The applicability of statistical post-processing techniques for quantitative precipitation forecast in the Huaihe River Basin

The upper Huaihe River Basin above Bengbu. Statistical post-processing is essential for accurate quantitative precipitation forecasts. This study evaluates the applicability of post-processing methods, including statistical, time series, and spatiotemporal models, considering factors such as topography, climate, and seasonality. Firstly, ECMWF forecasts data and topographic variables serve as predictors, with CN05.1 observed data as validation. Then Post-processing models calibrate raw forecasts, evaluated using RMSE, CC, MAE, and the TOPSIS method for comprehensive scoring. Key findings include: (1) The spatiotemporal prediction model based on deep learning, particularly ConvLSTM outperforms other types of models in overall performance, particularly during the flood season, especially in regions with complex topography, and in areas with highly variable climate. (2) Statistical models are more effective in stable conditions, providing reliable predictions when precipitation fluctuations are minimal, but they tend to underperform during extreme precipitation events. (3) Seasonal variation, terrain complexity, and climate patterns significantly impact model performance. Dry seasons yield lower errors and higher correlations, while flood seasons, hilly areas, and humid zones with intense, variable precipitation result in larger errors, especially for statistical models. Overall, this study provides insights into the selection and optimization of precipitation forecast models, with applications in climate change adaptation, extreme weather management, disaster prevention, and resource planning. • Evaluated performance for post-processing of quantitative precipitation forecast. • Verified the effectiveness of ConvLSTM across terrains, climates, and seasons. • Revealed QRF and STGCN as alternatives for short- and long-term stages, respectively.