowing (annually in the second half of June), fertilization (65 412 g/m2 of commercial NPK fertilizer) and dominant removal (Molinia caerulea plants 413 were manually removed annually). In each plot, the biomass of each species was 414 measured annually. More information: Lepš (2014). 415 416 40. The dataset is issued from an experiment (Long Term Experiment SOERE- 417 ACBB) located on a grassland in Theix, France (LAT: 45.72, and LON: 3.02). This 418 dataset consists of eight plots, where each plot was sampled an average of 8.0 times. 419 The experiment evaluated, on one hand, the effect of the intensity of grazing with two 420 treatments with cattle rotational grazing at high (Ca+) or low (Ca-) level of herbage 421 utilisation; these two treatments did not receive any mineral fertilisation. On the other 422 hand, it also evaluated the effect of nutrient availability, comparing two treatments 423 conducted under fixed cutting regime (three cuts/per year), one with fertilization 424 (NPK fertilizer) and the other without fertilization. The presence/absence of each 425 plant species was measured using 40 pin-points regularly spaced along fixed transects. 426 Complementarily, at each pin-point, 6 points are distributed to species according to 15 Page 16 of 147 Journal of Vegetation Science 427 visual estimation of their volume. Dataset owner: Frédérique Louault (INRA-UREP). 428 More information: Louault et al. (2017). 429 430 41. The dataset is issued from an experiment belonging to the Sevilleta LTER and 431 located on Chihuahuan desert in Sevilleta National Wildlife Refuge, New Mexico, 432 USA (LAT: 34.27, and LON: -106.68). This dataset consists of six plots, where each 433 plot was sampled an average of 14.3 times. More information: 434 https://portal.lternet.edu/nis/mapbrowse?packageid=knb-lter-sev.200.174699. 435 436 42. The dataset is issued from an experiment located on hyper-oceanic coastal 437 grasslands in United Kingdom (LAT: 57.27, and LON: -7.40). This dataset consists of 438 48 plots, where each plot was sampled an average of 6.8 times. The experiment 439 evaluated six treatments: 1) vertebrate grazing exclusion, 2) burial box with no sand 440 added, 3) buried to 10 cm, 4) buried to 20 cm, 5) windbreak - shelter from prevailing 441 SW winds, 6) no treatment. The cover values (%) of all individual plant species were 442 estimated annually from 2004 to 2010. Data owners: Robin Pakeman (James Hutton 443 Institute. Aberdeen) and Jack J. Lennon (School of Biological Sciences, Queen's 444 University Belfast). 445 446 43. The dataset is issued from an experiment located on a grassland in Cleish and 447 Kirkton, United Kingdom (LAT: 56.29, and LON: -4.07). This dataset consists of 16 448 plots, where each plot was sampled an average of 6.0 times. The experiment evaluated 449 ungrazed and sheep-grazed plots to maintain three different levels of sward height. In 450 each plot, the inclined-point quadrat method (32·5° to the horizontal) at 20 locations 451 (with a minimum of 25 contacts per location) was used to measure each species. More 452 information: Hulme et al. (1999). 453 454 44. The dataset is issued from an experiment located on a grassland in Bell Hill and 455 Cleish, United Kingdom (LAT: 55.80, and LON: -2.84). This dataset consists of eight 456 plots, where each plot was sampled an average of 7.0 times. In each plot, the inclined- 457 point quadrat method (32·5° to the horizontal) at 20 locations (with a minimum of 25 458 contacts per location) was used to measure each species. More information: Grant et 459 al. (1996a). 460 16 Journal of Vegetation Science Page 17 of 147 461 45. The dataset is issued from an experiment located on a grassland in Cleish and 462 Sourhope, United Kingdom (LAT: 55.81, and LON: -2.86). This dataset consists of 463 seven plots, where each plot was sampled an average of 6.0 times. There were 464 different treatments where cattle or sheep density was adjusted twice a week to 465 maintain the vegetation height between tussocks. In each plot, the inclined-point 466 quadrat method (32·5° to the horizontal) at 20 locations (with a minimum of 25 467 contacts per location) was used to measure each species. More information: Grant et 468 al. (1996) and Common et al. (1998). 