Occurrence Dynamics and Prediction of Coleoptera and Lepidoptera in China Using Multiple Machine Learning Models.

The temporal dynamics of forest-associated insects are closely associated with climatic variability, and identifying their temporal patterns and climatic associations is important for understanding insect population dynamics. In this study, automatically recorded Coleoptera and Lepidoptera in Rui'an City, Zhejiang Province, China, were selected as the target insect groups. Based on automatic insect monitoring and meteorological data, the temporal characteristics of recorded insect abundance and their associations with climatic factors were analyzed. Five machine learning models, including Multiple Linear Regression, K-Nearest Neighbors, Random Forest, Support Vector Regression, and Generalized Additive Model, were then developed and compared for predicting recorded insect abundance. The results showed that Coleoptera exhibited periodic fluctuations with cycles of approximately 6-20 days, whereas Lepidoptera displayed persistent seasonal variation. Air temperature, specific humidity, and shortwave radiation were significantly and positively correlated with recorded insect abundance, while wind speed showed a weak negative correlation. Among the five models, Random Forest achieved the best prediction performance, with MAE, RMSE, and R2 values of 1.37, 4.85, and 0.82 for Coleoptera and 0.57, 1.16, and 0.84 for Lepidoptera, respectively. Independent-year validation using 2023 data showed reduced predictive performance compared with random validation, indicating limited temporal generalization under the available observation period. These findings demonstrate that integrating temporal characteristics with climatic factors can support the prediction of automatically recorded insect abundance, while longer-term monitoring data are needed to further evaluate model robustness and temporal transferability.