Analysis of Potential Distribution of Spodoptera Frugiperda In Northwest

Spodoptera frugiperda (J. E. Smith) (Lepidoptera: Noctuidae), a newly invaded pest that breaks out fast and severely, causes a serious threat to the national security of food production. In this study, the MaxEnt model was used to predict the potentially suitable distribution area of S. frugiperda in Northwest China. The potential distribution of S. frugiperda was predicted using meteorological factors from the correlation analysis. According to the result, a satisfactory AUC value in the MaxEnt model indicates that the prediction model has good accuracy, which is su�cient for predicting the �tness zone of S. frugiperda in Northwest China. The prediction results show that the potential distribution risk of S. frugiperda is high in western Gansu, eastern Qinghai, Shaanxi, most regions of Ningxia, and part regions of Tibetan, and it also exists in Hami, Yili, Bozhou, Urumqi, Hotan, and Aksu in Xinjiang, and more than 60% of Northwest China are suitable distribution areas for S. frugiperda. As China's major wheat and maize production area, Northwest China is a crucial prevention area for S. frugiperda. Clarifying the potential geographical distribution of S. frugiperda in Northwest China is essential for early warning as well as prevention and control.


Introduction
Spodoptera frugiperda (J.E. Smith) (Lepidoptera: Noctuidae), native to the tropical and subtropical regions of the Americas, is a major agricultural pest widely distributed in the American continent 1 .Since 2016, the insect has invaded more than 60 countries and regions, including Africa, Asia, and Oceania 2 .In December 2018, S. frugiperda adults invaded Yunnan, China, from Myanmar 3 .Moreover, on January 11, 2019, its larvae were rst witnessed damaging maize in Jiangcheng County, Yunnan Province, China.After that, S. frugiperda, with its powerful long-distance migration ability, spread rapidly and caused damage across the country 2,4 .As of August 31, 2020, S. frugiperda has been found in 1,338 counties of 27 provinces nationwide.Among them, only adult insects were seen in 21 counties of Ningxia, Liaoning, Inner Mongolia, Beijing, and Tianjin.In a survey in late August, the larvae were found in 865 counties of 22 provinces, and the area where larvae occurred was 3.68 million mu [5][6][7] .Thus, it is essential to clarify the potential distribution of S. frugiperda in China for early warning as well as comprehensive prevention and control 2 .
S. frugiperda has a strong ight capacity.Under suitable wind conditions, S. frugiperda adult can even travel 1,600 km within 30 h; in China, S. frugiperda is also predicted to migrate long distances 8,9 .S. frugiperda has a wide host range and can damage more than 300 species of plants, including rice, sorghum, millet, sugarcane, vegetable crops, and cotton.The larvae mainly feed on tender tips, drill the base of the stem, and damage the maize ears, resulting in reduced yield or even no harvest 10 .If proper control measures are not taken, S. frugiperda will cause severe yield losses.Based on the feeding preferences for host plants, S. frugiperda can be divided into two haplotypes: the maize strain, which mainly feeds on and harms maize, cotton, and sorghum; the rice strain, which mainly feeds on and harms rice and various forage grasses 11 .S. frugiperda that has invaded China belongs to the maize strain and damages maize mainly by feeding them in summer 7,12 .In the winter of 2019, S. frugiperda was found to invade in partial areas by feeding on wheat and barley 13 and in Guangxi and Yunnan by infesting sugarcane 14 .S. frugiperda is highly adaptable to environments and highly reproductive.Female S. frugiperdas can mate and spawn in a high frequency.Up to 2,300 eggs can be spawned at a time, depending on their nutritional conditions.In China, since S. frugiperda poses a serious long-term threat to the security of national food production, it is therefore classi ed as a major emergent pest on maize 8 .MaxEnt is a species distribution prediction model based on the maximum entropy theory and has been widely used due to its short running time, user-friendly operation, and stable running results.In recent years, using this model, global searchers have systematically studied pests such as Drosophila melanogastes 15 , Sirex noctilio 3 , batocera lineolata 16 , and S. frugiperda.Using the MaxEnt model and ArcGIS, researchers have predicted the potentially suitable area of S. frugiperda in Yunnan, China 17 .
