Downscaling and fusion of satellite products: A case study of Lantsang River Basin

Satellite precipitation products have been one of the most important technologies with wide coverage more suitable for areas without data. However, their performance cannot fully meet the harsh requirement of the high resolution and precision precipitation data in recent years. It is very necessary to downscale and then integrate the satellite products with the ground observation data for better data quality. In this study, the Lantsang River Basin in the southern Tibetan Plateau was taken as an example, particularly considering topographic, geographic and vegetational elements. The geographically weighted regression (GWR) model was established with the spatial downscaling data of the tropical rainfall measurement mission (TRMM) satellite and the precipitation estimation from remotely sensed information using artificial neural networks-climate data record (PERSIANN-CDR). The GWR downscaling model was supposed to improve the correlation accuracy between satellite products and ground observation precipitation data. After that, the ensemble Kalman filter was used to take the inverse distance weighted (IDW) interpolated data of the ground meteorological station as the observed values of the fusion and then fused the downscaling TRMM and PERSIANN-CDR data to further improve the accuracy of precipitation data. The results show that: 1) The mean value of determination coefficient（R²） of PERSIANN-CDR monthly precipitation increased from 0.35 to 0.75 after GWR downscaling. At the same time, the root mean square error (RMSE) and mean absolute error (MAE) decreased by 13.98 and 10.13 mm, respectively. There was a significant increase in the correlation degree of the PERSIANN-CDR satellite precipitation product in all months after GWR downscaling. Meanwhile, the R² of TRMM monthly precipitation increased from 0.53 to 0.85, and the RMSE and MAE decreased by 11.49 and 15.50 mm, respectively. A significant improvement was achieved in the months with the low correlation degree for the surface meteorological stations before downscaling, such as May, June, and December, where the R² reached 0.67 or above after downscaling. In addition, the two types of products presented the more significant effects on the accuracy evaluation in the dry season (from November to April), compared with the rainy season (from May to October). It infers that the GWR greatly improved the monitoring performance of these two types of satellite precipitation products on precipitation in the dry season. 2) The accuracy was improved better than before after the data integration and downscaling from the ground stations. Furthermore, the ensemble Kalman filter was used for the data fusion of down-scaled products. The overestimation of precipitation was enhanced at ground meteorological stations by satellite products, especially with the less uncertainty of the data after fusion, indicating the high precision fusion. In summary, the downscaling and fusion can be expected to increase the spatial resolution and accuracy of data. The high spatial resolution of satellite products was also achieved in the high correlation with the precipitation data observed on the ground.