1 Freshwater ecotoxicity assessment of pesticide 2 use in crop production: Testing the influence of 3 modeling choices 4 Nancy Peñaa,b*, Marie T. Knudsend, Peter Fantkec, Assumpció Antóna and John E. 5 Hermansend a 6 IRTA. Torre Marimon, Ctra c-59 Km. 12.1 E-08140 Caldes de Montbui (Barcelona), Spain. b 7 Institute of Environmental Science and Technology (ICTA), Universitat Autónoma de 8 Barcelona (UAB), E-08193 Bellaterra (Barcelona), Spain. c 9 Quantitative Sustainability Assessment Division, Department of Management Engineering, 10 Technical University of Denmark, Bygningstorvet 116, 2800 Kgs. Lyngby, Denmark d 11 Department of Agroecology, Aarhus University, DK-8830 Tjele, Denmark 12 13 Corresponding author: Nancy Peña 14 *e-mail: nancy.pena@irta.cat; phone: +34 934 674 040 ext. 1203 15 Complete corresponding address: 16 Institute for Food and Agricultural Research and Technology-IRTA 17 Torre Marimon, ctra. C-59, Km 12.1 E-08140 Caldes de Montbui, Barcelona, Spain. 18 This is an Accepted Manuscript of the article Peña, N., Knudsen, M. T., Fantke, P., Anton, A., & Hermansen, J. E. (2019). Freshwater ecotoxicity assessment of pesticide use in crop production: Testing the influence of modeling choices. Journal of Cleaner Production, 209, 1332-1341, available online at: https://doi.org/10.1016/j.jclepro.2018.10.257. ©2018. This manuscript version is made available under the CC-BY-NC-ND 4.0 license. 1 19 ABSTRACT 20 Pesticides help to control weeds, pests and diseases contributing, therefore, to food 21 availability. However, pesticide fractions not reaching the intended target may have adverse 22 effects on the environment and the field ecosystems. Modeling pesticide emissions and the 23 alignment with characterizing associated impacts is currently one of the main challenges in 24 Life Cycle Assessment (LCA) of agricultural systems. To address this challenge, this study 25 takes advantage of the latest recommendations for pesticide emission inventory and impact 26 assessment and frames a suitable interface for those LCA stages and the related mass 27 distribution of pesticide avoiding a temporal overlapping. Here, freshwater ecotoxicity 28 impacts in the production of feed crops (maize, grass, winter wheat, spring barley, rapeseed 29 and peas) in Denmark are evaluated during a 3-year period, testing the effects of inventory 30 modelling choice and recent updates of the characterization method (USEtox). Potential 31 freshwater ecotoxicity impacts were calculated in two functional units to consider crop impact 32 profiles and cultivation intensity. According to the results, ecotoxicity impacts decreased over 33 the period, mainly because of the reduction of insecticide active ingredients (e.g., 34 cypermethrine). Three different emission modelling choices were tested; they differ on the 35 underlining assumptions and data requirements. The median results for the resulting emission 36 fractions vary ~4 orders of magnitude for the different models. Main aspects influencing 37 impact results are the interface between inventory estimates and impact assessment, and the 38 consideration of intermedia processes, such as crop growth development and pesticide 39 application method. Statistical differences were found in the impact results with 2 of emission 40 model tested, thereby indicating the influence of modelling choices on ecotoxicity impact 41 assessment. 42 43 Keywords: Pesticide emission factors, inventory modeling, ecotoxicity characterization, life 44 cycle impact assessment (LCIA), feed crops, agriculture.*1 1 Abbreviations Gly_agri: Total agricultural use of glyphosate AI: Active ingredient Hrb: Herbicides AS: Alternative scenario Ins: Insecticides BS: Reference scenario IS: Impact scores CF: Characterization factors LAI: Leaf area index DK: Denmark NAP: National Action Plans EF: Effect factor Pgr: Plant growth regulators FF: Fate factor XF: Exposure factor Fun: Fungicides GAP: Good agricultural practices 2 45 1 INTRODUCTION 46 With the increased global demand for agricultural products for food, fiber and bioenergy, and 47 the interrelated concerns on the environmental impact hereof, there is a need to have efficient 48 tools to evaluate the environmental performance profiles of agricultural production, to 49 facilitate a move towards more sustainable production systems. Life Cycle Assessment (LCA) 50 is a widely applied and standardized framework to quantify potential impacts of products and 51 systems along their entire life cycles. One of the main challenges in assessing the 52 environmental performance of agricultural systems in LCA is modeling emissions from 53 pesticide use and the subsequent coupling with the impact characterization model (van Zelm 54 et al., 2014). Over the past years, a significant number of LCA studies on agricultural systems 55 were conducted; however, ecotoxicity impacts as currently modelled may lead to inconsistent 56 results and wrong conclusions in few cases (e.g., comparing conventional and organic 57 farming), mostly due to the lack of agreement and precise definitions on the modeling 58 framework for this impact category (Notarnicola et al., 2017). 59 The development of the life cycle inventory (LCI) analysis and subsequent life cycle impact 60 assessment (LCIA) (e.g., pesticide emission quantification and related characterization of 61 ecotoxicity impacts) are the core phases of any LCA study. The robustness and reliability of 62 the LCA results depend mainly on the quality and representativeness of the LCI and LCIA 63 data and models selected. Different modeling options, hence, will affect the impact profiles of 64 a study, and this is especially relevant for agricultural systems (Anton et al., 2014). 