The fall armyworm (Spodoptera frugiperda), a highly polyphagous and destructive agricultural pest, has posed severe threats to food security in East Africa since its invasion. Identifying habitat suitability for the fall armyworm is critical for early prevention and control. However, existing studies have not adequately incorporated phenological information, which limits the accuracy of habitat suitability extraction. This study constructed a dataset integrating phenological and other environmental factors. Using both historical occurrence records and pseudoabsence points generated by the “E-H” method, habitat suitability models were developed based on four machine learning algorithms (SVM, RF, XGBoost, and LightGBM) to extract the habitat suitability for the fall armyworm in East Africa from January to December 2022. The results showed that LightGBM outperformed other models across multiple evaluation metrics, achieving an accuracy of 85.7%, an Area Under the Receiver Operating Characteristic Curve of 0.936, a Kappa value of 0.871, and a true skill statistic of 0.821. Compared to models without phenological information, the improved model significantly enhanced the accuracy of habitat suitability extraction and better captured the potential distribution of the fall armyworm. By incorporating phenological data, the study also conducted a detailed analysis of the generational dynamics of the fall armyworm in East Africa, providing valuable insights for control measures. Additionally, habitat suitability extraction at a finer scale was performed using Sustainable Development Goals Science Satellite 1 thermal infrared spectrometer data, which preliminarily validated its stability and superiority in fall armyworm habitat suitability applications. These findings offer scientific support for food security and sustainable agricultural development.
Wang et al. (Thu,) studied this question.