Received: 9 November 2021 Revised: 21 February 2022 Accepted: 1 March 2022 DOI: 10.1111/tbed.14504 ORIGINAL ARTICLE Targeting the search of African swine fever-infected wild boar carcasses: A tool for early detection Alberto Allepuz1 Mark Hovari2 Marius Masiulis3,4,5 Giovanna Ciaravino1 Daniel Beltrán-Alcrudo2 1 Department of Animal Health and Anatomy, Universitat Autònoma de Barcelona (UAB), Abstract Barcelona, Spain This study analyses the temporal and spatial distribution of found dead African swine 2 Food and Agriculture Organization (FAO), Regional Office for Europe and Central Asia, fever (ASF)-positive wild boar carcasses from 2017 to January 2021 in affected Budapest, Hungary European countries: Bulgaria, Estonia, Germany, Hungary, Latvia, Lithuania, Romania, 3 Emergency Response Division, State Food Poland, Serbia and Slovakia. During this period, a total of 21,785 cases were confirmed and Veterinary Service, Vilnius, Lithuania 4 in 19,071 unique locations. The temporal analysis of aggregated cases per month evi- National Food and Veterinary Risk Assessment Institute, Vilnius, Lithuania denced that most countries located in southern latitudes showed a higher number of 5 Dr. L. Kriauceliunas Small Animal Clinic, cases between January and April, whereas in northern latitudes there was no clear Veterinary Academy, Lithuanian University of temporal pattern. The space–time K-function evidenced a space–time clustering in Health Sciences, Kaunas, Lithuania the ASF-positive wild boar carcasses, which was most prominent within distances of Correspondence 2 km and within 1 week. A Bayesian hierarchical spatial model was calibrated to eval- Alberto Allepuz, Department of Animal Health and Anatomy, Universitat Autònoma de uate the association between the probability of finding ASF-positive wild boar car- Barcelona (UAB), Barcelona, Spain. casses and landscape factors (i.e. the presence of a path and paved road), land use and Email: alberto.allepuz@uab.cat wild boar abundance. Results showed the highest likelihood of finding ASF-positive The views expressed in this information wild boar carcasses in areas of transition between woodland and shrub, green urban product are those of the author and do not areas and mixed forests. The presence of a path and a higher abundance of wild boar necessarily reflect the views or policies of FAO. (Mark Hovari and Daniel Beltrán-Alcrudo) also increased slightly the odds of finding an ASF-positive dead wild boar. In summary, this paper aims to provide recommendations to design a search strategy to find ASF- infected wild boar carcasses, which is a crucial activity in the management of the dis- ease, not just for surveillance purposes (i.e. the early detection of an introduction and the regular monitoring to understand the epidemiology and dynamics), but also for control, namely the disposal of infected carcasses as a virus source. KEYWORDS African swine fever, early detection, epidemiology, surveillance, wild boar 1 INTRODUCTION then spread to other European countries (Sánchez-Vizcaino et al., 2013). The disease was eradicated from Europe in 1995 with the African swine fever (ASF) is a disease that affects all members of the exception of the Italian island of Sardinia, which remained endemic Suidae family. The disease was first identified in 1921 and since then it (Sauter-Louis et al., 2021). In 2007, ASF genotype II was introduced has been circulating mainly in Sub-Saharan Africa. In 1957 and 1960, into Georgia from where it spread gradually westwards until it reached ASF virus of genotype I arrived to Europe (Spain and Portugal) and the European Union (EU) in early 2014, namely Lithuania and Poland This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2022 The Authors. Transboundary and Emerging Diseases published by Wiley-VCH GmbH. Transbound Emerg Dis. 2022;1–11. wileyonlinelibrary.com/journal/tbed 1 2 ALLEPUZ ET AL . (Mačiulskis et al., 2020). Since then, multiple countries in Europe, 2 MATERIAL AND METHODS but also Asia and America have been affected by genotype II with outbreaks in wild boar, domestic pigs or both. The persistence of the 2.1 Study area and origin of data disease in wild boar, the lack of an effective vaccine or treatment, and the high case-fatality rate represent a serious challenge for the global The area of study included the following European countries: Bulgaria, pig sector. At present, biosecurity, movement control and the stamping Estonia, Germany, Hungary, Latvia, Lithuania, Romania, Poland, Ser- out of animals are the only tools to fight the disease in domestic pig bia and Slovakia. Data were provided by the national competent vet- farms (Sánchez-Cordón et al., 2018; EFSA et al., 2021). erinary authorities and covered all ASF-positive wild boar carcasses Finding ASF-positive wild boar carcasses is a crucial activity in the found dead, excluding hunted animals and road kills. To ensure that management of the disease, not just for surveillance purposes, but also data would be comparable in terms of their quality, spatial resolu- for control, namely the disposal of infected carcasses as a source of tion and level of detail, only countries reporting to the EU’s Ani- virus. When it comes to ASF surveillance and early detection in wild mal Diseases Information System (ADIS) were chosen. In fact, all eli- boar, it has been repeatedly proven that sampling and testing found gible countries under such criteria (i.e. reporting to ADIS and with dead wild boar is much more efficient than testing hunted wild boar ASF outbreaks in wild boar) were selected, with the exception of Bel- or road kills, even when the later may intuitively seem more conve- gium and the Czech Republic, where, due to the small areas initially nient. This is explained because the vast majority of wild boar that get affected by ASF, surveillance was particularly intense as compared infected with the ASF virus will die within days, leaving a very short to other countries. In fact, both countries managed to contain the time window of opportunity to detect the virus in a healthy-looking ani- disease and eventually eradicate it and regain freedom. The Italian mal, i.e. whether incubating animals or the few that survive the infec- island of Sardinia was also excluded because of the different genotype tion. Moreover, as soon as wild boars start presenting clinical signs, involved. they tend to hide and rest, which largely prevents them from being Data included all eligible events in the target countries. Many coun- hunted. This has very important implications when trying to find the tries revised their data collection and only since 2017 they started to disease in wild boar, both in already infected countries that try to provide precise geo-coordinates for each found dead wild boar. There- understand the epidemiology and evolution of the disease, but partic- fore, the data analysed for countries already affected at the time starts ularly in newly infected countries, where early detection is critical for in 2017. having a chance at successful control. The active search of carcasses in countries or regions at high risk of ASF, for example, across the border from infected areas, is the most efficient way to early detect the intro- 2.2 Covariates included in the model duction of the disease into ASF-free wild boar populations. Wild boar that have died of ASF represent a continuous source of OpenStreetMap data from the area of study was downloaded infection for other animals, as the virus might remain infectious in the from Geofabrik (https://www.geofabrik.de/). Paths correspond to carcass for an extended period of time, depending on the environmen- those categorized as paths or bridleways in OpenStreetMap. Paved tal conditions. It has been reported that a frozen carcass can maintain roads included secondary, tertiary and unclassified roads, as well infectious ASF virus for several months enabling the virus to overwin- as those with an agricultural use. Water included both water lines ter and to initiate a new outbreak when the defrosted carcass is vis- (i.e. rivers and streams) and water bodies (i.e. reservoirs or wet- ited the following spring by a susceptible wild boar or free-ranging pig. lands). Land use was extracted from the Corine Land Cover map, Therefore, the safe removal of carcasses from the environment and created by the European Environment Agency under the Euro- their disposal is an important measure to avoid ASF spread by reduc- pean Union’s Earth Observation Programme, named Copernicus ing the local maintenance of the virus (FAO, 2019). The EU developed [©European Union, Copernicus Land Monitoring Service, 2018, Euro- an ASF strategic approach to prevent and control the spread of the dis- pean Environment Agency (EEA)]. This map has a resolution of ease and eventually to eradicate ASF from the EU. One of the compo- 100 × 100 m grid and was last updated in 2018. For our study, we nents of this strategy is finding, testing and disposal of ASF-infected used the 44 classes included on level 3 that corresponded to five carcasses (Anonymous, 2020). Optimizing the search (and disposal) of main land use groups: artificial surfaces, agriculture, forests and semi- ASF-infected carcasses should contribute to the eradication of the dis- natural areas, wetlands and water bodies. Wild boar abundance was ease. retrieved from the ENETWILD consortium (2020) at a resolution of However, there are few studies that have attempted to identify in 2000 × 2000 m grid. which areas it is more likely to find ASF-infected carcasses. Similarly, there are currently no instructions or recommendations on where to look for the dead wild boar. The objective of this study was to describe 2.3 Data management the temporal and space–time distribution of ASF-positive wild boar carcasses reported from 2017 to January 2021 in Europe and to iden- Data were projected into ETRS89-extended/LAEA Europe. For each tify those landscape factors that increase the likelihood of finding these point in which a wild boar had been found dead, the distance to the carcasses. This will in turn enable optimization ASF surveillance efforts nearest path, paved road, water line or water body, was calculated by and strategies. creating a SpatiaLite database for each of these layers. A structured ALLEPUZ ET AL . 