Introduction
The desert locust, Schistocerca gregaria is known to be the oldest migratory insect (Song et al., Reference Song, Foquet, Mariño-Pérez and Woller2017), capable of covering distances of 150–200 km daily with the airflow (Yuga and Wani, Reference Yuga and Wani2022). They feed on over 400 types of plants, including wheat, corn, rice, potatoes, and various other crops and each locust can consume fresh green leaves equivalent to its body weight daily (FAO, 2020; Yuga and Wani, Reference Yuga and Wani2022). Thriving in semi-arid and arid environments, desert locusts find suitable breeding sites in bare lands (Latchininsky and Sivanpillai, Reference Latchininsky, Sivanpillai, Ciancio and Mukerji2010) and are widely distributed in African and the Middle Eastern deserts. They pose a threat to food security and significantly impact local agricultural production (Brader et al., Reference Brader, Djibo, Faye, Ghaout, Lazar, Luzietoso and Babah2006; Kimathi et al., Reference Kimathi, Tonnang, Subramanian, Cressman, Abdel-Rahman, Tesfayohannes, Niassy, Torto, Dubois, Tanga and Kassie2020). Desert locusts affect at least 1/10th of the world's population directly or indirectly (Latchininsky et al., Reference Latchininsky, Piou, Franc, Soti, Baghdadi and Zribi2016).
With the advancement of civilisation, the excessive use of fossil fuels has led to a tremendous emission of greenhouse gases, including carbon dioxide, significantly contributing to global warming (Aminu et al., Reference Aminu, Nabavi, Rochelle and Manovic2017). Studies indicate that global warming may expand pest geographical distribution, increase overwintering survival rates, boost reproductive rates, and alter plant–pest interactions (Skendžić et al., Reference Skendžić, Zovko, Živković, Lešić and Lemić2021). Simultaneously, it provides an opportunity for pests to move from one region to another in search of food and colonise new areas (FAO, 2020). Research has indicated a close correlation between the occurrence and migration of desert locusts with climate factors. Climate change, such as rainfall, can cause seasonal disasters of desert locusts (Uvarov, Reference Uvarov1966). In 2020, a desert locust plague in East Africa was attributed to a cyclonic storm in the Arabian Peninsula in 2018. The abnormal climate brought additional rainfall to the Arabian Desert, increasing the hatching rate of desert locust eggs. Abundant rainfall also promoted vegetation growth, providing ample food supply for desert locusts. The combination of warm weather and abundant rainfall triggered an 8000-fold increase in locusts in the Arabian Desert (Stone, Reference Stone2020). However, 2019 saw a decrease in rainfall in the Arabian Peninsula. As a result, the vegetation failed to meet the needs of high-density desert locusts, leading to the eastward migration of desert locusts and causing locust infestations in bordering areas, such as the India–Pakistan border in 2020 (Salih et al., Reference Salih, Baraibar, Mwangi and Artan2020). There is a close correlation between the occurrences of desert locusts and the environment. Nevertheless, their rapid escalation into outbreaks in the short term and the uncertainty surrounding outbreak area make prediction efforts particularly crucial. Since the 1900s, efforts have been made to forecast desert locusts, mainly through remote-sensing technology (Klein et al., Reference Klein, Oppelt and Kuenzer2021). This involves on-site investigations to record the growth conditions and population dynamics of desert locusts, combined with historical ecological data to predict the distribution of desert locusts in the near future (Cressman, Reference Cressman2013).
