Improving daily precipitation estimation using a double triple collocation-based (DTC) merging framework

Abstract

The availability of accurate precipitation data is crucial for water resources management, disaster prevention, and related research. While gridded products offer precipitation information at high spatial resolution, they still exhibit significant errors in precipitation estimation. The merging of multi-source gridded products has become a mainstream approach for improving precipitation estimation. However, many existing frameworks rely on gauge observations to estimate the merging weights, which limits their applicability in data-scarce regions. Moreover, these frameworks predominantly focus on enhancing precipitation estimation rather than on precipitation events. This study proposes a novel Double Triple Collocation-based (DTC) merging framework, which combines time–space TC (TC_2D)-based precipitation rate merging with categorical triple collocation (CTC)-based rain/no-rain merging. The objective is to minimize errors in precipitation estimation and enhance the detection capability for precipitation events without relying on rain gauge observations. Given that the TC_2D-based method is initially applied to precipitation merging, its effectiveness must be verified by comparing it with classic TC-based merging approaches (TC_Space and TC_Time). Taking the Jiulong River Basin (JRB) as a case study, the performance of the DTC and its comparative objects was evaluated with three triplets composed of independent precipitation products. The results indicated that all merged precipitation products outperform their parent products. Furthermore, the merged precipitation datasets, after being corrected using rain/no-rain time series generated by CTC-based merging, showed enhanced capability in detecting precipitation events. The performance of merged precipitation products was found to be highly dependent on the quality of the satellite precipitation products (SPPs) within the triplets. This study provides a promising approach for generating high-quality precipitation datasets, particularly in regions with limited observation data availability.