3 query language (SQL) query was created among them to extract dis- TA B L E 1 Distribution by country and year of the data used in the tances. For spatial modelling purposes, a buffer of 2000 m was created study, i.e. African swine fever-positive wild boar found dead in target countries between January 2017 and January 2021 around each location in which a wild boar was found dead. The area covered by the buffer was divided into a grid of 500 × 500 m. The loca- Country 2017 2018 2019 2020 2021 Total tions of ASF-positive wild boar carcasses were superimposed on this Bulgaria 6 68 980 15 1069 grid and cells that intersected with those points were classified as pos- Estonia 185 12 1 6 204 itive and otherwise, negative. Similarly, paths and paved roads were Germany 372 130 502 superimposed on this grid to identify if they were present in each of Hungary 155 2038 4444 57 6694 the grid cells. Water lines and water bodies from OpenStreetMap were Latvia 774 269 31 83 6 1163 not superimposed on this grid, as this land use was already present in the Corine Land Cover map data (i.e. level 3 classes: water courses and Lithuania 1065 1035 197 57 2354 water bodies). The zonal statistic plugin was used to obtain the maxi- Poland 738 2415 2470 2692 8315 mum and most frequent values of wild boar abundance and land use, Romania 125 385 682 50 1242 respectively, in that 500 × 500 m grid. All these analyses were done Serbia 27 84 111 with Quantum GIS 3.18 (QGIS Development Team, 2021). Slovakia 21 110 131 Total 2762 4017 5211 9453 342 21,785 2.4 Temporal analysis and space–time analysis The forecast library (Hyndman & Khandakar, 2008; Hyndman et al., ential (SPDE; Lindgren et al., 2011) and calculated from a matrix of 2021) in R was used to describe the temporal trend of the number of Euclidean distances between grid centroids using Delaunay triangu- ASF-positive wild boar carcasses between January 2017 and January lation (Cameletti et al., 2013; Simpson et al., 2011). This model was 2021. To construct the time-series dataset, we used the date when car- solved by using the R-INLA package (Schrödle & Held, 2011). To assess casses were confirmed to be infected by ASF by the national reference the association of the variables included in the model with the prob- laboratories. Dates were aggregated by month. The number of cases ability of finding ASF dead wild boar in a grid, 95% credible intervals per month along the different years, and the number of cases found (CR) were obtained from the exponential of the mean, 2.5% and 97.5% each month in the whole study period were described. percentiles of the posterior probability distribution of the regression The space–time K-function, as described by Diggle et al. (2015) was coefficients. We considered a variable to be associated if the proba- used to describe the excess of risk that could be attributed to an ASF- bility was over 95%, that is, if the 95% CR was greater or lower than positive wild boar carcass as a function of distance and time. In case of 1. If greater, the variable increased such probability, and if lower, it no space–time clustering (i.e. when cases occur independently in space decreased it. and time) the K-function at each distance and each temporal increase To validate the ability of the model to discriminate between grids in is equal to the product of the K-function in space and the one in time. which it was more likely to find wild boars dead due to ASF, the sta- The multiplication of the difference between the observed K-function tus (i.e. the classification of a grid as positive or negative) was removed in space and time by the product of the space and time K-functions is from 30% of randomly selected grid cells. In those cells, the status called the proportional increase in risk or excess of risk due to the pres- was predicted by the model and compared with their original value by ence of space–time interaction. Using the splancs R package, we calcu- means of a receiver operating characteristic (ROC) curve constructed lated this value over a space–time grid of 5 km times 2 months using using the pROC package (Robin et al., 2011) in R. The area under that intervals of 500 m and 1 week, respectively. To illustrate any elevated curve (AUC) is related to the performance of the model. AUC values disease risk attributable to space–time interaction this value was plot- greater than .8 and between .7 and .8 are indicative of good and mod- ted as a surface over a space–time grid. erate discriminate capacities respectively. 2.5 Spatial model 3 RESULTS A Bernoulli distribution was used to model the probability of find- 3.1 Descriptive, temporal and space–time ing ASF-positive wild boar carcass in each grid cell. The logit trans- analysis results formation was used to link such probability with specific explanatory variables. A backward and forward stepwise procedure based on the Table 1 and Figure 1 show the number and location of ASF-positive wild Akaike information criterion (AIC) was used to select the best model. boars found dead in target countries between January 2017 and Jan- Once the best model was selected, it was extended by adding ran- uary 2021. dom spatially structured and unstructured components. The spatially During this period, a total of 21,758 cases of ASF-positive in wild structured random effect was defined by a stochastic partial differ- boar carcasses were confirmed in 19,071 unique locations (i.e. in some 4 ALLEPUZ ET AL . F I G U R E 1 Location of the data used in the study, i.e. African swine fever-positive wild boar found