Maximum entropy (MaxEnt) modelling, a programme that models species distributions from occurrence records of ‘presence-only’ species (Phillips et al., Reference Phillips, Dudík and Schapire2004), analyses the relationship between the species record locations and environmental characteristics. Locations with similar environmental characteristics are predicted as potential distribution areas for the species (Elith et al., Reference Elith, Phillips, Hastie, Dudík, Chee and Yates2011; Sillero, Reference Sillero2011; Ahmed et al., Reference Ahmed, Mcinerny, O'Hara, Harper, Salido, Emmott and Joppa2015; Li et al., Reference Li, Li, Li and Liu2020; Muyobela et al., Reference Muyobela, Pirk, Yusuf and Sole2023). MaxEnt model demonstrates a notable advantage, requiring a relatively low number of species sampling points (≥5 points) for accurate predictions, making it a user-friendly and computationally efficient approach with high accuracy (Phillips and Dudík, Reference Phillips and Dudík2008). Wen et al. (Reference Wen, Lu, Nie, Sun, Liu, Huang and Ye2023) analysed the habitat suitability of Oedaleus asiaticus on the Mongolian Plateau; the results showed that the suitable area of O. asiaticus was increasing. Accumulated precipitation and surface temperature are identified as the main driving factors for the changes in the distribution of O. asiaticus. Abou-Shaara et al. (Reference Abou-Shaara, Amiri and Parys2022) analysed the potential habitable areas of the pest Plecia nearctica offering insights that can provide information for making control measures. Until now, MaxEnt has been widely used in predicting suitable areas for insects. In this study, we used MaxEnt model and employed environmental data from known distribution points to identify critical factors influencing their distribution. After factor optimisation, we analysed the potential geographical distribution of desert locusts under the present and future climate parameters, aiming to make accurate predictions in advance and guide efforts for the prevention and control of desert locusts.
Materials and methods
Species occurrence data
The occurrence data of desert locusts were sourced from the Locust Hub (https://locust-hub-hqfao.hub.arcgis.com/) of the Food and Agriculture Organization of the United Nations (FAO). To mitigate spatial autocorrelation between occurrence locations (Verbruggen et al., Reference Verbruggen, Tyberghein, Belton, Mineur, Jueterbock, Hoarau, Gurgel and Clerck2013) and address sampling bias induced by clustering effects, the collected occurrence data were loaded into ArcGIS. This process involved removing non-natural points eliminating duplicate coordinates in latitude and longitude. To avoid the bias caused by unbalanced data points, we checked and confirmed that each grid cell had only a single distribution point based on climatic precision. Ultimately, 1915 occurrence data points of desert locusts were used for modelling (table S1, fig. 1).
Figure 1. Global occurrence records of locusts. Each red point refers to an occurrence record.
Environmental data
The environmental data were sourced from the global climate database, WorldClim (https://worldclim.org/), with a spatial resolution of 2.5 arc-minutes and encompassing 43 environmental factors, comprising 19 bioclimatic variables and 24 climate variables. The current climate data were composed of historical climate data from 1970 to 2000. The BCC-CSM2-MR model was selected for future climate data (Wu et al., Reference Wu, Lu, Fang, Xin, Li, Li, Jie, Zhang, Liu, Zhang and Zhang2019). The study also chose the climate change scenario data for the years 2030 (average for 2021–2040), 2050 (average for 2041–2060), 2070 (average for 2061–2080), and 2090 (average for 2081–2100) under the sustainable development pathway (SSP1–2.6), and the fossil fuel-intensive development pathway (SSP5–8.5) (Popp et al., Reference Popp, Calvin, Fujimori, Havlik, Humpenöder, Stehfest, Bodirsky, Dietrich, Doelmann, Gusti, Hasegawa, Kyle, Obersteiner, Tabeau, Takahashi, Valin, Waldhoff, Weindl, Wise, Kriegler and van Vuuren2017; Meinshausen et al., Reference Meinshausen, Nicholls, Lewis, Gidden, Vogel, Freund, Beyerle, Gessner, Nauels, Bauer and Canadell2020).