65 Quantifying the chemical emissions to the environment in the LCI phase is typically based on 66 generic assumptions, often based on standard emission factors (e.g., expressed in percentages 67 of applied mass) or dynamic models based on specific application scenarios that describe the 68 emission distribution of organic pesticides. The consensus effort on the delimitation between 69 pesticide emission inventory and impact assessment for LCA already provides guidelines on 70 what should be quantified in those LCA steps but explicitly exclude how to do it avoiding 3 71 recommendations on specific models (Rosenbaum et al., 2015). The implications of choosing 72 different emission models in the LCA of crop production have been discussed for some 73 agricultural systems (Goglio et al., 2018; Schmidt Rivera et al., 2017; van Zelm et al., 2014). 74 However, no studies are addressing the influence of the pesticide emission modeling 75 approach, nor the evaluation of recent developments in impact assessment methods to 76 determine pesticide ecotoxicity impact profiles in different crop production systems. 77 Thus, there is a need to test different choices on how to quantify pesticide emission fractions 78 (i.e. different modeling approaches) and the recent developments on the recommended 79 method for freshwater ecotoxicity characterization in the production of feed crops. 80 The purpose of the present study is to contribute to the evaluation of the ecotoxicological 81 burden on freshwater ecosystems from pesticide use in crop production using the pesticide 82 use in Denmark as example. It is focused on assessing the influence of pesticides on the 83 environmental impact profiles of feed crops (maize, grass, wheat, barley, rapeseed and peas) 84 during the period 2013-2015, testing the effects of the LCI choice and the developments of 85 LCIA methodology. 86 2 MATERIALS AND METHODS 87 This study followed the LCA methodology to evaluate the potential ecotoxicity impacts on 88 freshwater ecosystems from pesticide use in Denmark´s (DK) crop production. This bottom- 89 up analysis focuses on the evaluation and influence of pesticide application on the 90 environmental impact profiles of maize, winter wheat, grass, spring barley, rapeseed and peas 91 during the period 2013-2015, testing the effects of the choice of the emission modeling 92 framework and the recent updates of the characterization method. For the later, we use the 93 global consensus model USEtox (http://usetox.org). 94 2.1 Definition of ecotoxicity impact scores 95 The quantification of ecotoxicity impact scores for freshwater ecosystems includes 1) detailed 96 LCI reporting on the pesticide active ingredient (AI), application methods, time and mass, 4 97 location, agricultural practices and crop stage development; 2) quantified AI emission 98 fractions for both on-field and off-field; and 3) mesures to avoid double counting of 99 multimedia transfers considered in the quantification of emission fractions and the impact 100 assessment fate modeling (Rosenbaum et al., 2015). Accordingly, the freshwater ecotoxicity 101 impact scores (IS) can be described as: 102 =∑, ( , ∙ , ) (1) 103 Where CFi,x is the characterization factor for freshwater ecotoxicity [PAF m3 d kg ], 104 and mi,x is the mass of AI x emitted to compartment i per area treated [kg ha-1]. 105 Potential freshwater ecotoxicity impacts (IScrop_ha) [PAF m3 d ha-1] were determined in 106 relation to 1 hectare [ha] of crop in a given year t within 2013 and 2015 (cultivation intensity). 107 Additionally, freshwater ecotoxicity impact profiles at country or regional level (IScrop) [PAF 108 m3 d crop-1] from pesticide use were derived from the product of crop impact scores and the 109 total crop area in a given year in DK. 110 The interface between LCI and LCIA and related mass distribution for pesticide application in 111 crop production are presented in Figure 1. This approach follows the proposed framework for 112 pesticide inventory and impact assessment (Rosenbaum et al., 2015; van Zelm et al., 2014). 113 Figure 1. 114 This interface considers the boundaries between the emission inventory and impact 115 assessment, setting also spatial and time dimensions, to quantify the AI emission fractions (in 116 air, freshwater and soil) and characterize ecotoxicity impacts, avoiding any overlap or double 117 counting of the chemical fate process. Furthermore, the emission flows, both on and off the 118 field, are clearly indicated and their link to the characterization factors for the impact pathway 119 (i.e. freshwater ecotoxicity). 120 2.2 Pesticide emission inventory 121 Pesticide application practices in DK for the selected crops were determined. Concrete active 122 ingredients were used throughout the study, meaning, that the chemical that is the biologically 5 123 active part in any pesticide was assessed (European Commission, 2017). The mass applied per 124 AI was derived from the annual statistical report on pesticide use by crop in DK for 2013 125 (Ørum and Samsøe-Petersen, 2014), 2014 (Ørum and Hossy, 2015) and 2015 (Ørum and 126 Holtze, 2017); for further information see Supporting information (SI-1). We addressed 127 nearly 60 different AIs from four distinct target classes, herbicides (Hrb), plant growth 128 regulators (Pgr), fungicides (Fun), and insecticides (Ins). Additionally, glyphosate (CAS- 129 RN107-83-6) use is not allocated to any specific crop cultivation, and it was assessed as the 130 total agricultural use of the AI per 1 hectare [ha] in a given year, hereafter identified as 131 (Gly_agri). All AI identification (CAS registry numbers-RN and names), and class are reported 132 in SI, Table S1. 133 2.3 Pesticide emission quantification 134 Crops are treated by foliar spray application (typically boom sprayers), and the reported DK 135 statistics on pesticides were used for agricultural practices. The agricultural field is considered 136 as part of the ecosphere and emissions to environmental media after spraying were modeled 137 via initial distribution (primary processes like initial drift deposition) and secondary emission 138 transfers (e.g., re-volatilization after deposition). The total emission fraction of an AI [kg kg-1] 139 is quantified as the sum of the fractions emitted to air, freshwater and soil: 140 = = _ + _ + _ . + _ . (2) 141 Where fem is the fraction of the applied mass of pesticide that becomes an emission to the 142 environment, mem the mass emitted, mapp the mass of pesticide applied, fem_air the fraction of 143 applied mass that is emitted to air, fem_fw the fraction of applied mass that is emitted to 144 freshwater, fem_soil.agri the fraction of applied mass that is emitted to on-field soil and fem_soil.other 145 the emission fraction reaching off-field soil and other surfaces. 