1. Introduction
Kiwifruit is one of the most popular fruits worldwide due to its special flavor and abundant nutrient contents, especially vitamin C, minerals, and dietary fiber [
1,
2]. Kiwifruit is a perennial cash crop, which can provide continuous considerable economic benefit to growers [
3]. At present, kiwifruit occupies an important position in the international fruit business. In leading producing countries, such as China, New Zealand, and Italy, kiwifruit is the main source of income for local families, making it a vital parameter in their economy [
4]. The kiwifruit industry in Sichuan has experienced unprecedented development since 2008, with its current output ranking second in China [
5]. Among the cultivated varieties, red-fleshed kiwifruit, especially “Hongyang”, has been widely cultivated, with a planting area reaching 50,000 hectares until 2018. Kiwifruit brown spot caused by
Corynespora cassiicola is a fungal airborne disease [
6,
7], which is the second most significant disease, followed by kiwifruit bacterial canker disease, in Sichuan [
8]. The disease spreads rapidly, and the incidence can reach 90% to 100% during the harvest stage. In some orchards, functional leaves are almost entirely lost, leading to the premature sprouting of autumn shoots, and the fruits lose water severely, shrivel, and soften, losing commodity value and causing huge yield loss [
8]. With the continuous cultivation of highly susceptible varieties, kiwifruit brown spot has become an escalating concern and a significant limiting factor for industry development. Therefore, understanding the occurrence regularity and distribution regionalization of the disease is of utmost importance.
Previous studies on the epidemic of plant diseases and pests have been limited by technical barriers and other factors, restricting their spatial development to relatively small scales, typically within a confined infected field. However, recent advancements in geographic information system (GIS) technology have enabled the analysis of large-scale spatial data, which makes it possible to regionalize the occurrence and development regularity of diseases and pests in large-scale regions, even on a global scale. Zou et al. [
9] regionalized the oversummer range of wheat powdery mildew using the GIS ordinary kriging method combined with the digital elevation model (DEM) of China. Kistner-Thomas et al. [
10] assessed the relationship between grasshopper density survey data and 72 biologically relevant GIS-based environmental variables and developed a regression model to predict the mean density of an adult grasshopper from 2012 to 2016. At present, in the research on the suitable regionalization of diseases and pests, correlation analysis, weight coefficient distribution, and assignment of each impact factor are first carried out with mathematical methods; then the comprehensive factors are graded by the index classification method, the data are rasterized by GIS, the value of the vacant area is inserted, and the regionalization results are finally determined [
11]. The accuracy of the model primarily relies on the key factors. Accurately understanding occurrence regularity could lay a good foundation for the subsequent distribution regionalization. Peng et al. [
12] successfully identified the occurrence and epidemic regularity of wheat stripe rust in Nanchong, achieving 100%, 98%, and 95% accuracy rates for short-term, medium-term, and long-term predictions, respectively. Chen et al. [
13] used trajectory analysis and an effective accumulated temperature model to simulate the migration process and development progress of
Spodoptera frugiperda, clarifying its migration path and occurrence regionalization in China.
The maximum entropy theory was the most objective criterion for selecting the statistical characteristics of random variables. Based on this theory, the maximum entropy (MaxEnt) model is a quantitative analysis tool, which has been widely applied in plant protection due to its stable operation, simplicity, rapid calculation, and high accuracy [
14,
15,
16]. Zhang et al. [
17] collected geographic location information on
Prunus salicina and one of the brown rot pathogenic species (
Monilinia fructicola), applying the MaxEnt model to simulate their potential suitable distribution in China. Wang et al. [
18,
19] utilized the MaxEnt model based on distribution information and environmental variables to investigate the suitability of kiwifruit bacterial canker disease in Sichuan and predict its potential distribution under climate change. Evaluating the accuracy of the model is an essential step, accomplished by employing different evaluation methods and standards with effective evaluation indexes. Among these, the area under the receiver operating characteristic (ROC) curve (AUC) has been widely used. Wei et al. [
20] used the AUC to evaluate the accuracy of the MaxEnt model in predicting the potential distribution of maize chlorotic mottle virus (MCMV) under historical and future climatic conditions. Cho et al. [
21] evaluated the accuracy of spatial (regression kriging) and nonspatial (MaxEnt) models to simulate the distribution of two invasive plant species (
Ambrosia artemisiifolia and
Ambrosia trifida) by AUC.