dead in target countries between January 2017 and January 2021 cases, several animals were reported in the same coordinates). The Twenty five per cent of carcasses were found within 755, 161 and year with more detected cases was 2020. Poland, followed by Hungary, 298 m of a path, paved road or water line/body, respectively. Tables 3 detected the most positives. and 4 show the most frequent land use in each grid cell together with The number of ASF-positive wild boar carcasses per month between the presence of a path or paved road and the abundance of wild boar 2017 and 2021 in the target countries included in this study can be according to the presence or not of ASF-positive wild boar carcasses in found in Figure 2. them. The temporal pattern of each country was heterogeneous. Despite Green urban areas, transitional woodland-shrub areas, mixed and this apparent heterogeneity, when plotting cumulative cases per broad-leaved forests and sport/leisure areas were the land uses with month in the whole period of study (Figure 3), there was a pattern a higher proportion of positive grid cells. Wild boars were only slightly whereby countries in southern latitudes (i.e. Bulgaria, Hungary, Roma- more abundant in those grids in which dead ASF-positive wild boar had nia and Serbia) showed a higher number of cases from January to April, been found (versus those without). with the exception of Romania, where there was also a high number of Odds ratio and their 95% credible intervals (CI) for each of the risk cases in November–December. On the other hand, there was no clear factors from the hierarchical Bayesian model together with the random temporal pattern for countries in northern latitudes (i.e. Estonia, Latvia, effects are presented in Table 5. Lithuania or Poland), with the exception of Poland, where the numbers Among land use categories, considering non-irrigated arable lands of cases were slightly higher in winter. as the baseline, the model showed the highest likelihood of finding Figure 4 shows the plot of the proportional increase in risk from ASF-positive wild boar carcasses in areas of transition between wood- the space–time K-function. This plot evidences the existence of space– land and shrub, green urban areas and mixed forests, with odds ratios time clustering in the data, which translates in an increase in risk, which around 3 times higher. The presence of a path and a higher abundance is most prominent within 2000 m and within 1 week. of wild boar also increased slightly the odds of finding an ASF-positive dead wild boar. On the other hand, the presence of paved roads was not retained in the model as it did not influence the likelihood of find- 3.2 Spatial model results ing ASF-positive carcasses. For model validation, we randomly selected 30% of cells in which Table 2 shows the distance between ASF-positive wild boar carcasses we removed the ASF status and estimated the area under the ROC to the nearest path, paved road, water line or water body. (receiver operating characteristic) curve (i.e. AUC). Results showed ALLEPUZ ET AL . 5 FIGURE 2 African swine fever-positive wild boar found dead in target countries per month (January 2017 to January 2021) F I G U R E 3 Number of African swine fever-positive wild boars found dead in target countries per month. Germany and Slovakia are not shown due to the low temporal frame for which cases have been found 6 ALLEPUZ ET AL . F I G U R E 4 Proportional increase in risk due to space–time clustering with the K-function. The elevated surface illustrates the excess in risk for finding African swine fever-positive dead wild boar TA B L E 2 Distance (in meters) from each African swine active search of dead wild boar and the subsequent disposal of infected fever-positive wild boar carcass to the nearest path, paved road, water carcasses. Results from our study might contribute to increase the effi- line or water body ciency of the search of infected carcasses by allowing to target those Min Q1 Median Q3 Max areas in which it is more likely to find ASF-positive dead wild boars. Path 0 755 2136 4376 22,634 Results from this study showed that some landscape factors (and wild Water * 0 298 887 1759 17,374 boar abundance to a lesser degree) increased the likelihood of find- ing ASF-positive wild boar carcasses and could therefore be used to Paved 0 161 394 800 8049 map those areas that should be prioritized to search for them. In the Min: minimum; Q1: first quartile; Q3: third quartile; Max: maximum. Czech Republic, Cukor et al. (2020) also attempted to identify those *The shortest distance to a water line or water body. factors linked to the location in which ASF-positive wild boar carcasses were found. In their study, they determined that most ASF-infected 77% (IC95%: 69.4%–84.2%), indicative of a model with a moderate carcasses were found in forest and especially in young forest areas. capacity to discriminate between ASF-positive and ASF-negative grid These results were explained by the fact that wild boars may choose cells (Figure 5). such areas to die, since they offer silence, cover and lower densities of other animal species. Similarly, our model also showed higher odds of finding ASF-infected carcasses in certain forests (i.e. mixed, broad- 4 DISCUSSION leaved or coniferous) and areas of transition between woodlands and shrub, which consist of young plants. Moreover, studies conducted in Early detection is of paramount importance to contain any outbreak. It Poland have also identified woodlands as areas with a risk of ASF occur- applies to all transboundary diseases and to both livestock and wildlife. rence (Podgórski et al., 2020). Therefore, these types of land use should Two key control measures recommended for ASF in wild boar are the be targeted in the search of ASF-infected carcasses. ALLEPUZ ET AL . 