Precipitation forecasting and feature correlation analysis based on the ARIMA-GM(1,1) model

Abstract

This study addresses the ambiguity in defining “rainy spells” through a threshold-based approach aligned with the National Precipitation Grade Standard. A hybrid ARIMA-GM(1,1) time series forecasting model was developed to mitigate single-model limitations in handling anomalies and noise. Cross-validation determined optimal weights (ARIMA(0,1,1): 0.3849; GM(1,1): 0.6151) for combined prediction. Empirical validation using Hangzhou’s 2025 Qingming Festival precipitation data (predicted: 1.84/0.8/0.7 mm vs. actual: 0 mm) demonstrated superior performance with reduced mean absolute error compared to individual models. A ridge regression framework was established by analyzing 2006–2025 meteorological datasets (daily temperature extremes, precipitation) and phenological records through Spearman correlation analysis. The model predicted 2026 flowering periods. This methodology provides a robust framework for meteorological event forecasting and agricultural phenology modeling