469 470 46. The dataset is issued from an experiment located on a moorland previously on the 471 Burnhead heft at the Redesdale Experimental Farm in Northumberland, United 472 Kingdom (LAT: 55.37, and LON: -2.45). This dataset consists of 12 plots, where each 473 plot was sampled an average of 6.0 times. The 12 plots were divided in three areas 474 with different grazing treatments: ungrazed, sheep-grazed (three levels: 0.4, 0.8 and 475 1.2 ha-1 yr-1). In each plot, the inclined-point quadrat method (32·5° to the horizontal) 476 at 20 locations (with a minimum of 25 contacts per location) was used to measure 477 each species. More information: Pakeman & Nolan (2009). 478 479 47. The dataset is issued from an experiment located on a heather moorland at 480 Dundonnell near Ullapool and at Claonaig, near Tarbert Loch Fyne, Argyll and Bute, 481 United Kingdom (LAT: 57.35, and LON: -5.55). This dataset consists of 17 plots, 482 where each plot was sampled an average of 6.0 times. The experiment had different 483 sheep grazing and exclusion treatments: 1) low at 0.4 sheep ha-1 yr-1, 2) moderate at 484 0.8 sheep ha-1 yr-1, 3) high at 1.2 sheep ha-1 yr-1, 4) fenced against both cattle and 485 sheep, and 5) fenced against cattle, also 6) sheep and cattle recorded from the open 486 hill. In each plot, the inclined-point quadrat method (32·5° to the horizontal) at 20 487 locations was used to measure each species. More information: Pakeman & Nolan 488 (2009). 489 490 48. The dataset is issued from an experiment located on a grassland in the Ordesa- 491 Monte Perdido National Park, Spain (LAT: 42.67, and LON: -0.06). This dataset 492 consists of four plots, where each plot was sampled an average of 19.0 times. The 493 point intercept method at 20 locations was used to measure each species. 17 Page 18 of 147 Journal of Vegetation Science 494 In each plot, the point intercept method was used annually to measure vegetation 495 along two perpendicular transects (a total of 400 sample points). More information: 496 Pardo et al. (2015). 497 498 49. The dataset is issued from an experiment located in Soto de Viñuelas, Spain 499 (LAT: 40.60, and LON: -3.63). This dataset consists of 68 plots, where each plot was 500 sampled an average of 11.5 times. In each plot, all plant species was recorded using 501 presence/absence data in five quadrats of 400 cm2 each from 1980 to 1995. Dataset 502 owner: Begoña Peco (Ecology Department Autonomous, University of Madrid). 503 504 50. The dataset is issued from an experiment located on a shrubland in Garraf, Spain 505 (LAT: 41.30, and LON: 1.82). This dataset consists of nine plots, where each plot was 506 sampled an average of 17.0 times. Three experiment evaluated three treatments: 1) 507 control, 2) warming (metallic curtain covering the plots during the night), and 3) 508 drought (transparent curtain covering the plots during rainfall). Number of contacts 509 per plot was used to quantify each species. Dataset owners: Josep Penuelas, Marc 510 Estiarte and Romà Ogaya (Global Ecology Unit CREAF-CSIC-UAB). 511 512 51. The dataset is issued from an experiment belong to the Jornada LTER (JRN- 513 LTER) and located in Chihuahuan desert, Jornada Basin Experimental Range, New 514 Mexico, USA (LAT: 32.00, and LON: -106.00). This dataset consists of 734 plots, 515 where each plot was sampled an average of 24.0 times. In each plot, the biomass of 516 each species was calculated from field measurement of individual species cover and 517 height. More information: 518 https://portal.lternet.edu/nis/metadataviewer?packageid=knb-lter-jrn.2100011001.49. 