Kiwifruit brown spot primarily occurs during high temperature and high humidity seasons, and initially appears in late June and lasts until October. On the other hand, kiwifruit bacterial canker disease is a low-temperature disease, occurring from November to May the following year. Both diseases significantly threaten the healthy development of the kiwifruit industry and substantially impact the variety layout. Meanwhile, their phenological periods and the physiological and biochemical characteristics are completely different. Therefore, the development of this study can provide more scientific and rational guidance for the variety layout of kiwifruit in Sichuan. At present, the occurrence regularity of kiwifruit brown spot in Sichuan has not been clearly elucidated, resulting in blind prevention and control and serious pesticide abuse in production, which is not conducive to the healthy development of the industry. The international scientific community has dedicated many efforts to enhancing resilience and sustainability in agriculture, with a particular emphasis on reducing pesticides [
22], including in kiwifruit cultivation [
23,
24]. Disease prediction is a prerequisite for disease management and plays an important role in integrated pest management [
25]. The database established based on long-term monitoring is the foundation for an accurate disease prediction [
26]. Early preparation for prevention can not only improve the effectiveness and benefits, but also reduce unnecessary costs and environmental pollution caused by pesticide abuse [
27]. Moreover, studies on kiwifruit brown spot are still in the initial stage and mainly focus on pathogen identification, population diversity, resistance of cultivars, and disease control method [
3,
6,
7,
28]. However, a suitable regionalization based on climate conditions remains unknown, warranting further study.
In this study, the epidemic of kiwifruit brown spot in the main producing region was monitored over the long term, and the potential distribution of this disease in Sichuan was employed to predict it using the MaxEnt model. The objectives of this study were to (1) clarify the disease occurrence regularity to provide rational guidance for the variety layout; (2) to evaluate the key environmental variables affecting the distribution of this disease; (3) to determine the suitability of this disease in Sichuan, aiming to provide a theoretical basis for the prediction, precise prevention, and control of the disease.
2. Materials and Methods
2.1. Sources of Software and Map
2.2. Determination of the Occurrence of Kiwifruit Brown Spot
During 2012–2022, the occurrence and distribution of kiwifruit brown spot in the main production regions in Sichuan were continuously assessed. Detailed geographical information (longitude, latitude, and altitude), cultivar, and disease occurrence were recorded. The assessment involved six prefecture-level administrative regions, including Chengdu, Ya’an, Guangyuan, Deyang, Mianyang, and Meishan, and was conducted at the end of the disease logistic phase, namely, mid- to late September.
2.3. Measurement of Disease Severity
Each experimental site consisted of five plots, with three trees per plot, totaling fifteen trees. Each tree was examined from five directions (east, west, south, north, and middle), with two branches per direction. From each branch, five leaves were selected, starting from the base upwards, resulting in a total of fifty leaves per tree. The standard of classification for disease severity is presented in
Table 1. The disease index (DI) was calculated according to the following formula [
3]. The disease level was divided into three categories [
8]: high (DI ≥ 66.67), moderate (33.33 ≤ DI < 66.67), and low (0 < DI < 33.33).
2.4. Correlation Analysis
The Pearson correlation between variety and disease index was determined by SPSS 21.0. The altitude and disease index of the “Hongyang” variety were extracted for correlation analysis.
2.5. Acquisition and Processing of Distribution Information
Kiwifruit brown spot disease primarily occurred in red-fleshed kiwifruit, widely cultivated in Sichuan. It has also been reported in Guangxi, Hubei, and Chongqing, exhibiting strong regional characteristics. Therefore, the other distribution information was mainly obtained by searching published papers [
3,
7,
28], inquiring the kiwifruit disease and pest prediction and forecast reports issued by local plant protection stations and agricultural technology and meteorological cooperative service reports, and consulting local plant protection staff. Through the above procedure, 108 occurrence records were evaluated. By incorporating the 122 available experimental sites collected during the initial stages, a total of 230 distribution points were obtained.
The distribution points with specific information were directly applied, while those lacking latitude and longitude were inquired in the global geographic information integrated database GeoName and the Baidu coordinate picking system to extract the coordinate information. The buffering area analysis function of ArcGIS was used to calculate the distance between the grid center and distribution points. Only the closest record to the center was retained within the same grid. The above distribution records in the order of species, longitude, and latitude were imported into Excel, with a positive north latitude and east longitude and a negative south latitude and west longitude. After the screening process, 225 valid points were retained for constructing the models (
Figure 1), and the species distribution information was transferred to the CSV file required by MaxEnt.
