Predicting the Potential Distribution of Apple Canker Pathogen (Valsa mali) in China under Climate Change

Apple valsa canker (AVC), caused by Valsa mali, is a serious wood disease of apple trees. The pathogen decays the barks and branches of trees and ruins entire orchards under severe conditions. However, studies have rarely focused on the suitable habitat of the pathogen, especially on a relatively large scale. In this study, we applied the maximum entropy model (MaxEnt 3.4.1, Princeton, NJ, USA) to predict the distribution of V. mali using climate factors, topographic factors, and soil factors under current and future climate scenarios. We measured the area of suitable habitat, change ratio of the suitable habitat area, increase and decrease maps under climate change, direction and distance of range shifts from the present to the end of the 21st century, and the contribution of environmental variables. The results showed that the area of suitable habitat is currently 183.46 × 104 km2 in China, among which 27.54% is moderately suitable habitat (MSH) and 13.13% is highly suitable habitat (HSH). Compared with current distribution, the area of MSH and HSH increases in future and the change ratio are positive. The Shared Socioeconomic Pathways (SSPs) 3–70 is considered the optimum climate scenario for V. mali. The suitability of V. mali increased mainly in Northwest, North, and Northeast China. V. mali will shift to the northwest with climate change. The shift distance optimistically increased from the SSP1–26 to the SSP5–85, with the biggest shift distance of 758.44 km in the 2090s under the SSP5–85 scenario. Minimum temperature of the coldest month (bio6) was the most critical climate factor affecting the distribution of the pathogen, and topographic factors played a more important role than soil factors. This study demonstrates that the potential distribution of V. mali is vitally affected by climate change and provides a method for large–scale research on the distribution of pathogens.


Introduction
With improving of living standards and the development of cultivation, apples have become one of the most popular fruits worldwide. During the past two decades of the 21st century, China has become the biggest apple-producing country [1], both in terms of farm area (averaging 1.95 × 10 6 ha) and yield (averaging 3.09 × 10 7 tons) (http://www.stats.gov.cn/). However, the yield per unit area is lower than half of France's and Italy's (http://www.fao.org/statistics/en/) and apple valsa canker (AVC) is considered one of the major factors causing the low quality and low yield per unit area [2,3]. Apple valsa canker (AVC), caused by the necrotrophic ascomycete Valsa mali [4], is a serious apple wood threat [49] of V. mali, a pathogen that may result in a potentially greater amount of damage under climate change conditions, is of vital importance for developing new procedures to control AVC. In this article, we aim to (1) predict the potential distribution of V. mali in China in the 21st century and (2) determine the direction of range shifts under climate change.

Species Occurrence Data and Environmental Variables
We gathered 93 occurrence points with GPS by our field survey, 12 occurrence points from 55 published studies from the Web of Science (WOS), and 168 occurrence points from 71 published reports from the China National Knowledge Infrastructure (CNKI). Removing widely marked points and duplicated sites, we gathered 211 available points. The points from published studies were positioned by GoogleEarth-pro if their GPS data were not recorded. To lessen the spatial autocorrelation effect, we finally obtained 158 rarefying data points ( Figure 1 and Table A1 of Appendix A) with the criterion that there would be no more than one occurrence point in a grid cell (the distance between occurrence data > 5 km). We initially download 19 bioclimatic variables (bio1-bio19, 2.5 arc-min resolution, 5 km) from WorldClim v2.1 [54] (https://www.worldclim.org/data/index.html), Digital Elevation Model (DEM,~1 km) data and three soil texture data (~1 km) from the Data Center for Resource and Environment Sciences (RESDC, http://www.resdc.cn/), and one soil type data from the Harmonized World Soil Database (HWSD, http://www.fao.org/soils--portal/en) (Tables A2 and A3). The data from WorldClim include current conditions  and future climate data (2030s: 2021-2040, 2050s: 2041-2060, 2070s: 2061-2080, and 2090s: 2081-2100). Variable selection is an effective way to deal with model overfitting [55,56]. We selected bio1 and bio12 as the climate factors based on the principal component analysis (PCA , Table A2) [37] of all bioclimatic variables (bio1-bio19), and then added bio6, bio11, and bio15 according to the incidence trends and physiology requirements [4,11] of V. mali. In our study, the topographic data (including elevation, aspect, slope, and curvature) in the grid were calculated from DEM data by ArcGIS 10.2 (ESRI, Redlands, CA, USA). We also chose soil data, including soil texture (sand, silt, and clay) and soil type (FAOSoil: including sand%, silt%, clay%, pH 2 O, N%, CaCO 3 %, etc., of topsoil and subsoil), as the vital environmental data in our study. To avoid multicollinearity [36,57], we conducted a correlation analysis of selected bioclimatic, topographic, and soil data and removed those variables with a correlation coefficient greater than 0.8 (Table A3) [55]. We resampled all environmental data into 2.5 arc-min spatial resolution for the accuracy of the prediction and finally retained 11 environmental variables (Table 1) to model the current distribution of V. mali.
We chose the Beijing Climate Center Climate System Model (BCC-CSM2-MR) from nine global climate models (GCMs) as the bioclimatic variables in our study because it has been widely used in Asia [48,61], especially in China [36]. Considering the stability of the terrain [55], we assumed these relative variables would not change in this century. To provide plausible descriptions of future climate change, we used the same topography variables and soil data along with the similar bioclimatic variables of different periods to predict the future distribution of V. mali.

