Abstract: The precipitation in mountain area has important research value. However, the uneven distribution and the lack of the rain gauge stations in mountain area result in the difficulties in the study of precipitation which is related to the ecological, hydrological and meteorological processes. High-resolution satellite rainfall data provides a great opportunity to monitor precipitation frequently over large and remote areas. The Tropical Rainfall Measuring Mission (TRMM)is one of these products and can provide reliable precipitation data for regional hydrology study at the monthly and yearly scales. Unfortunately, in some small region, especially the mountain region with abundant rainfall and complex terrain, the TRMM rainfall product cannot be used directly because of its coarse resolution (0.25°). Therefore, it is very necessary to improve its resolution. In this paper, taking Shaanxi Qinling-Daba Mountains which were further divided into four different terrain zones as the study area, the relationship between TRMM and NDVI explored by geographically weighted regression (GWR)was used to construct the precipitation downscaling model, it could produce 1 km downscaled precipitation data in yearly time scale; then, selecting year 2010 as a typical case, the monthly downscaled precipitation was obtained from the annual downscaled precipitation data according to the proportion of the monthly precipitation over annual precipitation in TRMM; finally, the effects of topography on the reliability of the downscaled results were evaluated. In the study, the accuracies of the TRMM precipitation and GWR downscaled precipitation were evaluated by using a network of 23 rain gauges over Shaanxi Qinling-Daba Mountains, and the R, RMSE and BIAS were calculated to evaluate the data. The results show as follows: (1)the downscaled precipitation data improves the spatial resolution (from 0.25° to 1 km), which can provide more details about the precipitation; (2)the R of annual downscaled precipitation data is 0.88, and at monthly scale, the R is 0.93, the GWR downscaled precipitation has a good correlation with the data of rain gauges and the result is reliable; (3)the value of the downscaled precipitation is smaller than that of the TRMM data, and the downscaled result underestimates the precipitation in most cases; (4)the variation degree of the accuracy of the downscaled precipitation is the smallest in the Qinling Mountains where the altitude is the highest in the study area; (5)under similar topographical conditions, GWR downscaled results perform better at higher altitude. The GWR model can be used to the spatial downscaling of the TRMM rainfall product in Shaanxi Qinling-Daba Mountains, the downscaled results could provide higher spatial resolution precipitation for the study and application of regional hydrology.

Key words: TRMM precipitation data, NDVI, GWR model, Downscaling, Shaanxi Qinling-Daba Mountains