7 TA B L E 3 Most frequent land use and presence of a paved road or path in grid cells where African swine fever-positive wild boar carcasses were found (i.e. positive), versus neighbouring cells in which they were not found (i.e. negative) Variable Category Pos Neg Proportion (%) Land use Green urban areas 16 128 11.1 Transitional woodland-shrub 1107 14,224 7.2 Mixed forest 2197 30,485 6.7 Broad-leaved forest 3677 54,526 6.3 Sport and leisure facilities 37 566 6.1 Water courses 52 1012 4.9 Coniferous forest 1790 35,995 4.7 Inland marshes 82 1827 4.3 Land principally occupied by agriculture, with 626 14,608 4.1 significant areas of natural vegetation Vineyards 66 1567 4 Discontinuous urban fabric 450 11,063 3.9 Pastures 1133 30,659 3.6 Mineral extraction sites 11 312 3.4 Industrial or commercial units 45 1295 3.4 Fruit trees and berry plantations 101 2991 3.3 Complex cultivation patterns 325 11,752 2.7 Natural grasslands 62 2262 2.7 Non-irrigated arable land 2868 114,139 2.5 a Freq_low 15 842 2.3 Water bodies 79 3817 2 Rice fields 9 463 1.9 Peat bogs 12 851 1.4 Paved Yes 6322 144,437 4.19 No 8439 190,946 4.23 Path Yes 1726 21,207 7.53 No 13,035 314,176 3.98 a Land uses with less than 200 observations in the dataset were grouped in this category. TA B L E 4 Descriptive statistics on wild boar abundance in grid cells with and without African swine fever-positive wild boar carcasses N Mean SD Min Q1 Median Q3 Max Positive (with) 14,841 54.8 24.7 3.7 36.2 47.2 67.8 181.3 Negative (without) 339,186 49.3 22.9 2.7 33.8 41.9 58.8 181.3 Searching near water courses or water bodies has also been rec- ing an ASF-infected carcass near water might be higher during the hot- ommended, as infected wild boar, when developing clinical signs such ter periods of the year, when animals need more drinking water and as fever and dehydration, search for humid environments and water cooler resting places, often associated to water sources. Perhaps also (Podgórski et al., 2020). Indeed, Cukor et al. (2020) described that the abundance of water (streams and rivers) is important and animals in around 60% of ASF-infected carcasses were found up to 100 m from more arid areas may tend to remain closer to the water. Consequently, water sources. However, we did not find such a clear association, as the recommendation for searching near water sources might depend only 25% of the ASF-positive carcasses were found within 298 m from on the period of the year and on the land uses present in the target area. water sources. The association with distance to water might be influ- Other landscapes such as green urban areas (OR of 3.0) and sport enced by other factors, such as temperature. The probability of find- and leisure facilities (OR of 1.5), or the presence of a path in the grid 8 ALLEPUZ ET AL . TA B L E 5 Fixed and random effects included in the hierarchical Bayesian model, odds ratio (OR), standard deviations (SD) and their 95% credible intervals (CI) Credible Credible OR SD 2.5% 97.5% Land usea Transitional woodland-shrub 3.1 0.0492 2.8 3.4 Green urban areas 3.0 0.3670 1.4 6 Mixed forest 2.9 0.0395 2.7 3.2 Broad-leaved forest 2.5 0.0377 2.3 2.7 Inland marshes 2.4 0.1449 1.8 3.2 Coniferous forest 2 0.0444 1.8 2.2 Land principally occupied by 2.0 0.0567 1.8 2.2 agriculture, with significant areas of natural areas Water courses 1.7 0.1877 1.2 2.5 Natural grasslands 1.6 0.1604 1.1 2.1 Pastures 1.5 0.0456 1.4 1.7 Rice fields 1.5 0.3857 0.7 3.1 Sport and leisure facilities 1.5 0.2503 0.9 2.4 Mineral extraction sites 1.5 0.3725 0.7 2.9 Industrial or commercial units 1.4 0.1865 1 2 Vineyards 1.4 0.1718 1 1.9 Complex cultivation patterns 1.2 0.0746 1.1 1.4 Discontinuous urban fabric 1.2 0.0702 1.1 1.4 Fruit trees and berry plantation 1.0 0.1441 0.8 1.4 Water bodies 0.9 0.1466 0.7 1.3 b Freq_low 0.8 0.3072 0.4 1.4 Peat bogs 0.8 0.3890 0.3 1.5 Pathc Path presence 1.1 0.0414 1 1.2 WBq2e 1.1 0.0377 1.1 1.2 d e Wild boars abundance WBq3 1.3 0.0390 1.2 1.4 e WBq4 1.3 0.0440 1.2 1.4 Coefficients SD Random effects Spatial-structured random 720.9 0.956 effect Non-spatial structured random 0.0106 0.007 effect a Reference category was ‘non-irrigated arable land’. b Land uses with less than 200 observations in the dataset were grouped in this category. c Reference category was ‘path absence’. d Reference category was the first quartile. e WBq2, WBq3 and WBq4 refer to the second, third and fourth quartile of the wild boar abundance distribution. (OR of 1.1), were also highlighted by the model as areas where it is more scapes should probably not be targeted to search for ASF-infected car- likely to find ASF-infected carcasses. Probably these results respond casses, since they are already indirectly found by people passing by, but to the higher human