Species Distribution Model Evaluation
In this study, we used MaxEnt v3.4.1 [25], a machine learning algorithm program [62,63], to estimate the potential geographic distribution [47,64] and range shifts [36] of V. mali because of its wide use [24,27,65] and higher accuracy [66] compared with other SDMs [67] in predicting the distribution of species. Occurrence data (without absence data [15]) and environmental data [63,67,68] are required to run MaxEnt. The area under the receiver operating characteristic curve (AUC) is usually used to estimate the accuracy of prediction [69], and Jackknife test results were used to evaluate the relative importance of environmental variables [63]. In general, AUC values range from 0 to 1 [25]: ≤0.5 indicating bad prediction, >0.5 and ≤0.7 indicating fair prediction, >0.7 and ≤0.9 indicating good prediction, and >0.9 and ≤1 indicating excellent prediction [69,70].
In addition to the default settings, for sub-setting the present data, 25% of the occurrence data were used as test data and 75% as training data [45,71]. To predict the potential distribution of V. mali, 17 single models (i.e., one for the current distribution and 4 SSPs × 4 periods for the future) were built with parameters of 1000 maximum iterations and repeated 5 times for each process to reduce uncertainty [37]. Hence, we finally built a total of 165 model results (80 of them were replications for corresponding future predictions) in this step.
To delineate the presence and absence maps of V. mali, we averaged the probability values and created binary maps (suitable or not suitable) with "maximum test sensitivity plus specificity (Max sss, close to 0.05)" [29,35,72]. The presence/absence maps show the spatial distribution of V. mali. We then reclassified the distribution map into low suitable habitat (LSH), moderately suitable habitat (MSH), and highly suitable habitat (HSH) with break values of 0.2, 0.5, and 1.0, respectively.

Spatial and Statistical Analysis
The centroids, calculated by the averaging coordinates (latitude and longitude) of all suitable pixels in distribution maps [73], have been widely used in assessing the distributional changes in binary SDMs [45]. To illustrate the suitable habitat area changes of V. mali, (1) we measured the core range shifts by calculating centroids of suitability in different periods (present, 2030s, 2050s, 2070s and 2090s) and SSPs (SSP1-26, SSP2-45, SSP3-70, and SSP5-85), and demonstrated the direction of range shifts of different periods (i.e., 2030s-present, 2050s-2030s, 2070s-2050s, and 2090s-2070s) in different climate scenarios; (2) we calculated the area change ratio of suitable habitats with the equation Cr = (Aa − Ap)/Ap, where Cr represents the change ratio, and Aa and Ap represent the distribution area (1 × 10 4 km 2 ) of the following period and previous period [36]; (3) we distinguished the suitable area without suitability change from the unsuitable area and tested Moran's index to measure the spatial autocorrelation of potential distribution with the help of ArcGIS 10.2. To fit the location of the study area, all geographic maps in our study were projected onto the Asia North Albers Equal Area Conic Projection.