To address multicollinearity among environmental variables and improve the simulation accuracy of the MaxEnt model (Feng et al., Reference Feng, Park, Liang, Pandey and Papeş2019), the ‘SDMtune’ package was used for appropriate variable selection (Vignali et al., Reference Vignali, Barras, Arlettaz and Braunisch2020). The dataset was divided into 25% testing data and 75% training data to train the MaxEnt model. Environmental variables with a correlation coefficient greater than 0.8 were removed based on their contribution percentages (fig. 2A). Finally, ten biologically meaningful climate variables were selected for modelling, including mean diurnal temperature range (bio02), temperature annual range (bio07), mean temperature of the coldest quarter (bio11), precipitation of the driest month (bio14), precipitation seasonality (bio15), precipitation of the warmest quarter (bio18), precipitation of the coldest quarter (bio19), precipitation of February (prec02), precipitation of April (prec04), and precipitation of June (prec06).
Figure 2. Model optimisation. (A) Heat map showing the correlation coefficient matrix of ten selected environmental variables. (B) Optimal model selection for predicting potential suitable areas of desert locusts.
Optimisation of model parameter
The calibration, creation, and evaluation of the model were conducted using the ‘Kuenm’ package in R (Cobos et al., Reference Cobos, Peterson, Barve and Osorio-Olvera2019), in which the two most critical parameters are feature combination (FC) and regularisation multiplier (RM) (Merow et al., Reference Merow, Smith and Silander2013). Here, FC covers five main features: linear features, quadratic features, product features, threshold features, and hinge features, with a total of 29 other parameter combinations. Similarly, RM parameters were set with intervals of 0.1, ranging from 0.1 to 4, resulting in a total of 40 RM values. This yielded 1160 (29 FC × 40 RM) parameter combinations. Model performance was assessed based on the omission rate (E), model complexity, and the Akaike information criterion (AICc) (Sugiura, Reference Sugiura1978; Akaike, Reference Akaike, Petrov and Csaki1998). A smaller omission rate (E) should be less than 5%, while a lower AICc value indicates a better fit between the model and the data. For delta_AICc, a lower value indicates more accurate predictions, and the model is deemed optimal when delta_AICc = 0. Ultimately, the model exhibited the highest performance when RM = 0.3 and FC = QP (Quadratic features, Product features), with E = 0.048 (table S2, fig. 2B).
Based on MaxEnt 3.4.4, the potential suitable areas of desert locusts were predicted. This study randomly selected 25% of the desert locust data as the testing set, while the remaining 75% were used as the training set. After ten runs, the output prediction results were averaged, with the area under curve (AUC) as the evaluation metric for model accuracy. Higher AUC values correlate with greater accuracy (Phillips et al., Reference Phillips, Anderson and Schapire2006). AUC values between 0.5 and 0.7 indicate low prediction accuracy, between 0.7 and 0.8 indicate moderate accuracy, and above 0.8 indicate high accuracy. The jackknife method was employed to test the importance of each environmental factor in the suitable area distribution of desert locusts (Guan et al., Reference Guan, Li, Ju, Lin, Wu and Zheng2021). A larger regularisation training gain value in the jackknife method indicates a greater influence of that factor. Finally, the ASC (ASCII File) result files were imported into ArcGIS to delineate the suitable areas of desert locusts.
Suitable area division and spatial pattern changes
The minimum training presence logistic threshold was chosen as the classification threshold for non-suitable and suitable areas (Escalante et al., Reference Escalante, Rodríguez-Tapia, Linaje, Illoldi-Rangel and González-López2013). The output values from MaxEnt software range from 0 to 1, where values closer to 1 indicate a higher likelihood of species presence. Given that the value of the minimum training presence logistic threshold was 0.0649 (table S3), the suitable area range was divided into four levels: non-suitability area (0–0.0649), low-suitability area (0.0649–0.35), moderate-suitability area (0.35–0.55), and high-suitability area (0.55–1.0).
The output from MaxEnt software was loaded into ArcGIS, and the reclassify tool in spatial analysis was used for visual representation of the predicted results. Finally, the proportions of different suitability levels were calculated.
Results
Modelling result evaluation
The prediction model of suitable areas for desert locusts exhibited an average AUC value of 0.900 over ten runs (fig. 3A), indicating higher prediction accuracy of the MaxEnt model and its suitability for predicting the suitability of areas for desert locusts.