146 Primary distribution 147 The primary distribution processes between compartments occur during the initial minutes 148 after pesticide application. This primary process are emission by wind drift (fd_lost), pesticide 6 149 deposition process and the fraction intercepted by the crop or weed. Since the fractions from 150 initial distribution to environmental media should sum up to 100% of the applied mass, 151 considering losses via degradation during the initial minutes negligible, the aggregated 152 emission fractions will be equal to one (Fantke et al., 2011a; Juraske et al., 2007). 153 Consequently, the crop/weed interception fraction (fint_crop) of an AI directly after the 154 application is given by: 155 _ =1− _ + _ . (3) 156 The fraction lost by wind drift fd_lost [kg/kg], depends on the application method, i.e. the spray 157 equipment and elevation, and wind speed. Based on models for conventional spray equipment 158 on field crops and deposition curve parameters assuming good agricultural practices (GAP), 159 the fd_lost was fixed to a value of 0.1 (Gil et al., 2014; Gil and Sinfort, 2005; Gyldenkrne et al., 160 1999; van de Zande et al., 2007). The soil deposition fdep_soil.agri [kg kg-1], depends on crop- 161 specific leaf area index (LAI), thereby also affecting fractions reaching soil surfaces of the 162 treated field area (Fantke et al., 2011b). With an exponential model (Gyldenkærne et al., 163 2000; Juraske et al., 2007), based on crop growth stage and capture efficacy, the fraction 164 reaching the soil surface is described as: × 165 _ . = (4) 166 Where kp is the capture coefficient [-] and set to 0.55 for pesticide spray solutions prepared 167 with adjuvants (Gyldenkrne et al., 1999). Pesticide target class and specific application time 168 were used to define crop-specific growth stages in the selected crops. The LAI was derived 169 for plant growth regulators, insecticides and fungicides distinctly as a value dependent on the 170 target class/crop growth stage/application time combination, (Fantke et al., 2011b; Itoiz et al., 171 2012; Olesen and Jensen, 2013); for herbicide application on weeds the corresponding LAI of 172 0.5 is used. This value is based on the reported leaf cover factor for fallow lands (Panagos et 173 al., 2015). Further details presented in SI, Table S2. 174 Secondary distribution 7 175 The subsequent secondary emission transfers include re-volatilization after deposition and 176 off-field emissions allocation. The volatilization from fractions deposited in the different 177 compartments is derived from the default Tier 1 emission factors per AI from their vapor 178 pressures (Webb et al., 2016) see Table S1 and S3 in SI. The emission factor emF was 179 calculated for each AI (see, SI Table S1), the inter-media transfer and the final emission 180 factors are presented in SI, SI-1 and SI-2. Finally, the water to soil area ratio for DK (0.016) 181 was used to allocate the off-field emissions (i.e. drift fraction deposited in off-field surfaces) 182 see SI, Table S2. This value is based on reported data of the Danish ministry of environment 183 (Stockmarr and Thomsen, 2009). 184 2.4 Freshwater ecotoxicity characterization 185 For assessing the ecotoxicity of pesticides on freshwater ecosystems, we followed the LCIA 186 emission-to-damage framework that links emissions to impacts through environmental fate, 187 exposure and effects (Jolliet et al., 2004). According to (Hauschild and Huijbregts, 2015; 188 Rosenbaum et al., 2008) characterization factors CF for freshwater ecotoxicity of chemical 189 emissions can be expressed as: 190 CF , = → , × , × , (5) 191 Where FFifw,x is the fate factor in [d] describing the mass transport, distribution and 192 degradation in the environment. The ecosystem exposure factor, XFfw,x, is defined as the 193 bioavailable fraction of a chemical in freshwater; and an effect factor (EFfw,x) expressing the 194 ecotoxicological effects in the exposed ecosystems integrated over the exposed water volume. 195 CFs were estimated with USEtox 2.02 as characterization model, with the specific European 196 landscape dataset (i.e. representing DK conditions) (Fantke et al., 2017; Westh et al., 2015). 197 New CFs for 10 additional AIs, following the procedure in Fantke et al. (2017) were derived. 198 A detailed description of the resulting CF and the data used can be found in SI, SI-3. 199 Furthermore, the recent developments for the characterization model between USEtox 200 versions 1.01 and 2.02 were evaluated. 8 201 2.5 Sensitivity analysis definition 202 Two types of local sensitivity tests were conducted. First, a scenario sensitivity analysis was 203 performed to test the effect of LCI modeling choices on the impact profile of the selected 204 crops on the three-year period. Three scenarios were considered, the above-described scenario 205 was selected as a reference case (BS) and two alternative scenarios (AS1-AS2) that represent 206 different modeling approaches to quantify emissions from pesticide use. The alternative 207 scenario AS1 followed Margni et al. (2002), which represents a usually used pesticide 208 emission modeling, and furthermore is one of the first approaches that account for pesticide 209 emission distribution in different environmental media in LCA studies for agricultural 210 systems. In this approach, the pesticide emissions are distributed in environmental media 211 based on fixed share percentages. They assume that the fraction of AI emitted to the soil will 212 be 85% of the total application, 5% will stay on leaves and the remaining 10% is lost into the 213 air across crops and pesticides. The second tested scenario AS2 represents fixed emission 214 fractions dependent on the foliar spray application and drift distributions for field crops. This 215 approach was chosen to represent a modeling framework where the initial distribution (i.e. 216 application method and crop relation) is taken into account but also allowing the inclusion of 217 field emissions in the assessment (Balsari et al., 2007; Felsot et al., 2010; Gil and Sinfort, 218 2005). Table 1 displays the emission fractions in the three scenarios considered. 