The Projected Distribution of V. mali
We obtained the distribution maps of V. mali under current and future climate scenarios (Figures 2 and 3) and calculated the area of suitable habitats ( Figure 4). In the current condition in China, with a 0.932 value of Moran's index, the areas of unsuitable habitat (USH), low suitable habitat (LSH), moderately suitable habitat (MSH), and highly suitable habitat (HSH) were founded to be 772.70 × 10 4 km 2 , 108.85 × 10 4 km 2 , 50.52 × 10 4 km 2 , and 24.09 × 10 4 km 2 , respectively. The HSH area of SSP1-26 showed a U-shape from the 2030s to the 2090s, while those of SSP2-45 and SSP5-85 were an inverted U-shape ( Figure 4). It is worth noting that, compared with the current distribution, almost all of the change ratios of LSH and MSH under four climate scenarios are positive in the 21st century, except for −7.92% of the present-2030s period under SSP3-70 ( Figure 5a). Correspondingly, the change ratio of HSH was more complicated: the SSP3-70 climate scenario decreased with the average ratios of −21.83% (the 2030s), −14.77% (the 2050s), and −13.17% (the 2070s) and increased with 33.57% (the 2090s), while the SSP2-45 climate scenario increased with the average ratios of 24.76% (the 2030s), 28.32% (the 2050s), and 6.42% (the 2070s) and decreased with −16.78% (the 2090s) (Figures 5a  and 6a). Under the SSP1-26 scenario, the area of LSH, MSH, and HSH increased in the present-2030s period and the 2070s-2090s period, and decreased in the middle of the century (2030s-2050s and 2050s-2070s periods, Figures 5b and 6b). The SSP3-70 scenario decreased the area of LSH, MSH, and HSH during the first 20 years (present-2030s) and increased the area from the 2030s to the 2090s except for the LSH of the 2050s-2070s period (Figure 5b). In the current condition, the potential suitable area (LSH, MSH, and HSH) of V. mali is distributed in all five traditional apple-producing regions, which is more serious in the Bohai Bay area, the Old Yellow River Basin area, and the Loess Plateau area of Northwest China (Figure 2). With global warming, the suitability of potential distribution shows a major increase in the Xinjiang area, North China, and Northeast China and a decrease in the Bohai Bay area, the Old Yellow River Basin area, and east of the Loess Plateau area (Figure 6a).
The centroid of range shifts showed that V. mali, compared with the current centroid, would shift to the northwest direction ( Figure 7). Under the SSP1-26, SSP2-45, SSP3-70 and SSP5-85 climate scenarios, the shift distance in the 2030s would be 199.18, 281.61, 238.10, and 269.55 km, respectively, while in the 2090s, the shift distance would be 388.97, 602.10, 572.55, and 758.44 km, respectively. From lower to higher SSP scenarios, the shift distance generally increased. The shift distance also increased under four climate scenarios from the present to the 2030s, the 2050s, the 2070s, and the 2090s.

Contributions of Environmental Factors to the Distribution of V. mali
The MaxEnt model provided satisfactory projections with an average AUC value of 0.952 ± 0.009 (test, Figure A1a) and 0.965 ± 0.013 (training). Bioclimatic data, topographic data, and soil data all played a vital role in determining the distribution of V. mali. Among environmental variables, bioclimate variables contributed the most (a total of 75.54%), with bio6 and bio12 providing 56.73% and 10.27% of the contribution (Figure 8). Topographic variables ranked as the second most important factors with the average contribution of elevation and slope being 8.44% and 7.23%, respectively, while soil variables ranked third, with sand and soil contributing 4.88% and 0.87% to the MaxEnt model, respectively. In general, the cumulative value of the top five important variables (bio6, bio12, elevation, slope, and sand) was 87.54%. Interestingly, bio11 had a small contribution to MaxEnt, with an average contribution of only 4.11%, but it varied from 0.01% to 19.87%.