519 520 52. The dataset is issued from an experiment located on a moorland on the Burnhead 521 heft at the Redesdale Experimental Farm in Northumberland, United Kingdom (LAT: 522 55.37, and LON: -2.45). This dataset consists of 10 plots, where each plot was 523 sampled an average of 6.0 times. The experiment had different grazing treatments: 524 summer grazing, winter grazing or year-round grazing (0.7 sheep ha-1 yr-1), year- 525 round grazing (1.4 sheep ha-1 yr-1), and no grazing. In each plot, the inclined-point 526 quadrat method (32·5° to the horizontal) at 20 locations (with a minimum of 25 18 Journal of Vegetation Science Page 19 of 147 527 contacts per location) was used to measure each species. More information: Hulme et 528 al. (2002) and Pakeman & Nolan (2009). 529 530 53. The dataset is issued from an experiment located on moorlands in Derbyshire, 531 United Kingdom (LAT: 54.69, and LON: -2.41). This dataset consists of 216 plots, 532 where each plot was sampled an average of 10.0 times. The experiment evaluated 36 533 treatments: no treatment; cut once per year; cut twice per year; herbicide sprayed; 534 herbicide sprayed in first year, cut in second; and cut in first year, sprayed in second. 535 Within each of these main plot treatments there were two sub-plot grazing treatments 536 - sheep grazing and no sheep grazing. Finally, there were three restoration treatments 537 applied at the sub-sub-plot level: untreated, Calluna moorland litter applied as litter, 538 and Calluna vegetation applied as cut brash. All these 36 treatments had 6 replicates. 539 In each plot, the species composition was recorded using point-quadrats (1 m-long 540 frame with 10 pin positions at 10 cm intervals, pin diameter = 2 mm). Dataset owner: 541 Rob Marrs (University of Liverpool). 542 543 54. The dataset is issued from an experiment belonging to the Environmental Change 544 Network (ECN) and located in the United Kingdom (LAT: 53.95, and LON: -3.23). 545 This dataset consists of 198 plots, where each plot was sampled an average of 11.1 546 times. In each plot (ten quadrats of 0.16 m2), the inclined-point quadrat method was 547 used to evaluate the vegetation annually. More information: Rennie et al. (2016) and 548 https://catalogue.ceh.ac.uk/documents/b98efec8-6de0-4e0c-85dc-fe4cdf01f086 and 549 https://catalogue.ceh.ac.uk/documents/d349babc-329a-4d6e-9eca-92e630e1be3f. 550 551 55. The dataset is issued from an experiment belonging to the Andrews Forest LTER 552 (AND-LTER) and located in a forest in the Oregon Cascade Range, USA (LAT: 553 44.22, and LON: -122.25). This dataset consists of 5 plots, where each plot was 554 sampled an average of 10.0 times. The vegetation cover (%) was visually estimated 10 555 times in a quadrat of 4 m2 for trees (vegetation > 60 cm tall) and 9 quadrats (0.1 m2) 556 for herb and low shrub (< 60 cm tall). More information: Rothacher (Rothacher 2013) 557 and https://portal.lternet.edu/nis/mapbrowse?packageid=knb-lter-and.3190.7. 558 559 56. The dataset is issued from an experiment belonging to the Park Grass permanent 560 grassland and located in Rothamsted, United Kingdom (LAT: 51.81, and LON: - 19 Page 20 of 147 Journal of Vegetation Science 561 0.37). This dataset consists of 74 plots, where each plot was sampled an average of 562 9.9 times. The purpose of the experiment was to evaluate different fertility and lime 563 treatments. Herbage was taken from six randomly located quadrats measuring 0.5 m x 564 0.25 m within each plot, resulting in a total sampling area of 0.75 m2 within each plot. 565 In each plot, the biomass of each species was measured annually in quadrats 566 (sampling area: 0.75 m2). More information: Crawley et al. (2005) and 567 http://www.era.rothamsted.ac.uk/Park. 568 569 57. The dataset is issued from an experiment located on a savannah in central Spain 570 (LAT: 40.38, and LON: -4.20). This dataset consists of 210 plots, where each plot was 571 sampled an average of 6.0 times. The experiment evaluated two types of pastures 572 (higher-productivity pastures and low-productivity pastures) and three treatments 573 (ungrazed, grazed by small herbivores, and grazed by large and small herbivores). In 574 each plot, the species cover (%) was visually estimated. More information: Rueda et 575 al. (2013). 