Besides, Several studies have shown the potential distribution of S. frugiperda in Central Asia 18 .Model prediction results well simulated the potential distribution of S. frugiperda, which is consistent with the actual occurrence.S. frugiperda occurs less frequently in the irrigated areas of Northwest China (such as Xinjiang and Tibet), and the prediction shows a reduced overlap and less risk.The potential distribution of S. frugiperda was explored using limiting environmental factors and the collection model prediction scheme 19 .However, their prediction results lack reference value, especially for the Middle East and East Asia.The potential distribution of S. frugiperda in Central Asia and China have been simulated, with uncertainties in single model predictions (MaxEnt model).Regions such as Liaoning, Hebei, Beijing, Tianjin, Shanxi, Shaanxi, Inner Mongolia, Ningxia, Gansu, Qinghai, and Xinjiang are suitable distribution areas for S. frugiperda in spring, summer, and autumn 18,20 .
China is one of the maize-producing countries, and maize is an important food crop and cash crop for China.Northwest China is a major production area of wheat and maize, so it is a crucial prevention area for S. frugiperda.On May 31, 2019, S. frugiperda was rst discovered in Yang County, Shaanxi Province.
After that, damagze occurred in 63 counties of this province 21 .On July 2, 2019, S. frugiperda larvae were investigated for the rst time in summer maize elds in Duanheba Village, Liangshui Township, Wudu District, Gansu Province 22 .Although S. frugiperda has not caused serious damage in these regions, it is imperative to clarify the potential geographic distribution of S. frugiperda in Northwest China to guide early warning as well as prevention and control.This study aims to provide a theoretical basis for the further control of S. frugiperda in China.

Results
Prediction of potential distribution areas of S. frugiperda.The red parts in Figure 1 (A1, A2) show the highly suitable distribution areas of the S. frugiperda.As shown in Figure 1, most areas in China, such as Fujian, Jiangxi, Hunan, Guangdong, Anhui, Jiangsu, Hubei, Guangxi, Zhejiang, Shanghai, Henan, Shaanxi, Shandong, Sichuan, Chongqing, Guizhou, Tianjin, Shanxi, are highly suitable distribution areas; Ningxia, Tibet, and Gansu in Northwest China are also highly suitable distribution areas; some areas in Xinjiang are moderately suitable distribution areas; besides, other areas in Northwest China are mostly nonsuitable distribution areas.According to the correlation analysis, Hami, Ili, Bortala Mongol Autonomous Prefecture, Urumqi, Hotan, Aksu in northwest China are moderately suitable distribution; eastern Tibet (Linzhi, Naqu, Ganzi, Yushu, Hercynian Mongolian autonomous region), Gansu, Qinghai, Shaanxi, Ningxia in most areas are highly suitable distribution; most of Xinjiang are non-suitable areas (Figure 1 (B2)).In China, the highly suitable, moderately suitable, low suitable, and non-suitable distribution areas of S. frugiperda cover 42.81%, 9.31%, 13.73%, and 34.16% of the studied area, respectively.In Northwest China, the highly suitable, moderately suitable, low suitable, and non-suitable distribution areas account for 28.03%, 13.60%, 18.77%, and 39.6% of the studied area, respectively.