219 Table 1. 220 Second, the sensitivity of the proposed modeling approach was tested by evaluating the 221 change in the impact scores (propagated from the change in emission fractions) as a function 222 of the variation of several input parameters by a factor of 2 larger of their initial values, one at 223 the time. Local sensitivity to input Sin [-] was further expressed as the effect on the model 224 output due to a change in an input parameter (for further details see SI, SI-5). 9 225 3 RESULTS AND DISCUSSION 226 3.1 Pesticides use in Danish crop production (2013-2015) 227 The AI considered in the study covers 98.3% of the total pesticide applications in terms of 228 mass applied for the selected crops: maize, winter wheat, grass, spring barley, rapeseed, peas 229 and the agricultural use of glyphosate (Gly_agri). The total pesticide use was 3165 tons in 2013, 230 1438 tons in 2014 and 2105 tons in 2015. The average pesticide application rates per crop 231 vary between 2 and 3 orders of magnitude (SI, Table S6). Grass is the crop with the lowest 232 application rates and pesticide use; together, fungicides and insecticides represent nearly 20% 233 of the total use in grass-2013; additionally, in 2014-2015, there was no use of insecticides, 234 and fungicides use was reduced by less than 2.5%. Gly_agri sum up to 2722 tons in the 3 235 years and represents near 40% of the total use of pesticides in DK. Winter wheat (2672 tons) 236 is the crop with higher pesticide use followed by spring barley (748 tons) (SI, Table S7). The 237 most used pesticide target class is Herb and prosulfocarb is the most used AI after Gly_agri 238 on this target class. 239 3.2 Ecotoxicity impact profiles of feed crops (2013-2015) 240 The IScrop from pesticide use decreased over the three years (Figure 2). The reduction of the 241 IScrop was more apparent in 2014 (59%) than in 2015 (33%) with respect to the base year 242 (2013). Most of the decrease in the IScrop was due to the non-use of a single substance: 243 cypermethrin. This insecticide was the major contributor to IScrop in 2013 across crops (e.g., 244 87% in maize, 60% in spring barley and 47% in winter wheat) and was no longer used in 245 2014-2015 (see Table S8 in SI). Furthermore, the fact that maize and grass did not require the 246 use of insecticides in 2015 also contributes to the reduction of IScrop, but it is essential to note 247 that this may be the result of unfavourable climatic conditions for the emergence of pests, 248 among many other different reasons. 249 Figure 2. 10 250 After winter wheat-2013 (1.6x109 PAF m3 d crop-1), spring barley-2013 (1.4x109 PAF m3 d 251 crop-1) and rapeseed-2013 (3.3x108 PAF m3 d crop-1) present the higher IScrop (Figure 2). The 252 larger IScrop in those crops is associated with the use of insecticides (e.g., cypermethrin, 253 pendimethalin and lambda-cyhalothrin) and fungicides (e.g., pyraclostrobin, azoxystrobin and 254 folpet), AIs with relatively high CF, and the more extensive cultivation practices (i.e. 255 cultivated area). Consequently, substance prioritization by LCA impact assessment helps to 256 identify potentially harmful AI for ecosystems and, with the restriction of their use or the 257 implementation of more sustainable practices, significant changes in the impact profiles of the 258 crops can be made more apparent (e.g., cypermethrin). In this sense, if farmers choose to use 259 pesticides AI causing lower impacts, the load on agricultural systems will decline, even if 260 they continue to spray their fields as usual for pests and disease control. Moreover, linking 261 this decision with integrated pest management (IPM) will further contribute to lowering the 262 ecotoxicological burden on freshwater ecosystems from pesticide use. 263 3.3 Pressure of pesticide impacts by hectare and class (2013-2015) 264 When calculating the potential ecotoxicity impacts on freshwater ecosystems per 1 hectare of 265 crop per year (IScrop_ha) [PAF m3 d ha-1] the cultivation intensity can be addressed, and thus, 266 their interaction of agricultural systems and practices is more apparent. Different ranking and 267 patterns than the presented in section 3.1 are found. Furthermore, the variations in pesticide 268 use (almost 3 orders of magnitude) and impact scores for individual AIs (up to 9 orders of 269 magnitude) are significant. Therefore, in the same year, the two indicators can move in 270 different directions (Figure 3), meaning that pesticide use or application rates is not an 271 adequate indicator of potential impacts (e.g., Gly_agri and rapeseed), since toxicity potentials 272 might be higher for pesticides that are applied in lesser amounts (Fantke and Jolliet, 2016). 273 Figure 3. 274 In terms of cultivation intensity, peas appeared as the crop with the highest pressure by 275 hectare cultivated in the entire period, with the maximum value (6440 PAF m3 d ha-1) in 2015. 11 276 In 2013 rapeseed, spring barley and winter wheat showed IScrop_ha between 64% and 54% 277 lower than peas, in 2014 the difference for the same crops was among 70% and 85% lower 278 and for 2015 all crops showed IScrop_ha 80% lower than peas (see Figure 4). 279 Figure 4. 280 The IScrop_ha for the study varies up to 3.5 orders of magnitude, and the substances 281 cypermethrin (Ins), aclonifen (Hrb), pendimethalin (Hrb) and lambda-cyhalothrin (Ins) 282 present the most significant contribution to IScrop_ha, which his nearly 70% (see Table S9 in 283 SI). The large IScrop_ha for peas-2015, almost double than precedent years, is mainly explained 284 by the bloated use of aclonifen (Hrb). This intensification of herbicide treatments in 2015 285 could be potentially associated with the emergence of weed infestation in peas productions 286 fields. Moreover, the sharp increment on IScrop_ha in part is explained by the dose increment by 287 hectare and the relatively high CF for direct emissions to surface water of aclonifen (SI, Table 288 S5), which is driven by a significant EF (1.3x10+4 PAF m3 kg-1). Furthermore, it is important 289 to note that even if some substances have a high CF; their use could be justified at low doses, 290 because of their agronomic importance and effectiveness of pest or disease control. 