Abstract: Precipitation data have became an indispensable part for agriculture, hydrological, meteorological, ecological and other environmental applications. Satellites obtain the earth's precipitation data from space through on-board sensors, which is playing a more and more important role in the data collection currently. Research increasingly suggests that satellite-derived precipitation products with their advantages in the continuity of spatial scale and high degree of prediction accuracy have vast space for development. It is well-known that use of the Tropical Rainfall Measuring Mission (TRMM) has been widely employed for obtaining global precipitation data recently due to its incomparable superiority to traditional method. However, the application is subject to certain restrictions by the relatively low spatial resolution (about 20?30km) of the data. Considering various influence factors such as spatial location and terrain and their spatial non-stationary characteristics, a case study on the application of mixed geographic weighted regression combined with Kriging interpolation (MGWRK) for spatial downscaling of the TRMM annual precipitation data was undertaken at Sichuan Province, Southwest China with a significant space differentiation of precipitation. And in the meantime, assessment of the downscaling results derived by different methods were carried out based on the measured data of 41 meteorological stations. Some results in this study showed that: (1) by use of the MGWRK model for downscaling, the spatial resolution of TRMM precipitation data was increased sharply from 0.25° (about 26km) to 1km, which can describe the spatial variation of precipitation more detailly and effectively in study area. (2) The MGWRK model not only attempted to use a combination of various auxiliary information with high-resolution such as spatial location and terrain, but also explored the characteristics of spatial stationary of the relationship between TRMM precipitation and its factors. From the assessment results of various downscaling approach to the TRMM data of mean annual values (1998?2017) and the two typical years’ values (the wet year at 1998 and the dry year at 2006), it was found that the MGWRK method can prove a higher accuracy compared with the OLS-based global regression Kriging (GRK) and the Bilinear resample (Bil) method and obtain a result that is more approximate to the original status. (3) The downscaling model presented in this paper considered the improvement of spatial resolution without compromising the maintaining accuracy and therefore it is obviously an approach available for the spatial downscaling of TRMM precipitation data in study area and contribute to define a foundation for the application of the TRMM data in small scale.

Key words: TRMM precipitation data, Spatial downscaling, Mix geographically weighted regression, Sichuan Province