资源环境科技发展态势分析平台

Very high-resolution (approximate to 1 km) precipitation datasets are needed in various applications, especially in the hydrological, environmental, agricultural, and biological fields, although they also have major roles in the climate sciences. Nevertheless, various studies have concurred that the spatial distribution of precipitation is influenced by land surface characteristics through non-stationary relationships, and that this characteristic is generally not considered fully by either global or local models using fixed combinations of variables (e.g. the normalized difference vegetation index and/or digital elevation models). This study introduced a new spatial downscaling algorithm, a divide-and-conquer method (called Cubist), to downscale Tropical Rainfall Measuring Mission (TRMM) satellite precipitation data over China from 2000 to 2015. In this endeavour, we considered the non-stationary relationships between precipitation and land surface characteristics at various scales and derived the following conclusions: (1) The relationships and related variables explaining the spatial distribution of precipitation varied with spatial scale. (2) The Cubist algorithm achieved greater accuracy than Multivariate Linear Regression (MLR) models, and it was found to compensate for the disadvantages of MLR overestimation and underestimation at low and high values, respectively. (3) In downscaling, the monthly Version 7 TRMM Multisatellite Precipitation Analysis (TMPA) datasets over China, the entire study area was separated into subregions driven by spatial datasets according to geographic similarities, which were similar to Chinese rainfall zones.