Adaptability of precipitation estimation method based on TRMM data combined with partial least squares downscaling in different landforms of Xinjiang, China

High-resolution precipitation data is beneficial to describe the spatial-temporal differentiation characteristics of regional precipitation. The purpose of this study was to explore the feasibility of using partial least squares downscaling method to improve the spatial resolution of Tropical Rainfall Measuring Mission’s (TRMM) data for estimating precipitation in different landform regions of Xinjiang, China (73°40′-96°23′E, 34°22′-49°10′N). The spatial resolution of remote sensing data was increased from 0.25°×0.25° to 250 m×250 m based on multi-source remote sensing data and meteorological station data, by introducing environmental factors such as relative humidity, topography, and latitude and longitude, and constructing a partial least squares downscaling model. The accuracy of precipitation estimation before and after downscaling was compared, and the temporal and spatial distribution characteristics of precipitation in Xinjiang and its responses to topography and geomorphology were discussed. The results showed that the original TRMM could estimate precipitation well for the whole Xinjiang and mountain and plain areas of Xinjiang. However, the estimation accuracy in basin was low. Compared with that of the original TRMM value, the accuracy of precipitation estimation was improved by the partial least squares downscaling model with the coefficient of determination (R2) increased from 0.74 to 0.85 and the root mean square error (RMSE) decreased by 0.26 mm. Moreover, the estimation after downscaling solved the problems of overestimation in the areas with low precipitation values and underestimation in the areas with high precipitation values. By downscaling method, the accuracy of precipitation estimation in stations at different altitudes and topography in the Urumqi River Basin was improved greatly. Compared to that before downscaling, the R2 increased from 0.06-0.91 to 0.39-0.95 and the RMSE decreased from 0.20-0.44 mm to 0.18-0.40 mm. It indicated that the downscaling method used for TRMM data was reliable in estimating precipitation in different altitudes and topography. The annual precipitation in different landform areas of Xinjiang was the highest in the areas with medium high mountain, followed by those with extremely high mountain, high mountain, low mountain, medium mountain, plain and basin. The ratio of area of basin, plain and mountainous was about 1:1.6:2.5. The precipitation in the areas with medium high mountain mainly occurred from June to September. The rich precipitation in low-altitude mountain areas and plain areas was from May to August and during this period the precipitation distribution was relatively uniform. For example, the months with the least precipitation in low mountain areas were only 10% less than the months with the most precipitation. The spatial distribution pattern of precipitation in Xinjiang was characterized by more in the north and less in the South. The multi-year average monthly precipitation in the northern of the Altai Mountains was higher than 20 mm, while the precipitation in the southeast of Tarim Basin and Tu-Ha Basin was little. In the study area, two obvious precipitation peak areas in each mountain (mountain group) were observed, however the height and scale of each mountain were different. Thus, the altitude of the peak area was different. The results can provide a valuable method to estimate regional precipitation for areas with scarce meteorological stations.