291 The contribution by pesticide target class to freshwater IScrop_ha can be observed in Figure 5. 292 Insecticides is the class that contributes in more significant proportion (56%) to impact scores, 293 followed by herbicides (36.4%) and fungicides (7%); plant growth regulators were not 294 included in Figure 5 as their contribution to IScrop_ha and IScrop_DK was lower than 1%. 295 Figure 5. 296 It is well known that pesticide treatments are a highly dynamic activity that varies year by 297 year. Although, it could be more static for herbicides than for the other classes (i.e. 298 insecticides and fungicides) that are more closely correlated with the specific climatic 299 conditions on the area and year of study and thus also the emergence of any specific pest or 300 disease. If these dynamics are to be considered in LCI and LCIA modeling choices, the 301 relevant data (on, e.g., pesticide treatment and crop characteristics) have to be consistently 12 302 reported (Fantke et al., 2016). As mentioned before IScrop_ha did not follow the same trends of 303 pesticide use, likewise, IScrop_ha did not correlate with use by crop (R2=0.0006) or by AI. 304 Similar trends of crop impacts on freshwater ecosystems (unallocated values by hectare and 305 year) are obtained by Nordborg et al. (2014) for the cultivation of maize, rapeseed and winter 306 wheat for biofuel feedstock production; Parajuli et al. 2017 for grass, maize and winter wheat 307 straw for bio-refinery, and Schmidt Rivera et al. 2017 for barley production in Italy and 308 Denmark. The studies above mentioned use PestLCI (version 1 or 2) as inventory model and 309 USEtox 1.01 as characterization method for the impact assessment. Therefore, using a less 310 data demanding a simplified approach could lead to same results for substance prioritization. 311 Despite the similarities in the trends of IScrop_ha, when comparing the results with the absolute 312 values of AI use per 1 ha in a given crop, the IScrop_ha are up to 2.2 orders of magnitude 313 higher; considering the uncertainty range of the characterization method (between 1-2 orders 314 of magnitude) this difference might be moderately significant, and more probably associated 315 with the difference in the LCI and the emission modeling framework. 316 3.4 Effects of modeling choices on ecotoxicity impact assessment 317 3.4.1 Comparing the LCI modeling choices 318 There are very different approaches and assumptions in order to provide emission estimates 319 for quantifying lifecycle emission inventories of pesticides in any LCA study involving 320 agricultural systems. The most simplified approaches are based on generic assumptions 321 regarding varying percentages for pesticide application, the modeling framework of Margni et 322 al., (2002) is used in several agricultural LCA studies. A different approach is the dynamic 323 emission modeling used in PestLCI. This model estimates emissions to three environmental 324 compartments: air, surface water and groundwater. It considers the agricultural field down to 325 1 m depth into the soil and up 100 m into the air as part of the technosphere, thus excluding 326 emissions to soil on-field and off-field (Birkved and Hauschild, 2006; Dijkman et al., 2012). 327 The main differences between the methods are the underlining assumptions, the definition and 13 328 alignment between LCI and LCIA and the data requirements for quantifying pesticides 329 emissions. In this sense, modeling approaches that allowed the inclusion of agricultural soil in 330 the assessment and that involve simplified assumptions for at least application methods were 331 selected in order to test the effects on the impact scores from the emission model choice. 332 The selected methodologies are described in section 2.5, and the results between the three 333 approaches (BS, AS1 and AS2) were compared between the five crops in the 3-year period. 334 The median results for fem in the BS are 2.5 and 1.5 orders of magnitude lower than the 335 emissions for the AS1 and AS2. When modeling fem_air the difference is smaller in comparison 336 with the variations of fem_fw between the three scenarios. Consequently, the variations in the 337 emission fractions lead to further changes in the estimated impact scores. 338 Results for IScrop_ha in [PAF m3 d ha-1] with the BS and the AS1 and AS2 are summarized in 339 Table 2. BS presented the lowest impact results across all crops and years; the highest impact 340 results appear in AS1, whereas, AS2 showed higher impacts than BS but within 1 order of 341 magnitude of difference. High variability in IScrop_ha results within BS and AS2 approaches 342 were observed. 343 Table 2. 344 The Tukey test was conducted to determine statistical differences in the impact assessment of 345 the three modeling approaches tested. The differences in results of BS and AS1 are 346 statistically significant. Meanwhile, the results for AS2 were statistically similar to BS. 347 The delineation between pesticide emission inventory and the impact assessment has shown 348 to have considerable influence on the estimation of ecotoxicity impacts of AI and the impact 349 profiles of crop production (Rosenbaum et al., 2015; van Zelm et al., 2014). However, that 350 alone is not the only explanatory reason for the lower IScrop_ha results. The consideration of 351 intermedia processes, crop growth development and application method allow for a more 352 accurate estimation of the real phenomena, which are also the aspects that usually have the 353 highest influence on LCI and LCIA models (Dijkman et al., 2012; Fantke et al., 2012). 