2.6. Acquisition and Processing of Environmental Variables
All environmental data were obtained for free from the WorldClim database (
http://www.worldclim.org/, accessed on 28 August 2023), which provides interpolated raster data based on global meteorological record information. The above data were in TIFF format with a spatial resolution of 2.5 arc-minutes, including 67 environmental variables, among which were 19 bioclimatic factors, monthly mean precipitation, monthly mean temperature, maximum and minimum temperature, and other bioclimatic indices (
Table 2). The environmental variables were extracted from the administrative zoning map of Sichuan as the base map. The TIFF format was transferred to the ASCII format required by MaxEnt using the format conversion function of ArcGIS. Initially, 67 environmental variables were extracted to build an initial model. The contribution rates and importance of environmental variables were determined by the jackknife method, and those rates of less than 1% were eliminated. The attribute values of environmental variables at 225 distribution points were extracted using the extraction analysis tool of ArcGIS. Pearson correlations between environmental variables were calculated using SPSS 21.0. Variables with a strong correlation were removed, and the relationship between kiwifruit brown spot and meteorological factors was considered to finally screen out variables.
2.7. Construction and Evaluation of MaxEnt Model
The distribution information of kiwifruit brown spot and environmental variables were imported into MaxEnt, and the climate response curve was created to analyze the relationship between each environmental variable and the distribution probability. The prediction map was drawn, and the importance of environmental variables was measured using the jackknife method. The random test percentage was set to 25% and repeated 10 times. The default values of the model were selected for other parameters, and the output path was set for modeling. In this study, AUC was used to evaluate the accuracy of the model simulation. The AUC value ranges from 0.5 to 1, with values closer to 1 indicating a stronger correlation between environmental variables and species distribution and a higher model accuracy. The evaluation criterion is 0.5 ≤ AUC < 0.6, fail; 0.6 ≤ AUC < 0.7, poor; 0.7 ≤ AUC < 0.8, general; 0.8 ≤ AUC < 0.9, good; AUC ≥ 0.9, excellent.
2.8. Geographic Division of Suitability
The ASCII format files output by MaxEnt were transferred to raster format files using the format conversion function of ArcGIS. The potential distribution map of kiwifruit brown spot was extracted from the administrative zone map of Sichuan as the base map, and then the spatial analysis tool was used for reclassification. According to previous studies [
29,
30], the suitable area was divided into four categories, displaying them in different colors: highly suitable area (
p > 0.66, red), moderately suitable area (0.33 <
p ≤ 0.66, orange), lowly suitable area (0.05 <
p ≤ 0.33, yellow), and unsuitable area (
p ≤ 0.05, white). The distribution area of each region and district (county) was calculated using the statistical analysis function of ArcGIS.
2.9. Field Evaluation of the Model
A field test of species distribution was conducted as the most direct and reliable method for model validation. In order to further verify the accuracy of the simulation results, additional 45 actual occurrence records from supplementary determination were introduced for the field test. The actual distribution points were mapped onto the reclassified map, and the corresponding suitable levels were extracted by ArcGIS, and then compared with the actual level of disease occurrence. Both are equally regarded as accurate.
4. Discussion
In this study, continuous monitoring of the epidemic of kiwifruit brown spot in Sichuan revealed that the occurrence of the disease was more severe in the red-fleshed varieties and less severe in the yellow-fleshed and green-fleshed varieties. These findings were consistent with the resistance evaluation of Huang et al. [
28] in the kiwifruit germplasm materials to kiwifruit brown spot. The study also highlighted the close relationship between variety layout and the occurrence and epidemic of kiwifruit brown spot. With the continuous expansion of the highly susceptible variety cultivation, kiwifruit brown spot has become the most significant fungal disease in the Sichuan-producing region. However, the impact of variety simplification becomes increasingly evident, and has led to an increased risk of disaster caused by the disease. At present, the prevention of the disease in production is still mainly dependent on chemical control, which is not conducive to the healthy and sustainable development of the kiwifruit industry. The breeding and utilization of resistant varieties are recommended as the most economical and effective measures to control the disease. Therefore, in highly suitable areas, the introduction of resistant varieties, such as Ruiyu and Jinyan, should be recommended. Additionally, the “technical regulations for comprehensive control of kiwifruit brown spot” [
31] formulated by Sichuan Agricultural University should be adopted for the scientific and efficient control of the disease. Consistent with the research conclusion of Cui [
32], the incidence and disease index were significantly negatively correlated with altitude; as the altitude increased, they decreased gradually. However, further research is needed to determine the altitude boundary for disease occurrence. Since WorldClim datasets are generated by integrating and interpolating the basic data of meteorological stations at different altitudes around the world, altitude information has been implied [
18]. Therefore, altitude was not selected as an environmental variable for the MaxEnt prediction model in this study.
