A flexible two-stage approach for blending multiple satellite precipitation estimates and rain gauge observations: an experiment in the northeastern Tibetan Plateau

Abstract. Substantial biases exist in the Satellite Precipitation Estimates (SPE) over complex terrain regions and it has always been a challenge to quantify and correct such biases. The combination of multiple SPE and ground observations would be beneficial to improve the precipitation estimates. In this study, a flexible two-step approach is proposed by firstly reducing the systematic errors of each SPE using rain gauge observations as references, and then merging the improved multi-SPE with a Bayesian weighting model. In the 1st stage, gauge references are assumed as a generalized regression function of SPE and terrain feature. In the 2nd stage, the weights assigned to the involved SPE are calculated according to the associated performance relative to gauge references. This blending method has the ability to exert benefits from multi-SPE in terms of higher performance and mitigate negative impacts from the ones with lower quality. In addition, Bayesian analysis is applied in the two phases by specifying prior distributions on the model parameters, which enables to produce posterior ensembles associated with their predictive uncertainties. The performance of the two-step blending approach is assessed using independent rain gauge observations during the warm season of 2014 in the northeastern Tibetan Plateau. Results show that the blended multi-SPE are significantly improved compared to the original individuals, especially during heavy rainfall events. This study can also be expanded as a data fusion framework in the development of high-quality precipitation products in high-cold regions characterized by complex terrain.