14 354 Furthermore, the consistency showed for trend results of others studies using PestLCI (a more 355 sophisticated emission modeling approach) compared to the BS results are satisfactory (see 356 section 3.3). Keeping in mind that such a model is much more data demanding and since 357 IScrop_ha represent potential impacts rather than actual damages, the substance prioritization 358 with a simplified method as the BS may serve as a first proxy in LCA studies when more 359 detailed data are lacking. 360 3.4.2 Variation from LCIA characterization method version 361 The range of variation for the CF of all AI in the study with USEtox 2.02 was almost 9 orders 362 of magnitude. FF and XF vary by near 2 orders of magnitude, while EF varies up to 7 orders 363 of magnitude indicating substantial differences in pesticide-specific ecotoxicity potential. The 364 variation in the CF for direct emissions to surface water, continental air or agricultural soil 365 was near to 10 orders of magnitude, but CF for direct emissions to continental air and 366 agricultural soil was lower than the CF for direct emissions to freshwater (3 and 2 orders of 367 magnitude, respectively). From which, the importance of modeling the impacts of the dose 368 applied, with a coherent coupling of the LCI to the LCIA model results (i.e. characterized 369 results). 370 Results for IScrop_ha in the base scenario (BS) and USEtox version 1.0 and 2.02 are 371 summarized in Table 3. The more substantial differences in the impact results from both 372 USEtox versions are the AI coverage, with version 1.01 covering fewer AI; thus, IScrop_ha 373 characterized with v 1.0 are lower in most of the cases due to AI coverage, as expected. 374 Furthermore, significant improvements and scientific consensus have been achieved for the 375 new features introduced in the USEtox version 2.02 among which substances and updated 376 substance data and continent-specific landscape parameters contribute to further improving 377 the accuracy in the quantification of CFs. An example of this, are the results for Peas 2013 to 378 2015, were all IA were included in both USEtox versions, and IScrop_ha were within the same 379 order of magnitude but between 3 to 6 times larger. 15 380 Table 3. 381 3.4.3 Sensitivity analysis 382 The results on the evaluation of ecotoxicity impact profiles in Danish crop production 383 demonstrate that modeling freshwater ecotoxicity impacts with the BS and USEtox 2.0 allows 384 to recognize trends of different pesticides treatments and burdens on freshwater ecosystems, 385 thus accounting for interactions between different compartments and a defined clear interface 386 between LCI and LCIA (Figure 1). 387 The variations of the emission fractions to air, surface water and soil were 6 orders of 388 magnitude. Given the input parameter sensitivity analysis presented in the Supplementary 389 material SI, SI-5, the primary sources of uncertainty in the proposed emission modeling 390 framework are identified as i) the application method and the drift fractions, and ii) the 391 allocation for the off-field emission, specifically the water to soil ratio (as shown in figure 6). 392 Although, the uncertainty range associated with pesticide emissions have not yet been 393 quantified and is beyond the scope of the present study. 394 The uncertainty of CFs (USEtox 2.02) due to emissions to air, freshwater and agricultural soil 395 is 176, 18 and 103 GSD2 (Rosenbaum, 2016). The major sources of uncertainty are 396 substances half-lives and ecotoxicity EF (Henderson et al., 2011). Furthermore, in comparison 397 with the FF and XF, the EF shows a substantial variation among the substances covered in 398 this study, explaining a large part of the variations in the CFs for the AI after emissions to 399 freshwater. 400 4 CONCLUSIONS 401 LCI modeling options do affect the ecotoxicological burden on freshwater ecosystems from 402 pesticide use, and directly affects substance prioritization in LCA studies. Furthermore, the 403 updated CF with the continent-specific landscape parameters contributes to a broader 404 assessment. In the case of scenario and sensitivity analysis, the main findings identified 405 application method and allocation for the off-field emission, as the main descriptors for 16 406 modeling emissions of pesticides. The use of the modeling framework presented in this study 407 allows delivering more robust results and accurately evaluation of ecotoxicity impacts. 408 Finally, to provide consumers and policymakers with more reliable information on the 409 environmental performances of agricultural systems, LCA studies need to include all relevant 410 emission outputs; therefore, a final consensus needs to be reached with a specific emission 411 model recommendation. 412 ACKNOWLEDGEMENTS 413 The authors gratefully acknowledge the PhD stage of Nancy Peña at the Agroecology 414 department of Aarhus University financed by the European Commission under the Seventh 415 Framework Program FP7-KBBE.2010.1.2-02, for the Collaborative Project SOLID 416 (Sustainable Organic Low-Input Dairying; grant agreement no. 266367). This work was 417 financially supported by a scholarship granted by the Colombian Government through 418 COLCIENCIAS. Authors would also thank the support from “CERCA Program Generalitat 419 de Catalunya” and Erica Montemayor for the proofreading. 420 APPENDIX A. SUPORTING INFORMATION 421 The following is the supplementary material related to this article. Detailed information of 422 scenarios, physicochemical properties and data on pesticide active ingredients, further 423 annotations on pesticide emission quantification, data and sources for the derivation of new 424 CFs, as well as supporting materials for results and sensitivity analysis included in the study 425 are provided in the Supporting information. 426 17 427 REFERENCES 428 Anton, A., Torrellas, M., Nuñez, M., Sevigne, E., Amores, M.J., Muñoz, P., Montero, J.I., 429 2014. Improvement of Agricultural Life Cycle Assessment Studies through Spatial Di ff 430 erentiation and New Impact Categories: Case Study on Greenhouse Tomato Production. 