576 577 58. The dataset is issued from an experiment located in Central Germany (LAT: 578 51.55, and LON: 10.07). This dataset consists of 14 plots, where each plot was 579 sampled an average of 14.9 times. In each plot, species vegetation cover (%) was 580 visually estimated. More information: Schmidt (2007). 581 582 59. The dataset is issued from an experiment located on a former arable field in the 583 Experimental Botanical Garden of the University of Göttingen, Germany (LAT: 584 51.56, and LON: 9.96). This dataset consists of six plots, where each plot was 585 sampled an average of 38.0 times. In each plot, species vegetation cover (%) was 586 visually estimated. More information: Schmidt (Schmidt 2006) and Bernhardt- 587 Römermann et al. (2011). 588 589 60. The dataset is issued from an experiment located in the Swiss National Park 590 (IUCN Ia reserve, LAT: 46.68, and LON: 10.22). This dataset consists of 41 plots, 591 where each plot was sampled an average of 12.2 times. In each plot, plant species 592 cover (%) was visually estimated. More information: Braun-Blanquet et al. (1931), 593 Schütz et al. (2000). 594 20 Journal of Vegetation Science Page 21 of 147 595 61. The dataset is issued from an experiment belonging to the Sevilleta LTER (SEV- 596 LTER) and located in Sevilleta National Wildlife Refuge, New Mexico, USA (LAT: 597 34.31, and LON: -106.49). This dataset consists of 95 plots, where each plot was 598 sampled an average of 9.8 times. The experiment was designed to evaluate the effect 599 of prescribed burning (two areas were left unburned as control treatments, and the 600 other plots were burned in different dates) and grazing exclusion (fenced and 601 unfenced). In each plot, the individuals present in 36 quadrats (0.1 m2) were recorded. 602 More information: https://portal.lternet.edu/nis/mapbrowse?packageid=knb-lter- 603 sev.148.131885. 604 605 62. The dataset is issued from an experiment belonging to the Sevilleta LTER (SEV- 606 LTER) and located in Sevilleta National Wildlife Refuge, New Mexico, USA (LAT: 607 34.33, and LON: -106.74). This dataset consists of 81 plots, where each plot was 608 sampled an average of 9.2 times. The experiment had three treatments: 1) control 609 plots (natural rainfall regime) 2) drought was induced by rainfall shelters, and 3) 610 watering was applied by redirecting the water from the nearby rainfall shelters. In 611 each plot, the plant cover (%) was estimated every spring. More information: 612 https://portal.lternet.edu/nis/mapbrowse?packageid=knb-lter-sev.147.167839. 613 614 63. The dataset is issued from an experiment belonging to the Sevilleta LTER (SEV- 615 LTER) and located on a grassland in Sevilleta National Wildlife Refuge, New 616 Mexico, USA (LAT: 34.33, and LON: -106.63). This dataset consists of 216 plots, 617 where each plot was sampled an average of 7.7 times. The experiment evaluated the 618 impact of prairie dog reintroduction (grazed and ungrazed areas) on vegetation. In 619 each plot, the plant cover (%) was estimated annually. More information: 620 https://portal.lternet.edu/nis/metadataviewer?packageid=knb-lter-sev.212.4. 621 622 64. The dataset is issued from an experiment belonging to the Sevilleta LTER (SEV- 623 LTER) and located on a woodland in Sevilleta National Wildlife Refuge, New 624 Mexico, USA (LAT: 34.37, and LON: -106.54). This dataset consists of 100 plots, 625 where each plot was sampled an average of 13.0 times. In each plot, the plant cover 626 (%) was visually estimated annually. More information: 627 https://portal.lternet.edu/nis/mapbrowse?packageid=knb-lter-sev.278.245672. 628 21 Page 22 of 147 Journal of Vegetation Science 629 65. The dataset is issued from an experiment belonging to the Sevilleta LTER (SEV- 630 LTER) and located in Sevilleta National Wildlife Refuge, New Mexico, USA (LAT: 631 34.37, and LON: -106.58). This dataset consists of 100 plots, where each plot was 632 sampled an average of 16.4 times. The experiment evaluated three treatments: 1) 633 control plots (untouched vegetation), 2) removal of all three dominant species (Larrea 634 tridentata, Bouteloua eriopoda, Bouteloua gracilis), and 3) removal of one dominant 635 species. In each plot, the plant cover (%) was visually estimated annually. More 636 information: https://portal.lternet.edu/nis/mapbrowse?packageid=knb-lter- 637 sev.168.192543. 