Evaluation of model accuracy.The AUC value was used to evaluate the accuracy of the model prediction, and the range is [0, 1].The larger the AUC value, the better the prediction effect, which indirectly re ects the sound prediction of the model.The ideal circumstance is that when the AUC value is 1, the distribution area predicted by the model is identical to the actual distribution area of S. frugiperda.The ROC curve was evaluated by the following benchmarks: when the value of AUC is between 0.5-0.6, the prediction result is failure; when the value of AUC is between 0.6-0.7, the prediction result is poor; when the value of AUC is between 0.7-0.8, the prediction result is fair; when the value of AUC is between 0.8-0.9, the prediction result is good; when the value of AUC is between 0.9-1, the prediction result is excellent.When the mean AUC value obtained by using 19 meteorological factors for the prediction analysis is 0.881 and when it is obtained by analyzing meteorological factors after ltering is 0.864, the prediction results are reasonable (Figure 2).Analysis of dominant environmental variables in uencing the potential distribution area of S. frugiperda.The Jackknife of the MaxEnt model was used to obtain the relative importance of different environmental variables on prediction.As shown in Figure 3, according to the tness analysis of S. frugiperda using 19 environmental variables, Bio19, Bio1, Bio11, and Bio18 have a greater in uence on the potential distribution of S. frugiperda.

Discussion
Since the invasion of S. frugiperda in Africa, it has caused annual losses of $2.481 billion to $6.187 billion 20 .For thousands of years, the planting industry has been dominant in China, and the signi cance of agriculture cannot be neglected.However, S. frugiperda poses a considerable threat to maize, wheat, and other main food crops in China.As of April 2020, the occurrence acreage of S. frugiperda reached 111.33 hm 2 in Yangdong District, Yangjiang City, Guangdong Province, including a heavy occurrence acreage of 13.33 hm 2 23 .Based on the meteorological data after conducting correlation analysis, the results reveal that 65% of areas of China are the potential distribution areas of S. frugiperda.In addition, over 60% of the areas in Northwest China are suitable distribution areas (28.03% in the highly suitable distribution areas, 13.60% in the moderately suitable distribution areas, and 39.6% in the low suitable distribution areas).Among those areas, Linzhi and part areas of Shannan in Tibet, western areas of Gansu, Shaanxi, and most areas of Ningxia are highly suitable distribution areas; sporadic areas of Hami, Yili, Bozhou, Urumqi, Hotan, and Aksu in Xinjiang are moderately suitable distribution areas.There are also potential distribution areas in other parts of Northwest China.
In this study, the MaxEnt model was used to predict the suitable distribution.After correlation analysis, the model results show that most areas in China, such as Fujian, Jiangxi, Hunan, Guangdong, Anhui, Henan, Guizhou, and Yunnan provinces, are highly suitable distribution areas.The above results are generally consistent with those in previous studies 20 .However, the study of the potential distribution areas in Northwest China differed somewhat from the previous studies 20 .It is concluded that Shaanxi, Gansu, Ningxia, and parts of Qinghai in Northwest China are low suitable distribution areas; most parts of Xinjiang and Tibet are non-suitable distribution areas but also have high tness zones.In contrast, in this study, most areas of Shaanxi, Qinghai, Gansu, Ningxia, and part of Tibet are considered highly suitable distribution areas; some parts of Xinjiang are moderately suitable distribution areas for the S. frugiperda.This classi cation may be led by the different sample sites.In this study, distribution data from CABI (www.cabi.org/ISC /datasheet /29810) were used to construct the model.CABI maps the approximate distribution of invasive species at a macroscopic scale, with a focus on national or provincial administrative centers.These administrative centers only represent the area where the species is found, which are signi cantly different from the speci c distribution points.Therefore, they cannot re ect the actual distribution of the species in detail 24 .
In this study, the distribution areas in the world of S. frugiperda were involved in the model analysis.The suitable distribution areas in Northwest China covered a relatively high proportion.Both this study and previous studies 20 reveal that S. frugiperda has strong migration ability in the suitable distribution areas in Northwest China.Northwest China, as an important maize production area, may still be a potentially suitable distribution area for S. frugiperda.Therefore, this insect should be closely monitored and controlled.