431 Environ. Sci. Technol. 48, 9454–9462. 432 Balsari, P., Marucco, P., Tamagnone, M., 2007. A test bench for the classification of boom 433 sprayers according to drift risk. Crop Prot. 26, 1482–1489. 434 doi:10.1016/j.cropro.2006.12.012 435 Birkved, M., Hauschild, M.Z., 2006. PestLCI-A model for estimating field emissions of 436 pesticides in agricultural LCA. Ecol. Modell. 198, 433–451. 437 doi:10.1016/j.ecolmodel.2006.05.035 438 Dijkman, T.J., Birkved, M., Hauschild, M.Z., 2012. PestLCI 2.0: A second generation model 439 for estimating emissions of pesticides from arable land in LCA. Int. J. Life Cycle Assess. 440 17, 973–986. doi:10.1007/s11367-012-0439-2 441 European Commission, 2017. Food Safety - Plants-Pesticides [WWW Document]. URL 442 https://ec.europa.eu/food/plant/pesticides_en (accessed 10.20.17). 443 Fantke, P. (Ed), Bijster, M., Guignard, C., Hauschild, M., Huijbregts, M., Jolliet, O., Kounina, 444 A., Magaud, V., Margni, M., McKone, T.., Posthuma, L., Rosenbaum, R.., van de 445 Meent, D., van Zelm, R., 2017. USEtox® 2.0 Documentation (Version 1) [WWW 446 Document]. URL http://usetox.org 447 Fantke, P., Arnot, J.A., Doucette, W.J., 2016. Improving plant bioaccumulation science 448 through consistent reporting of experimental data. J. Environ. Manage. 181, 374–384. 449 doi:10.1016/j.jenvman.2016.06.065 450 Fantke, P., Charles, R., Alencastro, L.F. de, Friedrich, R., Jolliet, O., 2011a. Plant uptake of 451 pesticides and human health: Dynamic modeling of residues in wheat and ingestion 452 intake. Chemosphere 85, 1639–1647. doi:10.1016/j.chemosphere.2011.08.030 453 Fantke, P., Juraske, R., Antón, A., Friedrich, R., Jolliet, O., 2011b. Dynamic multicrop model 454 to characterize impacts of pesticides in food. Environ. Sci. Technol. 45, 8842–8849. 455 doi:10.1021/es201989d 456 Fantke, P., Wieland, P., Juraske, R., Shaddick, G., Itoiz, E.S., Friedrich, R., Jolliet, O., 2012. 457 Parameterization models for pesticide exposure via crop consumption. Environ. Sci. 458 Technol. 46, 12864–12872. doi:10.1021/es301509u 459 Felsot, A.S., Unsworth, J.B., Linders, J.B.H.J., Roberts, G., Rautman, D., Harris, C., Carazo, 460 E., 2010. Agrochemical spray drift; assessment and mitigation—A review*. J. Environ. 461 Sci. Heal. Part B 46, 1–23. doi:10.1080/03601234.2010.515161 462 Gil, E., Gallart, M., Llorens, J., Llop, J., Bayer, T., Carvalho, C., 2014. Spray adjustments 463 based on LWA concept in vineyard. Relationship between canopy and coverage for 464 different application settings. Asp. Appl. Biol. 122, 25–32. 465 Gil, Y., Sinfort, C., 2005. Emission of pesticides to the air during sprayer application: A 466 bibliographic review. Atmos. Environ. 39, 5183–5193. 467 doi:10.1016/j.atmosenv.2005.05.019 468 Goglio, P., Smith, W.N., Grant, B.B., Desjardins, R.L., Gao, X., Hanis, K., Tenuta, M., 469 Campbell, C.A., McConkey, B.G., Nemecek, T., Burgess, P.J., Williams, A.G., 2018. A 470 comparison of methods to quantify greenhouse gas emissions of cropping systems in 471 LCA. J. Clean. Prod. 172, 4010–4017. doi:10.1016/j.jclepro.2017.03.133 472 Gyldenkærne, S., Ravn, H.P., Halling-Sørensen, B., 2000. The effect of dimethoate and 473 cypermethrin on soil-dwelling beetles under semi-field conditions. Chemosphere 41, 474 1045–1057. doi:10.1016/S0045-6535(99)00511-1 475 Gyldenkrne, S., Secher, B.J.M., Nordbo, E., 1999. Ground deposit of pesticides in relation to 476 the cereal canopy density. Pestic. Sci. 55, 1210–1216. doi:10.1002/(SICI)1096- 18 477 9063(199912)55:12<1210::AID-PS76>3.0.CO;2-6 478 Hauschild, M.Z., Huijbregts, M.A.J., 2015. Introducing Life Cycle Impact Assessment, in: 479 Hauschild, M., Huijbregts, M.A.J. (Eds.), The International Journal of Life Cycle 480 Assessment. pp. 66–70. doi:10.1007/BF02978760 481 Henderson, A.D., Hauschild, M.Z., Van De Meent, D., Huijbregts, M.A.J., Larsen, H.F., 482 Margni, M., McKone, T.E., Payet, J., Rosenbaum, R.K., Jolliet, O., 2011. USEtox fate 483 and ecotoxicity factors for comparative assessment of toxic emissions in life cycle 484 analysis: Sensitivity to key chemical properties. Int. J. Life Cycle Assess. 16, 701–709. 485 doi:10.1007/s11367-011-0294-6 486 Itoiz, E.S., Fantke, P., Juraske, R., Kounina, A., Anton, A., 2012. Deposition and residues of 487 azoxystrobin and imidacloprid on greenhouse lettuce with implications for human 488 consumption. Chemosphere 89, 1034–1041. doi:10.1016/j.chemosphere.2012.05.066 489 Jolliet, O., Müller-Wenk, R., Bare, J., Brent, A., Goedkoop, M., Heijungs, R., Itsubo, N., 490 Peña, C., Pennington, D., Potting, J., Rebitzer, G., Stewart, M., Udo de Haes, H., 491 Weidema, B., 2004. The LCIA Midpoint-damage Framework of the UNEP/SETAC Life 492 Cycle Initiative. Int. J. Life Cycle Assess. 9, 394–404. 493 Juraske, R., Antón, A., Castells, F., Huijbregts, M.A.J., 2007. Human intake fractions of 494 pesticides via greenhouse tomato consumption: Comparing model estimates with 495 measurements for Captan. Chemosphere 67, 1102–1107. 496 doi:10.1016/j.chemosphere.2006.11.047 497 Margni, M., Rossier, D., Crettaz, P., Jolliet, O., 2002. Life cycle impact assessment of 498 pesticides on human health and ecosystems. Agric. Ecosyst. Environ. 93, 379–392. 499 doi:10.1016/S0167-8809(01)00336-X 500 Nordborg, M., Cederberg, C., Berndes, G., 2014. Modeling potential freshwater ecotoxicity 501 impacts due to pesticide use in biofuel feedstock production. Environ. Sci. Technol. 48, 502 111379–11388. 503 Notarnicola, B., Sala, S., Anton, A., McLaren, S.J., Saouter, E., Sonesson, U., 2017. The role 504 of life cycle assessment in supporting sustainable agri-food systems: A review of the 505 challenges. J. Clean. Prod. 140, 399–409. doi:10.1016/j.jclepro.2016.06.071 506 Olesen, M.H., Jensen, P.K., 2013. Collection and evaluation of relevant information on crop 507 interception. EFSA Support. Publ. 10, 438E–n/a. doi:10.2903/sp.efsa.2013.EN-438 508 Ørum, J.E., Holtze, M.S., 2017. Bekæmpelsesmiddel- statistik 2015 Behandlingshyppighed 509 og pesticidbelastning , baseret sprøjtejournaldata, Orientering fra Miljøstyrelsen. 