Figure 3. (A) Model suitability test on the basis of AUC value in predicting desert locusts' suitable areas. (B) Importance of ten selected environmental variables evaluated by jackknife testing.
Factors influencing spatial distribution pattern of desert locusts
According to the contribution results of environmental variables (table S4), temperature annual range (bio07), mean temperature of coldest quarter (bio11), precipitation of February (prec02), and precipitation of driest month (bio14) were the major factors affecting the model prediction results. These factors displayed respective contributions of 43.6, 33.5, 8.3, and 4.5%, with a cumulative contribution of 89.9%. Temperature annual range (bio07) and mean temperature of coldest quarter (bio11), as shown in the results of the jackknife test (fig. 3B), exhibited the greatest gain in prediction effect, indicating their greatest influence on the model prediction results.
Analysis of the curves of the most important contributing environmental variables in the MaxEnt model of desert locusts (fig. 4A) shows that within a 1°C range of annual temperature variations, the occurrence probability of desert locusts is as high as 68.5% and decreases with an increase in the value of annual temperature variations. As the average temperature in the coldest season rises (fig. 4B), the occurrence probability of desert locusts declines and levels off at around 29°C, with an occurrence probability of about 40.0%. Desert locusts tend to avoid areas with excessive precipitation. When the precipitation in February (fig. 4C) is 1 mm, the probability of occurrence is the highest. As precipitation increases, the probability of occurrence decreases and tends to stabilise at 160 mm; higher rainfall in the driest month (fig. 4D) corresponds to a lower occurrence probability of desert locusts, with the occurrence probability close to 0 when it reaches 80 mm.
Figure 4. Response curves of the important contributing environmental variables in the MaxEnt model: (A) temperature annual range; (B) mean temperature of coldest quarter; (C) precipitation of February; (D) precipitation of driest month.
Distribution of potential suitable areas for desert locusts under current climate
Under the current climate, the major suitable areas for desert locusts are located in northern Africa and southwestern Asia, near the Horn of Africa (fig. 5). These areas include Morocco, Algeria, Libya, Egypt, Western Sahara, Mauritania, Mali, Niger, Chad, Sudan, Eritrea, Djibouti, Ethiopia, Central Africa, Cameroon, Nigeria, Benin, Togo, Ghana, Guinea, Bissau, Senegal, Uganda, Kenya, Syria, Jordan, Iraq, Saudi Arabia, Yemen, Oman, Iran, Afghanistan, Pakistan, India, and other regions. In the coastal areas of the Americas, there are low suitable areas, such as Mexico, Venezuela, Peru, Chile, etc. In Oceania, only Australia has low suitable areas. Europe also has low suitable areas distributed in very few areas near Africa, Portugal, Spain, Italy, Greece, and other regions. High-suitability areas are mainly distributed in Mauritania, Mali, Niger, Chad, Sudan, Ethiopia, Kenya, Somalia, Saudi Arabia, Yemen, Oman, India, and other regions.
Figure 5. Potential suitable areas for desert locusts under current climate conditions.
Statistically, the suitable areas for desert locusts cover an area of about 0.2976 × 108 km2, around 19.97% of the global land. High-suitability areas, mainly distributed in Mauritania, Mali, Nigeria, Chad, Sudan, Ethiopia, Kenya, Saudi Arabia, Yemen, Pakistan, India, etc., cover 0.0479 × 108 km2, 3.21% of the global land. Moderate-suitability areas cover 0.1005 × 108 km2, 6.75% of the global land, while low-suitability areas, mainly distributed in the Americas, Oceania, and western Asia, cover 0.1492 × 108 km2, 10.01% of the global land.
Changes in suitable areas of desert locusts under future climate conditions
Under the SSP1–2.6 scenario (fig. 6A, C), the total suitable areas for desert locusts exhibit a decreasing trend and become relatively stable after 2050. Similarly, the low- and moderate-suitability areas show a decreasing trend, with these areas being transformed into high-suitability areas, which exhibit an increasing trend until 2070. Under the SSP5–8.5 scenario (fig. 6B, D), the total suitable areas undergo a decrease followed by an increase. However, the high-suitability areas show expansion throughout the period.
