Merging Multisatellite and Gauge Precipitation Based on Geographically Weighted Regression and Long Short-Term Memory Network

To generate high-quality spatial precipitation estimates, merging rain gauges with a single-satellite product (SPP) is common approach. However, single SPP cannot capture the pattern of well, and its resolution also too low. This study proposed an integrated framework for multisatellite gauge precipitation. The integrates geographically weighted regression (GWR) improving estimations long short-term memory (LSTM) network estimation accuracy by exploiting spatiotemporal correlation between products gauges. Specifically, was applied to Han River Basin China generating daily estimates from data both four SPPs (TRMM_3B42, CMORPH, PERSIANN-CDR, GPM-IMAGE) during period 2007–2018. results show that GWR-LSTM significantly improves (resolution 0.05°, coefficient 0.86, Kling–Gupta efficiency 0.6) over original 0.25° or 0.1°, 0.36–0.54, 0.30–0.52). Compared other methods, whole basin improved approximately 4%. Especially in lower reaches River, 15%. In addition, this demonstrates multiple-satellite much better than partial multiple satellite observations.