510 Ørum, J.E., Hossy, H., 2015. Bekæmpelsesmiddelstatistik 2014. Orientering fra 511 Miljøstyrelsen nr. 13. 512 Ørum, J.E., Samsøe-Petersen, L., 2014. Bekæmpelsesmiddelstatistik 2013. Orientering fra 513 Miljøstyrelsen nr- 6. København. 514 Panagos, P., Borrelli, P., Meusburger, K., Alewell, C., Lugato, E., Montanarella, L., 2015. 515 Estimating the soil erosion cover-management factor at the European scale. Land use 516 policy 48, 38–50. doi:10.1016/j.landusepol.2015.05.021 517 Parajuli, R., Kristensen, I.S., Knudsen, M.T., Mogensen, L., Corona, A., Birkved, M., Peña, 518 N., Graversgaard, M., Dalgaard, T., 2017. Environmental life cycle assessments of 519 producing maize, grass-clover, ryegrass and winter wheat straw for biorefinery. J. Clean. 520 Prod. 142, 3859–3871. doi:10.1016/j.jclepro.2016.10.076 521 Rosenbaum, R., Anton, A., Bengoa, X., Bjorn, A., Brain, R., Bulle, C., Cosme, N., Dijkman, 522 T.J., Fantke, P., Felix, M., Geoghegan, T.S., Gottesburen, B., Hammer, C., Humbert, S., 523 Jolliet, O., Juraske, R., Lewis, F., Maxime, D., Nemecek, T., Payet, J., Rasanen, K., 524 Roux, P., Schau, E.M., Sourisseau, S., van Zelm, R., von Streit, B., Wallman, M., 2015. 525 The Glasgow consensus on the delineation between pesticide emission inventory and 526 impact assessment for LCA. Int. J. Life Cycle Assess. 20, 765–776. doi:10.1007/s11367- 527 015-0871-1 19 528 Rosenbaum, R.K., Bachmann, T.M., Gold, L.S., Huijbregts, M.A.J., Jolliet, O., Juraske, R., 529 Koehler, A., Larsen, H.F., MacLeod, M., Margni, M., McKone, T.E., Payet, J., 530 Schuhmacher, M., Meent, D. van de, Hauschild, M.Z., 2008. USEtox—the UNEP- 531 SETAC toxicity model: recommended characterisation factors for human toxicity and 532 freshwater ecotoxicity in life cycle impact assessment. Int J Life Cycle Assess 13, 532– 533 546. 534 Schmidt Rivera, X.C., Bacenetti, J., Fusi, A., Niero, M., 2017. The influence of fertiliser and 535 pesticide emissions model on life cycle assessment of agricultural products: The case of 536 Danish and Italian barley. Sci. Total Environ. 592, 745–757. 537 doi:10.1016/j.scitotenv.2016.11.183 538 Stockmarr, J., Thomsen, R., 2009. Water supply in Denmark. The Danish action plan for 539 promotion of eco-efficient technologies, Danish Ministry of Environment. 540 van de Zande, J.C., Michielsen, J.M.G.P., Stallinga, H., 2007. Spray drift and off-field 541 evaluation of agrochemicals in the Netherlands. 542 van Zelm, R., Larrey-Lassalle, P., Roux, P., 2014. Bridging the gap between life cycle 543 inventory and impact assessment for toxicological assessments of pesticides used in crop 544 production. Chemosphere 100, 175–181. doi:10.1016/j.chemosphere.2013.11.037 545 Webb, J., Hutchings, N., Amon, B., 2016. EMEP/EEA air pollutant emission inventory 546 Guidebook. 547 Westh, T.B., Hauschild, M.Z., Birkved, M., Jørgensen, M.S., Rosenbaum, R.K., Fantke, P., 548 2015. The USEtox story: a survey of model developer visions and user requirements. Int. 549 J. Life Cycle Assess. 20, 299–310. doi:10.1007/s11367-014-0829-8 550 551 20 552 Tables and table captions 553 Table 1. Comparison of pesticide emission fractions fem calculated by the BS (reference 554 scenario), AS1 (Margni et al. 2002) and AS2 (application method and crop relation). Emission scenarios Average fraction Standard deviation emitted [kg kg-1] on fractions BS fem_air 1.16x10-1 2.03x10-1 fem_fw 1.60x10-3 0 fem_soil.agri 3.75x10-1 3.11x10-1 fem_soil.other 8.70x10-2 2.01x10-2 AS1 fem_air 1.00x10-1 0 fem_fw 5.00x10-2 0 fem_soil 8.50x10-1 0 AS2 fem_air 1.70x10-1 0 fem_fw 1.00x10-2 0 fem_soil 4.50x10-1 0 555 556 21 557 Table 2. Comparison of scenarios to test different emission modeling approaches. Results for 558 potential freshwater ecotoxicity impact scores IScrop_ha in [PAF m3 d ha-1] in the base scenario 559 (BS) and alternative scenarios AS1 and AS2 560 BS AS1 AS2 Crop 2013 2014 2015 2013 2014 2015 2013 2014 2015 Maize 513 92 50 14370 2261 582 3041 475 138 Grass 17 11 13 219 141 169 51 31 37 Winter wheat 2210 434 551 58522 11790 14879 12410 2502 3154 Spring Barley 2086 458 631 64214 12888 18305 13514 2701 3808 Rape 1880 921 1394 56586 17682 33144 12244 4144 7267 Peas 3454 2928 6440 110166 69469 120016 23547 14653 26057 561 562 22 563 Table 3. Comparison of scenarios to test developments of LCIA characterization method. 564 Results for potential freshwater ecotoxicity impact scores IScrop_ha in [PAF m3 d ha-1] in the base 565 scenario (BS) and USEtox version 1.0 and 2.02 566 BS - USEtox 1.0 BS - USEtox 2.02 Crop 2013 2014 2015 2013 2014 2015 Maize 246 63 146 513 92 50 Grass 24 12 14 17 11 13 Winter wheat 1349 445 1223 2210 434 551 Spring Barley 758 267 390 2086 458 631 Rape 776 563 702 1880 921 1394 Peas 1483 1893 6080 3454 2928 6440 Glyphosate Agri-use 24 12 17 14 6 8 567 23 568 Figures and figure captions 569 570 Figure 1. Interface between LCI and LCIA for pesticide application in crop production. 571 572 24 573 574 Figure 2. Freshwater ecotoxicity impact profiles for crop production (2013-2015), impact 575 scores IScrop in [PAF m3 d crop-1]. *Glyphosate (CAS107-83-6) assessed as the total 576 agricultural use in Denmark. 577 25 578 579 580 Figure 3. Comparison between use of pesticide active ingredient (USE_crop) [tones] and 581 potential freshwater ecotoxicity impacts (IScrop_ha) [PAF m3 d ha-1] for 5 analyzed crops 2013 582 and *Glyphosate (CAS107-83-6) assessed as the total agricultural use in Denmark in 583 logarithmic scale. 584 26 585 586 Figure 4. Pressure of pesticide impact scores by hectare of crop cultivated for Danish crop 587 production (2013-2015), impact scores IScrop_ha in [PAF m3 d ha-1]. *Glyphosate (CAS107-83- 588 6) assessed as the total agricultural use in Denmark. 589 27 590 591 Figure 5. Share of freshwater ecotoxicity impact scores IScrop_ha in [%] by pesticide class herbicides (Hrb), 592 insecticides (Ins) and fungicides (Fun) taking as reference per crop IScrop_ha - 2013 as reference year. 593 28 594 595 Figure 6. Sensitivity to model input parameters of BS. Variation for ecotoxicity impact scores (IScrop_ha) in [PAF 596 m3 d ha-1] of Maize in 2013 (Mz-13) 597 29