Long-term Precipitation Estimation Combining Time-Series Retrospective Forecasting and Downscaling-Calibration Procedure

Long-term precipitation datasets with high spatial resolution are crucial for hydrological, meteorological, and ecological research. However, due to the coarse spatial resolution and relatively short observation period of satellite-derived precipitation, obtaining long-term precipitation data with high spatial resolution remains to be a challenging task. In this study, autoregressive integrated moving average models (ARIMA and ARIMAX) are applied to retrospectively forecast month and annual Tropical Rainfall Measuring Mission (TRMM) precipitation during 1982–1997 at a 0.25°-pixel scale. The downscaling-calibration procedure including geographically weighted regression (GWR) and ten-fold cross-validation geographical difference analysis (10CV-GDA) is employed to obtain long-term precipitation (1982–2018) at high spatial resolution over the Yangtze River Basin (YRB). The precision of supplementary TRMM precipitation generated by ARIMAX is significantly higher than those generated by ARIMA, and supplementary TRMM precipitation is considered suitable for the subsequent downscaling-calibration procedure. The 10CV-GDA calibration significantly improved the precision of the downscaled results in comparison to the results obtained without or with geographical difference analysis (GDA) calibration (coefficients of determination (R²): 0.86, mean absolute error (MAE):106.29 mm, and root mean square error (RMSE): 147.29 mm). The annual downscaled results are disaggregated into 1 km monthly precipitation, and the overall performances of precipitation estimation are excellent (R²: 0.88, MAE: 18.11 mm, RMSE: 31.87 mm). This study provides a novel framework for long-term precipitation estimates at high spatial resolution and has great potential for enhancing the utilizability of satellite-derived precipitation.