activity, which implies that any wild boar carcass rather increase the incentives of the public to report found dead wild will most likely be quickly found, rather than a predilection of wild boar. Citizen science and mobile application easing such public report- boar for those areas. An association between human population den- ing can assist detection efforts. Accordingly, a participatory workshop sity and the number of reports of ASF-positive carcasses has indeed with different experts in the field (Jori et al., 2020) highlighted that been reported elsewhere (Lim et al., 2021). Therefore, and despite ASF- good communication and transparent information directed to the pub- positive wild boar carcasses might be found in these areas, these land- lic was a powerful tool for improving passive surveillance against ASF. ALLEPUZ ET AL . 9 F I G U R E 5 Receiver operating characteristic (ROC) curve to test the ability of the model to discriminate between positive and negative African swine fever grid cells in the 30% of randomly selected cells in which their status against ASF was removed. AUC: area under the curve The size of the infected area is another important factor that influ- and resting places or water sources. The reasoning behind the 1 week ences the search of wild boar carcasses once ASFV has been confirmed temporal pattern might be explained by the fact that wild boar are in an area. The size of the infected area may vary greatly, which makes social animals who live in groups. Most times, several members of the the targeted search for dead wild boar very demanding in terms of time same group will become infected by ASF at about the same time. This and human resources. The minimum size of the infected area should implies they will all be dying clustered at approximately the same time be defined based on the geographical distribution of the disease, the (i.e. 1 week) and around the same area (2 km radius). Combining the wild boar population in the area and the presence of major natural or search in this spatial and temporal frame (focusing on the landscapes artificial obstacles to the movements of wild boars. Therefore, it can identified by the model) with other methods such as the use of hunting vary from a few square kilometres to even an entire country. Since dogs (Jori et al., 2020) or drones might also maximize the probability the search is time and resource consuming, it is critical to define the of carcass detection. area and period of time in which such search should be performed, to As highlighted by the human density factor mentioned above, it is optimize the most likely time and location. The space–time analysis important to stress that this model does not always point to the areas evidenced that after the first detection of an ASF-infected wild boar with more ASF-positive wild boar carcasses, but rather at the places in an area, the probability of finding ASF-positive carcasses was higher where such carcasses are more easily found, for example, close to up to 2 km and over the following week. This combination will offer paths, in areas often visited by people, and where vegetation is lower the best effort-success ratio. Indeed, many ASF-affected countries and/or thinner, thus allowing for a better visibility. While efforts were have guidelines for the search of newly infected areas that recommend done to utilize only data of high quality, by targeting countries that searching for at least 30 days and focusing on the wild boar feeding all collect and report data with precise geo coordinates and the same 10 ALLEPUZ ET AL . reporting standards/requirements (AIDS), there are a number of biases allows to generate search maps or strategies for wild boar carcasses, that are difficult or often impossible to avoid. Perhaps the most impor- which focus on the areas with a higher likelihood to find an ASF-positive tant bias relates to the nature of wild boar as a wildlife species, that wild boar carcass. Rather than covering whole territories, both the gen- is, the fact that they live freely, in unknown numbers and densities eration of maps and the subsequent search efforts should be based on and without movement restrictions. This implies that finding their car- risk assessment approach. Results also helps emergency preparedness casses when they die of ASF or any other diseases is a challenging pro- to make better simulation exercises for ASF in wild boar, by aiding to cess that translates in a high (but variable) degree of under-reporting, better determine where dead wild boar might be found. which will depend on the search effort (whether active or passive), but For free countries, the mapped areas should be those at a higher also on the type of land (e.g. how accessible it is or how thick is the veg- risk for ASF introduction, for example, border areas or specific hunt- etation). These will vary greatly between and even within countries. ing grounds. For infected countries, the rapid finding and subsequent Efforts were taken when selecting the targeted countries, by avoid- disposal of ASF-positive wild boar carcasses is one of the key recom- ing countries with very intensive search effort like the Czech Repub- mended measures to reduce the viral load in the ecosystem, which will lic or Belgium. The limited fenced