Changes on the Suitable Habitat Area of V. mali
Our results successfully modeled the current distribution of V. mali and indicated that the suitable habitats will be optimistically affected by the anticipated climate, topographic, and soil factors in 2030s, 2050s, 2070s, and 2090s, except several specific periods of HSH may be negatively affected (Figure 5a). At the species scale, enough experiments have been performed to integrate the effects of climate change and other environmental variables on species range shifts [37]. Species that are sensitive to changes in temperature and precipitation will probably vary greatly in area and range shifts [43], and those with a narrow geographic range might be too vulnerable to extinction [36]. In our case, both SSP2-45 and SSP3-70 had reasonable expansion trends, and SSP1-26 and SSP5-85 had bigger expansion areas across the four climate scenarios. However, SSP3-70 might be the more plausible scenario for V. mali, considering both the increase area of HSH and MSH and their change trends. In addition, we observed a significant increase of suitability in Northwest, North, and Northeast China, both in terms of suitable classes and potential area (Figures 4 and 6a). The corresponding increase of precipitation and temperature may be the internal dominate driver because the change of topographic factors, soil type, and soil texture factors are relatively slow in a short historical period [55]. With the continued increase of greenhouse gas emissions, a higher temperature increase is assumed from the lower emission scenario (SSP1-26) to the higher emission scenario (SSP5-85).
Previous studies have reported that species shifts to poleward in latitude and upward to higher elevation are the most frequent types of range shifts under the pressure of global warming [43,44,74]. In Southern Europe, studies of several forest pathogens showed the same shift trends due to climate change, resulting in the northward and upward spread [49]. However, the range shifts of V. mali showed an unexpectedly different direction, toward the northwest and lower elevations (Figure 7). Because the optimum temperature for V. mali to extend hyphae is approximately 20 • C [4], we speculate that the hydrothermal condition changes in suitable area mainly resulted in the unexpected shift type of the pathogen; affected by the southeast monsoon, the precipitation change caused by climate change may be the dominant factor. Global warming may lead to an average of 31% precipitation increase [40], which would have a significant impact on plant distribution patterns. In East Asia, although the southeast monsoon plays an important role relative to the open Northwest Pacific [75], the effects of the maritime climate are invariably weak in inland areas of Western China. In the arid and semiarid land of Western China, a slight increase in precipitation is enough to cause a dramatic change in the distribution of species. Considering the strong environment adaptability [11] of the conidial of V. mali, increasing precipitation may lead to a further expansion in the wild and in orchards than predicted. In addition, based on the range shift comparison of the present and future (Figures 6 and 7), V. mali is expected to shift northwestward in all of the warming scenarios, which is similar to the direction of the southeast monsoon. To the best of our knowledge, the northwestward shift of V. mali is consistent with the shift of Rhododendrons in Southwest of China, but is the opposite in terms of vertical shift [36]. Because of high mountains located in both Southwest and Northwest China, we speculate the different response strategies of V. mali and Rhododendrons to temperature and precipitation may be the main reason for their opposite shifts in vertical direction. Precipitation (bio12) was found to play a more important role in the distribution of V. mali than temperature (bio1) in our study (Figure 8), and considering the complex topography of the suitable area, we believe that the downward range shift is also acceptable. Thus, the unexpected shift of species may be not so unexpected when the conditions of the local climate is taken into account [74].

The Role of Environmental Variables in Effecting the Distribution of V. mali and Corresponding Strategies for Preventing AVC
Any amount of caution is understandable when interpreting model-based results [45], for there are too many environmental variables and parameters that must be considered. The effects of environmental variables on the distribution of species may be disproportionate [76]. Our results indicate that climate factors have a stronger regulating effect on the distribution of V. mali than topographic and soil factors, and the temperature variables (bio1, bio6, and bio11) have a more significant effect than precipitation variables (bio12 and bio15; Figures 8 and A1c). We assumed two possible reasons, and the first one focuses on the influence scale of environmental variable type. The predicted potential distribution of V. mali is spread over a wide area in Northern and Northwestern China. However, the change of topographic factors (i.e., aspect, elevation, slope, etc.) and soil factors (i.e., soil type and soil texture) generally affect the distribution of species in narrow spatial scale, while climate factors have been proved to be more influential on a broader scale [36,41]. Compared with climate factors, topographic and soil variables tend to affect the host of the pathogen. The second reason focuses on the infection strategy. Previous studies have confirmed that the primary infection paths of V. mali are pruning wounds (human factor) and rainwater (precipitation), and 80% of AVC is caused by pruning wounds, while wetness (precipitation) has not exhibited an essential effect [2,4]. The hyphal extension temperature of V. mali ranges from 5 to 35 • C (optimum at 20 • C), leading to the infection of the pathogen usually appearing in March, April, and May, declining from June and barely occurring in November and thereafter [4]. This is consistent with our results which show that temperature plays a more prominent role in affecting the distribution of V. mali than precipitation, and this explains why the minimum temperature of coldest month (bio6) contributed most to MaxEnt. Furthermore, at the wound, we speculate that water and temperature are probably complementary to V. mali infection, because in early spring, the alternations of ice melting and frost cracking often lead to worse AVC [77], especially of spur and bark wounds.
At the scale of the population level, the adaptive potential of the pathogen may reflect the effects of climate change [78], causing greater potential damage to trees [49]. Crabapple, the rootstocks of apple trees, has been identified along with apple seeds and seedlings as a potential inoculum source [79], which makes the control procedures of AVC more difficult. Therefore, to reduce the spread of V. mali, it is of vital importance to limit latent sources of infection: improving garden management methods to cultivate vigorous trees [3], coating new pruning wounds with protection film [4], and using special pruning tools for special gardens, as well as strengthening the management of diseased trees. In addition, the selection of biological agents [15], application of novel rapid assays [79], and nutritional dynamic monitoring of garden trees, especially leaf nutrient status of potassium content [3], are of great help to effectively decrease the occurrence and development of AVC, because suitable crop load management has been proved to increase the high-quality fruit percentage and plant resistance to AVC [80].