638 639 66. The dataset is issued from an experiment belonging to the Shortgrass Steppe 640 LTER (SGS-LTER) and located on grasslands and shrublands in Central Plains 641 Experimental Range, Colorado, USA (LAT: 40.85, and LON: -104.77). This dataset 642 consists of 18 plots, where each plot was sampled an average of 8.2 times. In each 643 plot, the plant cover was recorded on three permanent transects (1 m2: sum of plots 644 along the transect). More information: Stapp (Stapp 2013) and 645 https://portal.lternet.edu/nis/mapbrowse?packageid=knb-lter-sgs.140.17. 646 647 67. The dataset is issued from an experiment located in a beech forest near Göttingen, 648 Central Germany (LAT: 51.57, and LON: 10.32). This dataset consists of seven plots, 649 where each plot was sampled an average of 18.0 times. Four plots had a fertilization 650 treatment (NP addition) and three were the control plots. In each plot, the species 651 cover (%) was visually estimated. More information: Schmidt (2009). 652 653 68. The dataset is issued from an experiment located on a meadow near Zvíkov, 654 Czech Republic (LAT: 48.99, and LON: 14.61). This dataset consists of 40 plots, 655 where each plot was sampled an average of 10.3 times. The experiment evaluated four 656 treatments: 1) control (intact vegetation), 2) mycorrhizal grasses and forbs left, non- 657 mycorrhizal species weeded out, 3) mycorrhizal forbs remaining, everything else 658 weeded out, and 4) mycorrhizal grasses remaining, everything else weeded out. In 659 each plot, the species cover (%) was visually estimated annually. More information: 660 Šmilauer & Šmilauerová (2013). 661 22 Journal of Vegetation Science Page 23 of 147 662 69. The dataset is issued from an experiment located on a floodplain grassland in 663 Anloo and Taarlo, The Netherlands (LAT: 53.05, and LON: 6.66). This dataset 664 consists of 80 plots, where each plot was sampled an average of 28.9 times. In each 665 plot, the species cover (%) was estimated almost every year from 1973 to 2008. 666 Dataset owners: Christian Smit and Jan P. Bakker (Conservation Ecology Group, 667 Groningen Institute for Evolutionary Life Sciences). 668 669 70. The dataset is issued from an experiment located on a meadow in the north-eastern 670 Tibetan Plateau in Qinghai Province, China (LAT: 37.62, and LON: 101.20). This 671 dataset consists of 30 plots, where each plot was sampled an average of 9.0 times. The 672 experiment was designed to evaluate 10 nitrogen treatments (no N added and 9 673 combinations of three N forms and three N rates). In each plot, the species cover (%) 674 was visually estimated annually. More information: Song et al. (2012). 675 676 71. The dataset is issued from an experiment located on salt marshes of the 677 Schleswig-Holstein Wadden Sea National Park in Hamburger Hallig and 678 Westerhever, Germany (LAT: 54.49, and LON: 8.75). This dataset consists of 212 679 plots, where each plot was sampled an average of 18.7 times. There were two 680 treatments in Westerhever: natural condition and intensive grazing, and only natural 681 conditions in Hamburger Hallig. In each plot, the species cover was measured 682 annually using the Londo scale (percentage of vegetation cover) from 1997 to 2015 in 683 Hamburger Hallig and from 1995 to 2012 in Westerhever. Dataset owner: Martin 684 Stock (Wadden Sea National Park of Schleswig-Holstein). 685 686 72. The dataset is issued from an experiment located on a wooded savanna in 687 Laikipia, Kenya (LAT: 0.28, and LON: 36.87). This dataset consists of 18 plots, 688 where each plot was sampled an average of 14.7 times. The treatments were six 689 combinations (3 replicates) of cattle, wildlife, and mega-herbivore grazing. These 690 either allowed (1) the entry of all large mammalian herbivores, (2) all large 691 mammalian herbivores except mega-herbivores (elephants Loxodonta africana and 692 giraffe Giraffa camelopardis) to enter, or (3) excluded all large herbivores. In each 693 plot, vegetation was assessed annually by counting the number of pins hit by each 694 species over a ten-point pin frame at each station. More information: Veblen et al. 695 (2016). 