infected area in these two countries eventually translate in less spread of the disease and even its control allowed a clear shot at eradication (as it indeed happened), which trans- and eradication. lated in an active search of carcasses that probably lead to the detec- Easier than generating risk maps is the standardization of search tion of the majority of existing ASF-positive carcasses in the area. On parameters. Already described within the paper, just providing the key the other hand, countries with less resources and no economic incen- risk factors to hunting ground managers is a simple, yet powerful tool to tives for the reporting of carcasses were also excluded from the study focus search efforts where there are more chances of success, that is, (i.e. most countries outside the EU, except for Serbia), as the under- finding an ASF-positive wild boar carcass. The most important factors reporting is considered to be more severe than in study countries. identified by the model are (in order of importance): Wild boar management is another important factor, for example, the type of hunting (driven or not), the ban of supplementary feeding, the 1. Transitional woodland-shrub level of hunting biosecurity, the awareness and cooperation of hunters, 2. Mixed forest the magnitude of (economic) incentives to report, etc. All these differ 3. Broad-leaved forest between and even within countries and affect the way ASF spreads 4. Inland marshes in wild boar population and the chances of finding wild boar. Finally, 5. Coniferous forest ecological and climatic factors will also affect the wild boar popula- 6. Land principally occupied by agriculture tions, not just in their abundance (which was accounted through the 7. Water courses use of wild boar abundance variable), but also their movement pat- 8. Natural grasslands terns, behaviour and interactions. Factors related to the disease also 9. Pastures need to be accounted for. Although all countries are affected by the 10. Rice fields same genotype (II), there are various strains circulating (Nurmoja et al., 11. Vineyards 2017), and different levels of endemicity, which translate in different 12. Areas with the highest wild boar density clinical presentations, lethality and other epidemiological parameters. Also, the ASF status in domestic pigs (which may allow the disease to When trying to find carcasses around an already confirmed ASF- jump back and forward between domestic and wild populations) and infected wild boar, active searches should take place within 1 week other epidemiological factors cannot be excluded as potential biases. after the event and in a 2 km radius, focusing in those areas in which is more likely to find them. 5 CONCLUSION ACKNOWLEDGEMENTS The authors acknowledge the veterinary services from the different Finding ASF-positive wild boar carcasses is a crucial activity in the man- countries for providing outbreak data. Acknowledgements also to the agement of the disease, not just for surveillance purposes (i.e. the early Food and Agriculture Organization of the United Nations (FAO), which detection of an introduction and the regular monitoring to understand financed the study through an Emergency Regional Technical Coop- the epidemiology and dynamics), but also for control, namely the dis- eration Programme (TCP) project (TCP/RER/3704) on “African Swine posal of infected carcasses as a source of virus. This study, based on Fever emergency preparedness in the Balkans”. [Correction added on thousands of observations, can be translated into very practical appli- 31 March 2022, after first online publication: The preceding sentence cations in the early detection of ASF in wild boar populations. This is was corrected.] The views expressed in this information product are key to have a chance at the control and eradication of the disease in wild those of the authors and do not necessarily reflect the views or policies boar populations, which is otherwise extremely difficult and resource- of the Food and Agriculture Organization. consuming. Results pointed that efforts to find (and remove) additional ASF-positive wild boar carcasses after a confirmed case should be ETHICS STATEMENT devoted up to 2 km and over the following week. In addition, the model No animals were used in this project. ALLEPUZ ET AL . 11 DATA AVAILABILITY STATEMENT populations. Preventive Veterinary Medicine, 185, 105178. https://doi.org/ The data that support the findings of this study are available on request 10.1016/j.prevetmed.2020.105178 Lim, J. -. S., Vergne, T., Pak, S.-I. l, & Kim, E. (2021). Modelling the spatial from the corresponding author. The data are not publicly available due distribution of ASF-positive wild boar carcasses in South Korea using to privacy or ethical restrictions. 