Limitations of SDMs in Projecting the Distribution of Species
Our study obtained the potential distribution maps of V. mali in several periods under four climate scenarios and successfully established an example for predicting the suitable habitats of a plant pathogen. However, the accuracy of our simulation is also affected by several limitations. Firstly, the accuracy of input data affects the accuracy of model results. Both the accuracy of species occurrence data and the selection of the current climate template increase the uncertainties of prediction [36]. Secondly, the selection of variables requires more rigorous methods. To avoid collinearity, we screened the environmental factors involved in model operation, but those factors that were filtered out by the screening process may contain potentially important information affecting the distribution of species. At the same time, the screened environmental variables have unstable performance in different simulations (i.e., the range of bio6 and bio11 in Figure 8). Thirdly, the model may limit the accuracy of prediction. The MaxEnt model assumes that species exist in habitats with similar environmental conditions, such as temperature and precipitation [67,81]. However, in the biological world, the presence of species is affected by more complex factors [71], which are often difficult to fully and accurately consider together. Unlike the simulated distributions, species may fail to disperse due to geographic barriers and real conditions of target regions [25,28]. In addition, some unquantifiable factors may affect the accuracy of simulation results due to their nonquantifiable characteristics. Especially, for some host-specific species (i.e., some plant pathogens), it is even difficult to distinguish the climate suitability between the species and its host [31]. We only considered abiotic factors [42] in our study, and biotic factors (i.e., human-caused land use change [36]) may also lead to changes of suitable habitat. Optimizing the algorithms in order to utilize more important information and to make the model more intelligent will further benefit the accuracy of simulation.

Conclusions
This study provided the potential distribution simulation of the pathogen of apple canker, V. mali, in the current and other periods of the 21st century under four CMIP6 climate scenarios: SSP1-26, SSP2-45, SSP3-70, and SSP5-85. In the current period, the predicted area of suitable habitat (LSH, MSH, and HSH) is 183.46 × 10 4 km 2 . The suitable area will have different degrees of increase under all climate scenarios. In all scenarios, the range of potential suitable habitat shifted to the northwest, with the shift distance increasing from the 2030s to the 2090s, compared with the range of the present. The increase of suitability was found to be mainly in Xinjiang, and North and Northeast China, while the decrease was mainly in the Bohai Bay area, the Old Yellow River Basin area, and east of the Loess Plateau area. Till the end of this century, SSP1-26 has the biggest increase in total habitat area, and SSP3-70 was considered the optimum scenario for V. mali. We conclude that the dominant factors effecting the distribution of V. mali are temperature and corresponding precipitation, and temperature is considered to play a more important role. Acknowledgments: We acknowledge the groups of ArcGIS and MaxEnt for their contribution in making this simulation possible. WorldClim, RESDC, and FAO provided the raster data. We thank Hanqi Liu for his advice and assistance in data processing. Thanks to Runzhi Mao for his spiritual support. Especially, the authors would also like to thank the anonymous reviewers for their very important suggestions which have been particularly helpful in improving this paper.