23 Page 24 of 147 Journal of Vegetation Science 696 697 73. The dataset is issued from an experiment located on a coastal heathland in 698 Lurekalven, Norway (LAT: 60.70, and LON: 5.08). This dataset consists of 42 plots, 699 where each plot was sampled an average of 6.0 times. In each plot, all vascular plants, 700 bryophytes and lichens were recorded annually using frequencies (1 m x 1 m metal 701 frame divided into 16 subplots). More information: Vandvik et al. (2005). 702 703 74. The dataset is issued from an experiment located in Bonanza Creek LTER, 704 Alaska, USA (LAT: 65.00, and LON: -148.00). This dataset consists of 59 plots, 705 where each plot was sampled an average of 12.0 times. In each plot, the species cover 706 707 For (%) was visually estimated. More information: Viereck et al. (2010) and https://portal.lternet.edu/nis/mapbrowse?packageid=knb-lter-bnz.174.19. 708 709 75. The dataset is issued from an experiment located on desert steppes in Gobi Gurvan 710 Saykhan National Park, Mongolia (LAT: 43.61, and LON: 104.13). This dataset 711 consists of 18 plots, where each plot was sampled an average of 7.1 times. The 712 experiment evaluated two treatments: 1) exclusion of large ungulates, and 2) no 713 exclusion of large ungulates. In each plot, the species cover (%) was visually 714 estimated annually. More information: Wesche et al. (2010). 715 716 76. The 9dataset is issued from an experiment located on a floodplain grassland on 717 formerly arable land (LAT: 51.78, and LON: -1.31). From 1989 the site was divided 718 into nine plots of c. 0.4 ha over which three contrasting grazing management practices 719 (control, cattle and sheep) were randomly superimposed. These nine plots were 720 monitored in June of each year from 1991-2009. More information: Woodcock et al. 721 (2011). 722 723 77. The dataset is issued from an experiment located on a grassland in southeast 724 Estonia (LAT: 58.11, and LON: 27.07). This dataset consists of 55 plots, where each 725 plot was sampled an average of 8.9 times. The treatments were: fertilizer, sucrose and 726 control. In each plot, the species cover (%) was visually estimated annually. More 727 information: Liira et al. (2012). 728 24 Journal of Vegetation Science Page 25 of 147 729 Figure Table S2. Characteristics of the study sites. 730 ID Country Biome Habitats Duration CP MP 1 USA TGD Grassland 12.5 NO YES 2 USA WS Grassland 34.5 YES YES 3 USA TGD Grassland 13.5 YES YES 4 USA SD Grassland 11.2 YES YES 5 USA TGD Savanna 21.5 YES YES 6 USA TGD Grassland 8.0 YES YES 7 USA TGD Grassland and Shrubland 11.5 YES YES 8 South Africa SD Grassland 10.7 YES NO 9 South Africa SD Savanna 15.8 YES NO 10 South Africa TGD Savanna 14.7 YES NO 11 South Africa SD Savanna 16.0 YES NO 12 Hungary WS Savanna 14.5 YES YES 13 USA TF Grassland 7.0 YES YES 14 USA TF Savanna 6.2 YES YES 15 USA TF Grassland 24.8 YES NO 16 USA TF Grassland 22.0 YES YES 17 USA TF Grassland 14.8 YES YES 18 USA TGD Grassland 13.5 YES YES 19 Germany WS Grassland 6.0 YES NO 20 United Kingdom WS Shrubland 6.0 NO YES 21 USA TRF Forest 21.4 YES YES 22 Australia SD and WS Savanna 10.2 YES NO 23 United Kingdom WS Grassland 8.0 YES YES 24 France WS Grassland 28.0 NO YES 25 Russia BF Grassland 24.3 YES NO 26 United Kingdom TF Shrubland 12.0 YES YES 27 USA TF Grassland 10.0 YES NO 28 USA TGD Grassland 27.8 YES NO 29 Czech Republic TF Grassland 16.8 YES NO 30 Czech Republic TF Grassland 29.8 YES NO 31 USA WS Grassland 18.9 YES YES 32 Spain WS Grassland 12.0 YES YES 33 USA Tu Grassland 6.0 YES YES 34 USA Tu Grassland 6.0 YES NO 35 Germany WS Grassland 7.7 YES YES 36 Czech Republic WS Grassland 8.0 YES NO 37 France TF Grassland 13.0 NO YES 38 USA TGD Grassland 9.0 YES YES 39 Czech Republic WS Grassland 16.0 YES YES 40 France WS Grassland 8.0 YES YES 41 USA TGD Grassland, Shrubland and Savanna 