|其他摘要||Precipitation is a key climate variable of global water and energy circulation. And it also is an indispensable input parameter of water calculation, water resources assessment, climatic analysis, hydrological models research. Thus, there is great importance of high resolution and precision precipitation data. At present, in the Western China, the hydrometeorology observation stations are so spare that not satisfy the application requirements of watershed hydrological model, especially in the Tien mountainous with complicated topography and high altitude. Currently, the remote sensing technology could provide continuous distribution of precipitation data in space. And the TRMM data has already been widely used, which become an important method to study the precipitation variability. However, the low resolution and data accuracy problem of TRMM seriously restricted its wide application. While downscaling and correction technology to the remote sensing precipitation data can effectively solve the problem.
This study is taking Tianshan mountainous as the research object. First, we validated the applicability of TRMM3B43 precipitation data with measured data of eight meteorological stations in Tianshan mountainous; Then, corrected TRMM3B43 data based on different methods (ANN and CART) to get the high-precision precipitation data; Finally, according to the relation between TRMM3B43 precipitation and vegetation NDVI, slope, aspect, elevation, latitude and longitude, established downscaling regression model by CART algorithm. Using high resolution and precision precipitation data by secondary correction to analyze the characteristics of the spatial distribution of precipitation in Tianshan mountainous. Main conclusions are listed as follows:
(1) TRMM3B43 precipitation data has the basic applicability in Tianshan mountainous. In the study area, the determination coefficient (R2) between monthly precipitation data of TRMM3B43 and GPM and observed station were 0.76 and 0.87. There is a significant linear correlation relationship, illustrating that the precipitation data between TRMM3B43, GPM and observed station have good consistency on the whole. The precipitation data of TRMM3B43 and GPM can reflect the approximate precipitation in the study area. While, in a single station, the correlation between monthly precipitation data of TRMM 3B43, GPM and observed station is not high, and the error is large.
(2) According to the results of ANN and CART correlation algorithm, we know that: Whether on the whole or single station, the average of R2 between TRMM and observed precipitation data is more than 0.7 after correlation by comprehensively considering geographical factor and NDVI data. δ and RMSE are within the acceptable range. Illustrating that it’s a well effect that adjust TRMM data by two methods of ANN and CART. The fitting degree between the precipitation data of TRMM after correlation by CART algorithm and the measured values is higher than ANN. Compare with ANN, the precipitation data of TRMM after correlation by CART algorithm is closer to the measured values. Which can determine the effect of correction of CART algorithm is better than the ANN method, and CART calculation process is more simple and efficient.
(3) The method that based on CART algorithm to establish a regression model which considered NDVI, gradient, slope direction, vertical and longitude and latitude to downscale TRMM precipitation data is feasible. We can obtain annual precipitation data of resolution of 1 km of Tianshan mountainous through the downscaling and correction. Through the inspection of: Downscaling process improves the resolution of spatial, but with a certain error between the precipitation data of observed station. The monthly precipitation data is obtained by proportional coefficient method of decomposition. The determination coefficient (R2) between the downscaling of years of average precipitation data and measured precipitation data is 0.85, δ and RMSE were 15.20% and 53.72mm respectively. Compared with the unadjusted R2 increased 0.05, δ and RMSE reduced 5.42% and 10.58 mm respectively. Thus: the precision of downscaling precipitation data after correction is higher, and the method of correction reduced the error between downscaling and observed precipitation data effectively. Therefore, this research can get the precipitation data of high resolution and precision of Tianshan mountainous by the method of downscaling and correction.
(4) Use the downscaling precipitation data after correction to analyze the precipitation in three time scales of month, season and year in Tianshan mountainous. The spatial distribution shows: The most precipitation in Tianshan mountainous all the year is located in western, and the precipitation decreased gradually from the west to the east in Tianshan mountainous. The precipitation mainly concentrated in the central Tianshan mountainous in summer. The precipitation in the western area is more than the east in winter, and the precipitation in northslope of Tianshan mountainous is slightly higher than southslope. From the change of the time view: The precipitation is large in summer and small in winter in Tianshan mountainous, which seasonal is apparent. Among them, July is the largest and February is the least.|