Figure 6. Potential suitable areas for desert locusts under the future climate scenario: (A) under the SSP1–2.6 scenario and (B) under the SSP5–8.5 scenario. Changes in each suitable area of desert locusts from current to 2090: (C) under the SSP1–2.6 scenario and (D) under the SSP5–8.5 scenario.
In 2030, the high-suitability areas under the SSP1–2.6 scenario cover 0.0567 × 108 km2 (0.6%) more than the current area. Similarly, under the SSP5–8.5 scenario, the high-suitability areas cover 0.0606 × 108 km2 (1.85%) more than the current area. By 2090, the high-suitability areas under the SSP1–2.6 and SSP5–8.5 scenarios will respectively cover 0.0603 × 108 and 0.0891 × 108 km2, with expansions of 1.84 and 2.77% compared to the current area. In comparison with the current high-suitability areas, the future high-suitability areas will mainly expand in North African regions such as Algeria, Libya, Chad, Egypt; and India, Iran, Saudi Arabia, and Iraq.
A comparative analysis has been conducted between the potential high-suitability areas under different climates in 2090 and the current ones. The results showed that under the SSP1–2.6 scenario (fig. 7A), the stable areas of desert locusts cover an area of 0.0788 × 108 km2, with an increase of 0.0231 × 108 km2 in the high-suitability areas, mainly concentrated in Algeria, Chad, Sudan, Saudi Arabia, Nigeria, and other regions. Under the SSP5–8.5 scenario (fig. 7B), the stable areas of desert locusts cover an area of 0.0717 × 108 km2, with an increase of 0.0759 × 108 km2 in the high-suitability areas, mainly concentrated in Algeria, Niger, Libya, Chad, Egypt, Sudan, Saudi Arabia, Iran, Pakistan, and other regions. The high-suitability areas exhibit northward expansion in this scenario compared with SSP1–2.6. The regions of the stable areas are almost the same in both scenarios, but the increase in high-suitability areas under SSP5–8.5 is larger than that under SSP1–2.6 by 0.0528 × 108 km2.
Figure 7. Changes in high suitable areas of desert locusts in 2090: (A) under the SSP1–2.6 scenario and (B) under the SSP5–8.5 scenario.
Discussion
In this study, we collected 1915 representative occurrence data points and used an optimised model to predict the distribution area changes during climate change. Actually, the default parameters of the MaxEnt model were initially designed to simulate the actual spatial distribution of 266 species in six different regions. However, for the prediction of the spatial distribution areas of species, the model may face issues such as overfitting, higher complexity, and lower accuracy, which require further optimisation (Warren and Seifert, Reference Warren and Seifert2011; Radosavljevic and Anderson, Reference Radosavljevic and Anderson2014). It has been shown that adjustments to the model parameters made by delta_AICc can constrain the model complexity (Akaike, Reference Akaike, Petrov and Csaki1998). Larger values of delta_AICc indicate higher model complexity, and vice versa (Sugiura, Reference Sugiura1978). In our study, ‘Kuenm’ package in R was employed to optimise the MaxEnt model, with the parameter of E < 2 and the smallest delta_AICc value selected to avoid overfitting and improve prediction accuracy. Additionally, the selection of environmental variables is crucial for predicting the distribution of suitable areas. For the selection of environmental variables, the ‘SDMtune’ data package (Vignali et al., Reference Vignali, Barras, Arlettaz and Braunisch2020) was applied for correlation analysis of 43 variables to eliminate environmental variables with strong correlations. The optimised model exhibits an AUC value of 0.900, indicating a better fit between the model prediction and the actual spatial distribution, allowing for accurate prediction of the potential geographic distribution of desert locusts.