14.3 YES NO 42 United Kingdom TF Grassland 6.8 YES YES 43 United Kingdom TF and TRF Grassland 6.0 YES YES 25 Page 26 of 147 Journal of Vegetation Science 44 United Kingdom TF Grassland 7.0 YES YES 45 United Kingdom TF Grassland 6.0 YES YES 46 United Kingdom TF Shrubland 6.0 NO YES 47 United Kingdom TF Savanna 6.0 YES YES 48 Spain BF Grassland 19.0 YES YES 49 Spain TGD Grassland 11.5 YES NO 50 Spain WS Shrubland 17.0 YES YES 51 USA TGD Grassland, Shrubland and Savanna 24.0 YES NO 52 United Kingdom TF Savanna 6.0 NO YES 53 United Kingdom TF Shrubland 10.0 YES YES 54 United Kingdom TF and WS Grassland, Savanna and Forest 11.1 YES NO 55 USA TF Forest 10.0 YES NO 56 United Kingdom WS Grassland 9.9 YES YES 57 Spain TGD Savanna 6.0 YES YES 58 Germany WS Grassland 14.9 YES NO 59 Germany WS Grassland 38.0 NO YES 60 Switzerland BF Grassland and Forest 12.2 NO YES 61 USA TGD Savanna 9.8 YES YES 62 USA TGD Grassland, Shrubland and Savanna 9.2 YES YES 63 USA TGD Grassland 7.7 YES YES 64 USA TGD Forest 13.0 YES NO 65 USA TGD Grassland and Savanna 16.4 YES YES 66 USA TGD Grassland and Shrubland 8.2 YES NO 67 Germany WS Forest 18.0 YES YES 68 Czech Republic WS Grassland 10.3 YES YES 69 Netherlands WS Grassland 28.9 NO YES 70 China WS Grassland 9.0 YES YES 71 Germany WS Salt marsh 18.7 YES YES 72 Kenya WS Savanna 14.7 NO YES 73 Norway TRF Grassland 6.0 NO YES 74 USA BF Grassland and Savanna 12.0 YES NO 75 Mongolia TGD Grassland 7.1 YES YES 76 United Kingdom WS Grassland 18.0 NO YES 77 Estonia WS Grassland 8.9 YES YES 731 26 Journal of Vegetation Science Page 27 of 147 732 ID: Identification of the data set, biomes (TGD: temperate grassland desert, SD: 733 subtropical desert, WS: woodland shrubland, TF: temperate forest, BF: boreal forest, 734 Tu: Tundra, and TRF: temperate rain forest), Duration: Average number of years of 735 the dataset, CP: presence of plots where the long-term conditions prior to the 736 establishment of the sampling scheme were maintained throughout the sampling, MP: 737 presence of plots exposed to different treatments that altered the long-term conditions. 27 Page 28 of 147 Journal of Vegetation Science 28 Journal of Vegetation Science Page 29 of 147 738 Supporting Information to the paper 739 Valencia et al. Directional trends in species composition over time can lead to a 740 widespread overestimation of asynchrony. Journal of Vegetation Science. 741 742 Appendix S3. Application of the analyses shown in Fig. 1 of the main text to the 743 three remaining indices of synchrony. 744 745 Gross 746 Note: on panel (b) the mean synchrony values with the T3 approach per datasets are 747 significantly higher than without the T3 approach (p < 0.001, paired t-test) 748 749 750 751 Logvar 752 Note: on panel (b) the mean synchrony values with the T3 approach per datasets are 753 significantly higher than without the T3 approach (p < 0.06, paired t-test) 754 755 756 757 29 Page 30 of 147 Journal of Vegetation Science 758 Phi 759 Note: on panel (b) the mean synchrony values with the T3 approach per datasets are 760 significantly higher than without the T3 approach (p < 0.22, paired t-test) 761 762 30 Journal of Vegetation Science Page 31 of 147 763 Supporting Information to the paper 764 Valencia et al. Directional trends in species composition over time can lead to a 765 widespread overestimation of asynchrony. Journal of Vegetation Science. 766 767 Appendix S4. Application of the analyses shown in Fig. 3 of the main text to the 768 three remaining indices of synchrony. For each index, also, a table of number of 769 datasets with either positive or negative significant t-statitstic values is reported for 770 both manipulated and control plots (positive means that the T3 approach increased 771 synchrony; negative means that the T3 approach decreased synchrony). The grey area 772 in each panel reports and approximate are where t-statistic values were not significant 773 (‘ns’). 774 775 Gross For manipulated 'control' 25 25 (a) (b) 20 20 Review 15 15 T−statistic (synT3 − syn) 10 10 T3>synch 5 5 ns 0 0 T3synch 5 5 ns 0 0 For T3synch 5 5 ns 0 0 T3