Four environmental variables that exerted a strong in uence on the distribution points of S. frugiperda were obtained by correlation analysis, namely Bio19 (Precipitation of the coldest quarter ), Bio18 (Precipitation of the warmest quarter), Bio4 (Temperature seasonality), and Bio12 (Average annual precipitation).Besides, among the 19 environmental variables models, the environmental variables that had a greater impact on the distribution area were Bio19(Precipitation of coldest quarter ), Bio1 (annual mean temperature), Bio11 (Mean temperature of the coldest quarter), and Bio18 (Precipitation of the warmest quarter).This result shows that temperature and precipitation are the main factors affecting the distribution of S. frugiperda.A relatively low temperature with a large temperature difference helps prohibit the spread and establishment of the insect 17 .In addition, precipitation impacts the migration and dispersal of S. frugiperda.S. frugiperda can be divided into two haploid genotypes, including the maize strain and the rice strain.The former feeds mainly on maize, cotton, and sorghum, while the latter feeds more on rice and various forages.Both have caused signi cant damage not only to the economy of China but also the world 25 .In this paper, only 19 meteorological factors were used to analyze and predict the prediction model merely in an ideal and objective situation.However, the actual distribution of S. frugiperda was not only related to climatic factors, but anthropogenic activities, food, and natural enemies could impact its distribution.In future studies, a broad viewpoint is needed to achieve more accurate predictions 26 .

Materials And Methods
Distribution data and treatment of Spodoptera frugiperda.The distribution data of S. frugiperda in this study were divided into two parts: (1) valid data for S. frugiperda were downloaded from the Global Biodiversity Information Facility (GBIF) (http://www.gbif.org/)database and the species distribution data of the model were recorded.(2) Recorded data of China's reports.A total of 3,085 sample points were exported into a table after importing distribution point data using ArcGIS10.2software(Figure 4).The table was saved as a CSV le in the order of species name, the longitude and latitude of distribution points, and sample points were acted as the species distribution data of the model.(2) Correlation analysis: During the operation of MaxEnt, if there is a strong correlation between the environmental variables, the model will be excessively tted.Therefore, ArcGIS software was used to extract the values of 19 variables.Pearson correlation coe cient was used to calculate the correlation between variables whose correlation coe cient is higher than 0.9 were eliminated.
(3) Nineteen environmental variables were run in the MaxEnt model (except for the number of runs, the other settings were consistent with the operation of this experimental model), and then variables were screened according to the contribution rate and correlation analysis results of each variable in the initial model.After the above procedures, four variables (BIO19, BIO18, BIO12, bio4) were nally reserved for establishing the nal model 27 .
Software and map data.MaxEnt (version 3.3.3.)used in this paper was downloaded from the MaxEnt homepage (http: //www.cs.princeton.edu~schapire/MaxEnt).ArcGIS 10.2, developed by Environmental Systems Research Institute, Inc., was also used.The vector map scale of China is 1:4 million, which was downloaded from the National Geomatics Center of China (http://nfgis.nsdi.gov.cn/) for free.
Methods.The distribution points and environmental variables of S. frugiperda were imported in MaxEnt; 75% of the distribution points were randomly selected as training data, and 25% as test data.Other parameters in the Jackknife model were selected as default parameters, and the results predicted in this model were continuous raster data with values between 0 and 1.The output le format of MaxEnt was ASCII, which would be imported into ArcGIS to be converted to raster format.The Natural Breaks was used to classify the tness level of S. frugiperda into four categories according to the tness index (P) with Jenks' natural breaks method 28 : non-suitable distribution areas (P < 0.08), low suitable distribution areas (0.08 ≤ P < 0.25), moderately suitable distribution areas (0.25 ≤ P < 0.47), and highly suitable distribution areas (P ≥ 0.47).The vector maps of Shaanxi, Gansu, Ningxia, Xinjiang, and Tibet were used as the base map to conduct extraction by mask in ArcGIS.The Natural Breaks was used for classi cation again.Finally, the potential distribution areas of S. frugiperda in Northwest China were identi ed.

Declarations Figures
The in uence of environmental variables for S. frugiperda distribution prediction Figure 4 Distribution map of Spodoptera frugiperda Selection of environmental variables.(1) Using 19 environmental variables in the global prediction model: Data on environmental variables for global prediction models: Nineteen bioclimatological variables with a spatial resolution of 2.5' were originated from WORLDCLIM (http //www.worldclim.org/)version 1.4.