The study found that the distribution probability of desert locusts decreased with the increase in the average temperature in the coldest season, levelling off at 29°C. This suggests that despite the preference of desert locusts for high temperatures, overly high temperatures can also affect their survival, possibly because higher temperatures induce a decrease in the hatching rate or survival rate of their eggs. The egg-hatching rate of desert locusts was proportional to the temperature rise from 15 to 35°C, but too high temperature will reduce the hatching rate (Hunter-Jones, Reference Hunter-Jones1970). The 2020 desert locust outbreak was caused by abundant precipitation (Ceccato et al., Reference Ceccato, Cressman, Giannini and Trzaska2007; Wang et al., Reference Wang, Zhuo, Pei, Tong, Han and Fang2021; Zhao et al., Reference Zhao, Li, Huang, Dong, Geng, Ma and Chen2023). However, this study found that desert locusts avoid areas with excessive precipitation, and their distribution probability remains almost zero when precipitation reaches 80 mm in the driest month. Additionally, overly wet soil fails to meet their reproduction requirements. Previous studies have found that desert locusts will not lay eggs until precipitation exceeds 25 mm since sufficient moisture ensures egg hatching and larvae growth (Roffey and Popov, Reference Roffey and Popov1968). However, extreme rainfall could overwhelm locust eggs and result in death (Ackonor, Reference Ackonor1989; Dinku et al., Reference Dinku, Ceccato, Cressman and Connor2010). Only suitable temperature and humidity can provide the necessary conditions for desert locusts’ survival.
The spread and diffusion of organisms is a process influenced by various environmental factors, with outbreaks being the result of the combined effects of climate, vegetation, soil, and other factors. Previous studies showed extreme climate events can enhance the process of biological invasion (Bellard et al., Reference Bellard, Thuiller, Leroy, Genovesi, Bakkenes and Courchamp2013), and a series of chain reactions caused by global climate change have made invasive species adaptable to new habitats (Hoffmann and Sgrò, Reference Hoffmann and Sgrò2011), which raises threats to local production and economies. Given the influence by abnormal climatic conditions, in 2018, the Arabian Peninsula and the Indian Peninsula's desert areas witnessed an increase in rainfall, leading to the formation of seasonal lakes in the desert that provided favourable conditions for plant growth and also for the breeding and growth of desert locusts. The population of desert locusts considerably rose in a short period, resulting in the widespread locust plague in early 2020 (Salih et al., Reference Salih, Baraibar, Mwangi and Artan2020; Stone, Reference Stone2020). Additionally, as a migratory invasive insect, desert locusts can also spread through air currents (Wang et al., Reference Wang, Zhuo, Pei, Tong, Han and Fang2021), with their main dispersal routes influenced by the monsoon. In the summer, desert locusts move towards Central Asia and the Arabian Peninsula in the northeast direction with the help of the southwest monsoon, then continue to disperse eastwards to India and Pakistan, and northwards up to the surroundings of the Caspian Sea (Rainey, Reference Rainey1951), severely threatening agriculture and pastoral production. With great cautious, these newly emerged suitable areas should pay additional attention to monitor the migration and prevent the establishment of desert locusts.
Conclusion
We used optimised MaxEnt model and parameters to predict potential geographical distribution of desert locusts. The main climatic factors affecting the distribution of desert locusts are: temperature annual range, mean temperature of coldest quarter, precipitation of February, and precipitation of driest month. The results showed an increase of high risk area in the future climate conditions, especially in North Africa. The newly emerged suitable areas included Brazil, Argentina, Chile, Venezuela, America, Mexico, Spain, Turkey, and Australia. Our study provided important evidence for desert locusts’ monitoring and management.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0007485324000440.
Acknowledgements
This work was supported by the Special fund of National Key R&D Program of China (2021YFE0194800), the earmarked fund for CARS (CARS-34), Public-interest Scientific Institution Basal Research Fund (Y2022GH12), and International Science & Technology Innovation Program of Chinese Academy of Agricultural Sciences (CAAS-ZDRW202108).
Competing interests
None.