Currently, there is limited information on the occurrence and epidemic regularity of kiwifruit brown spot in large-scale areas. Our group has conducted some work on the epidemic dynamics in the early stage, mainly focusing on field disease monitoring and data collection, providing a certain foundation for the construction of prediction models. In this study, the MaxEnt model was employed to simulate and predict the potential distribution of kiwifruit brown spot in Sichuan, and suitability regionalization was conducted using ArcGIS. The suitable areas of each region were calculated, the distribution of central planting areas was analyzed, and the accuracy of the model was verified through field tests. The results indicated that the highly suitable areas of kiwifruit brown spot were mainly located in the Chengdu and Ya’an regions, while the unsuitable areas were mainly distributed in the Ganzi, Aba, and Liangshan regions. The Panzhihua region was entirely classified as an unsuitable area. According to the central planting areas of red-fleshed kiwifruit, Pujiang County and Mingshan District were entirely classified as highly suitable distribution areas. Both kiwifruit brown spot and bacterial canker disease had an adverse impact on the healthy development of the industry. Therefore, in the layout of varieties, both diseases should be comprehensively considered. Ma. [
33] demonstrated that the potential severe and suitable areas for kiwifruit bacterial canker disease were mainly distributed along the Longmen Mountains from south to north, connected with the Qinba Mountains, and concentrated in the Ya’an, Chengdu, and Guangyuan regions, which had a high contact ratio with the highly suitable areas for kiwifruit brown spot in this study. Wang et al. [
18] indicated that the unsuitable areas of kiwifruit bacterial canker disease in Sichuan were mainly located in the Ganzi, Liangshan, and Panzhihua regions, consistent with the findings of kiwifruit brown spot in this study. The occurrence regionalization of kiwifruit brown spot in Sichuan has been clarified, which not only provided a scientific and effective theoretical basis for the formulation of a prevention and control strategy, but also played an important role in the layout and development of the kiwifruit industry in combination with previous studies on kiwifruit bacterial canker disease.
A disease epidemic is a result of the interaction between host plants and pathogens under the influence of environmental conditions. Environmental conditions mainly include meteorological factors, soil conditions, tillage system, and cultivation measures, with meteorological factors playing an extremely important role. Many researchers have utilized modeling to select key meteorological factors for disease prediction. For instance, Chen et al. [
34] found that the amount of rainfall in spring significantly influenced the date of the grapevine downy mildew symptom onset. Chaulagain et al. [
35] used correlation analysis and stepwise logistic regression to identify afternoon humid thermal ratio (AHTR), temperature-based duration variables, and their interaction terms as the most significant variables associated with brown rust epidemics of sugarcane in Florida. Kiwifruit brown spot initially occurred in late June, rapidly spread in mid-July, and gradually slowed down until the end of October. In this study, the key environmental variables affecting the potential distribution were identified using the jackknife method, including bio3, tmin8, tmax7, tmin6, prec8, and tavg10, which completely coincide with the actual phenology period of kiwifruit brown spot. Additionally, kiwifruit brown spot develops rapidly under high temperature and humidity conditions, with excessive temperatures inhibiting its development [
3,
7], which is consistent with the response curve results that the temperature in June and August should not be too low, the temperature in July should not be too high, and the precipitation in August reaches a certain amount. Although our study did not include all factors contributing to the distribution, the selected environmental variables can provide a basis for the future refinement and assessment of a prediction model of kiwifruit brown spot.
An ecological niche model is an emerging technology based on ecological principles, according to the known geographical distribution information of species and corresponding environmental variables, using specific algorithms to calculate the niche demand of target species in the designated area and combining GIS technology to project its distribution probability onto the map [
36,
37]. However, species distribution models usually lead to overestimating or underestimating the species distribution, namely, false positive and false negative. In this study, the suitability levels predicted by the MaxEnt model were often higher than the actual disease occurrence level, resulting in many false positives. We supposed that the reason for this phenomenon might be attributed to the species distribution model only distinguishing “existence” or “nonexistence” when extracting distribution information, without considering the disease occurrence level, resulting in overfitting. Additionally, the distribution information in this study was mainly obtained through field investigations and local reports inquiry, with a total of 225 distribution points. The data were true, reliable, and relatively systematic, but there might be some omissions in their completeness, which might also cause some errors. Furthermore, previous studies have indicated that the occurrence and epidemic of leaf spot disease were not only affected by meteorological factors, altitude, and cultivars, but also closely related to abiotic factors, such as planting density, site conditions, and canopy density [
14,
38,
39]. Gonzalez-Dominguez et al. [
40] developed a model using weather and host phenology to predict the infection period and disease progression of
Phomopsis cane and leaf spot throughout the season, and validated its performance using ROC analysis (AUROC > 0.7). Ortega-Acosta et al. [
41] established a Weibull model using multiple abiotic factors to describe the epidemic dynamics of Roselle leaf and calyx spot induced by
C. cassiicola. Therefore, various factors should be considered comprehensively in future research to further improve the prediction model of kiwifruit brown spot.