2019–2020 National Surveillance Data. Animals (Basel), 11(11), 1208. https://doi.org/10.3390/ani11051208 ORCID Lindgren, F., Rue, H., & Lindstrom, J. (2011). An explicit link between Gaus- sian fields and Gaussian Markov random fields: The SPDE approach (with Alberto Allepuz https://orcid.org/0000-0003-3518-1991 discussion). Journal of the Royal Statistical Society. Series B, 73, 423–498. Marius Masiulis https://orcid.org/0000-0003-2779-5803 https://doi.org/10.1111/j.1467-9868.2011.00777 Giovanna Ciaravino https://orcid.org/0000-0002-5796-8093 Mačiulskis, P., Masiulis, M., Pridotkas, G., Buitkuvienė, J., Jurgelevičius, V., Jacevičienė, I., Zagrabskaitė, R., Zani, L., & Pilevičienė, S. (2020). The African swine fever epidemic in wild boar (Sus scrofa) in Lithuania (2014– REFERENCES 2018). Veterinary Sciences, 7, 15. https://doi.org/10.3390/vetsci7010015 Anonymous. (2020). Strategic approach to the management of African Nurmoja, I., Petrov, A., Breidenstein, C., Zani, L., Forth, J. H., Beer, M., Swine Fever for the EU, Working Document SANTE/7113/2015-Rev.2 Kristian, M., Viltrop, A., & Blome, S. (2017). Biological characterization of (Accessed: 29.04.2020) https://ec.europa.eu/food/system/files/2020- African swine fever virus genotype II strains from north-eastern Estonia 04/ad_control-measures_asf_wrk-doc-sante-2015-7113.pdf in European wild boar. Transboundary and Emerging Diseases, 64(6), 2034– Cameletti, M., Lindgren, F., Simpson, D., & Rue, H. (2013). Spatio- 2041. temporal modeling of particulate matter concentration through the Podgórski, T., Borowik, T., Łyjak, M., & Wot’zniakowski, G. (2020). Spatial SPDE approach. Advances in Statistical Analysis, 97, 109–131. https://doi. epidemiology of African swine fever: Host, landscape and anthropogenic org/10.1007/s10182-012-0196-3 drivers of disease occurrence in wild boar. Preventive Veterinary Medicine, Cukor, J., Linda, R., Václavek, P., Šatrán, P., Mahlerová, K., Vacek, Z., Kunca, 177, 104691. T., & Havránek, F. (2020). Wild boar deathbed choice in relation to ASF: QGIS.org (2021). QGIS Geographic Information System. QGIS Association. Are there any differences between positive and negative carcasses? Pre- http://www.qgis.org ventive Veterinary Medicine, 177, 104943. Robin, X., Turck, N., Tiberti, N., Lisacek, F., Sanchez, J. C., & Müller, M. (2011). Diggle, P. J., Chetwynd, A. G., Häggkvist, R., & Morris, S. E. (1995). Second- pROC: An open-source package for R and S+ to analyze and compare order analysis of space-time clustering. Statistical Methods in Medical ROC curves. BMC Bioinformatics, 12, 77. https://doi.org/10.1186/1471- Research, 4, 124–136. 2105-12-77 Nielsen, S. S., Alvarez, J., Bicout, D. J., Calistri, P., Canali, E., Drewe, J. A., Sánchez-Cordón, P. J., Montoya, M., Reis, A. L., & Dixon, L. K. (2018). African Garin-Bastuji, B., Gonzales Rojas, J. L., Herskin, M., Miranda Chueca, M. swine fever: A re-emerging viral disease threatening the global pig indus- Á., Michel, V., Padalino, B., Pasquali, P., Roberts, H. C., Sihvonen, L. H., try. The Veterinary Journal, 233, 41–48. https://doi.org/10.1016/j.tvjl. Spoolder, H., Stahl, K., Velarde, A., Viltrop, A., . . . Gortázar Schmidt, C. 2017.12.025 (2021). Scientific Opinion on the African swine fever and outdoor farm- Sánchez-Cordón, P. J., Montoya, M., Reis, A. L., & Dixon, L. K. (2013). African ing of pigs. EFSA Journal, 19, 6639. https://doi.org/10.2903/j.efsa.2021. swine fever (ASF): Five years around Europe. Veterinary Microbiology, 6639 233, 41–48. https://doi.org/10.1016/j.tvjl.2017.12.025 Acevedo, P., Croft, S., Smith, G. C., Blanco-Aguiar, O. A., Fernandez-Lopez, J., Sauter-Louis, C., Conraths, F. J., Probst, C., Blohm, U., Schulz, K., Sehl, J., Scandura, M., Apollonio, M., Ferroglio, E., Keuling, O., Sange, M., Zanet, Fischer, M., Forth, J. H., Zani, L., Depner, K., Mettenleiter, T. C., Beer, M., & S., Brivio, F., Podgórski, T., Petrović, K., Soriguer, R., & Vicente, J. (2020). Blome, S. (2021). African Swine Fever in Wild Boar in Europe—A Review. Validation and inference of high-resolution information (downscaling) of Viruses, 13, 1717. https://doi.org/10.3390/v13091717 ENETwild abundance model for wild boar. EFSA Supporting Publication Schrödle, B., & Held, L. (2011). Spatio-temporal disease mapping using INLA. 17, 23. https://doi.org/10.2903/sp.efsa.2020.EN-1787 Environmetrics, 22, 725–734. https://doi.org/10.1002/env.1065 FAO, OIE and EC. (2019). African swine fever in wild boar ecology and biose- Simpson, D., Illian, J., Lindgren, F., Sørbye, S. H., & Rue, H. (2011). Going curity. FAO Animal Production and Health Manual No. 22. Rome. off grid: Computationally efficient inference for log-Gaussian Cox pro- Hyndman, R., Athanasopoulos, G., Bergmeir, C., Caceres, G., Chhay, L., cesses. NTNU Technical report, 10. O’Hara-Wild, M., Petropoulos, F., Razbash, S., Wang, E., & Yasmeen, F. (2021). Forecast: Forecasting functions for time series and linear mod- els. R package version 8.15. https://pkg.robjhyndman.com/forecast/ Hyndman, R. J., & Khandakar, Y. (2008). Automatic time series forecasting: How to cite this article: Allepuz, A., Hovari, M., Masiulis, M., The forecast package for R. Journal of Statistical Software, 26(3), 1–22. Ciaravino, G., & Beltrán-Alcrudo, D. (2022). Targeting the https://www.jstatsoft.org/article/view/v027i03 search of African swine fever-infected wild boar carcasses: A Jori, F., Chenais, E., Boinas, F., Busauskas, P., Dholllander, S., Fleischmann, L., Olsevskis, E., Rijks, J. M., Schulz, K., Thulke, H. H., Viltrop, A., & Stahl, K. tool for early detection. Transboundary and Emerging Diseases, (2020). Application of the World Café method to discuss the efficiency of 1–11. https://doi.org/10.1111/tbed.14504 African swine fever control strategies in European wild boar (Sus scrofa)