Assessment of spatial and temporal variations in precipitation using mixing methods based on multiple precipitation products on the Chinese Loess Plateau

In the intricate and precise natural process of the hydrological cycle, precipitation plays a pivotal driving role as the core link in the mutual transformation between surface water and atmospheric water. From a hydrological perspective, precipitation is not only the most active input in the hydrological cycle but also the primary driving force behind watershed hydrological processes. Its spatiotemporal distribution characteristics directly influence the dynamic processes of hydrological elements such as surface runoff, infiltration, and evapotranspiration. Accurate precipitation data is crucial for advancing hydrological studies, supporting agricultural practices, improving flood prediction, enabling effective drought monitoring, and formulating strategies to adapt to climate change. However, the Loess Plateau region, with its complex terrain and strong spatial heterogeneity, Together with the sparse and unequal distribution of rainfall ground stations, makes large-scale and accurate studies on precipitation characteristics challenging. This study concentrated on the Loess Plateau, employing remote sensing and reanalysis datasets to derive spatially distributed precipitation data for the targeted area. A comparison was conducted among nine precipitation datasets against observed data to evaluate the feasibility of each product. Furthermore, the precipitation estimation capabilities of two mixing methodologies were analyzed. The analysis reveals that the GPM dataset delivers the most accurate outcomes, whereas both MERRA-2 and ERA5_Land consistently overestimate precipitation values. The incorporation of mixing techniques enhanced the precision of precipitation estimates; notably, the maximum R ( R max ) method produced interannual precipitation predictions with a maximum deviation of merely 10 mm from observed values, thus demonstrating its accuracy in interannual precipitation estimation within the Loess Plateau. Additionally, the Bayesian Model Averaging (BMA) method outperformed individual datasets in both spatial and temporal analyses. This method effectively estimates precipitation in the Loess Plateau by transforming site-specific weight values into grid formats. According to BMA estimates, average annual rainfall from 2001 to 2018 was around 445.2 mm, with a linear annual increase of 2.79 mm throughout this timeframe. Understanding the spatial heterogeneity characteristics and dynamic variation patterns of spatiotemporal precipitation is of significant theoretical and practical importance for formulating scientific soil and water conservation strategies and ecological restoration plans. In the Loess Plateau, a typical ecologically fragile region, precipitation, as a crucial hydro-meteorological element, directly influences the formation of surface runoff, the process of soil erosion, and the potential for vegetation recovery through its spatial and temporal distribution patterns.