• Introduced LOESS smoothing and first-derivative analysis to identify the key developmental stages of Chilo suppressalis . • Determined the screening scheme for sensitivity parameters specific to each pest development stage. • Constructed a stage-specific daily dynamic prediction framework to achieve spatiotemporally explicit extrapolation of C . suppressalis population dynamics at the regional scale. Chilo suppressalis is a major pest in rice-producing regions, posing serious threats to rice yield and quality. Existing pest prediction research generally ignore the stage-specific heterogeneity in population dynamics and neglect the synergistic effects among meteorological, soil, and rice physiological information. This makes it challenging to accurately characterize the complete dynamic changes in pest populations. In this study, we explicitly highlight three methodological innovations: (1) the use of a LOESS-based curve fitting and slope-change detection framework to objectively partition C . suppressalis population dynamics into three stages—population establishment, expansion, and outbreak; and (2) the integration of multi-source data, including trap monitoring, rice physiological indices derived from remote sensing, and ERA5 meteorological and soil variables, to construct stage-specific prediction models.; and (3) building upon this stage-based framework, we designed a targeted sensitivity-parameter screening scheme and developed daily dynamic prediction models using the Random Forest (RF) algorithm, which incorporate meteorological, soil, and crop physiological indicators. The results demonstrate that the proposed stage-specific prediction model achieves excellent performance. During the outbreak stage, the R 2 for all three experimental fields exceeds 0.9, with MAE below 23.16 and RMSE under 32.86. In the Jiulong field, stage-specific predictions show R 2 values above 0.89. Compared with Long Short Term Memory (LSTM) and Prophet models, RF exhibits superior stability and generalization, with test set R² consistently above 0.69, highlighting its robustness and reliability for stage-specific prediction of C . suppressalis population dynamics. These findings highlight the practical value of our approach for enhancing comprehensive pest forecasting and supporting targeted pest management.
Wu et al. (Thu,) studied this question.