Added value of merging techniques in precipitation estimates relative to gauge-interpolation algorithms of varying complexity

Data-fusion techniques leverage the strengths of multisource precipitation data and can significantly enhance the accuracy of precipitation estimates. However, the extent to which these techniques improve precipitation estimates (i.e., added value) compared to interpolation algorithms and the factors driving this improvement remain unclear. To address these gaps, this study compared the performance of two merging techniques, i.e., double machine learning (DML) and geographically weighted regression (GWR), with multiple interpolation algorithms in estimating precipitation across China. The interpolation algorithms vary in complexity and include typical methods (IDW and Kriging), semi-physical methods (GIDS, DAYMET, and MicroMet), and climatologically aided interpolation (CAI). We quantified the added value of the merging techniques over these interpolation algorithms and investigated the driving factors using a data-driven approach. Results indicate that the merging techniques outperform all the interpolation algorithms, regardless of their complexity. The merging techniques provide greater added value in gauge-scarce regions (e.g., Northeast China) than in gauge-rich regions (e.g., Northwest China). The magnitude of the added value from merging techniques is significantly influenced by the choice of interpolation algorithms due to their varying performance. Additionally, our data-driven model reveals that factors such as the amount of precipitation, number of wet days, performance of precipitation products, and gauge density are key drivers that negatively affect the added value of merging techniques. This study highlights the importance of integrating multisource data to improve precipitation estimates, especially in regions with sparse gauges, rather than relying solely on gauge-only interpolation.