Refine the Uncertainty of GPM IMERG Precipitation Product Accounting for the Inherent Error From Rain Gauges Estimations

Abstract

Satellite precipitation retrieval accuracy assessment requires reliable ground validation, yet conventional approaches using rain gauges as “truth” neglect representativeness errors inherent in point-to-area approximations. This study uses 7,253 rain gauges (2020–2024) over the Jianghuai monsoon region to quantify these errors and reassess Integrated Multi-satellite Retrievals for GPM (IMERG) performance. We show that at least 16 gauges per 0.2° grid are required for reliable area-mean precipitation estimates. Analysis reveals dual dependence of gauge representativeness errors on gauge density (n, number of gauges per grid cell) and rainfall intensity (RR): (a) errors decay exponentially with increasing n, following root mean square error (RMSE) = ae^−bn, where a and b are fitted coefficients; (b) errors increase with RR when n is held constant. Parameterized relationships enable error quantification across density gradients. Direct IMERG-gauge comparisons show that seasonal mean differences are negatively correlated with gauge density (Pearson's r = −0.33, p < 0.01), indicating that sparse gauge networks are a primary driver of apparent discrepancies. Error decomposition using gauge uncertainties yielded bounded IMERG retrieval errors (RMSE^B_min/max). Applying the same framework to Kling-Gupta efficiency (KGE) revealed similarly improved IMERG performance after removing gauge-induced uncertainties, reinforcing the internal consistency of our analysis. Crucially, incorporating gauge errors reduced significant discrepancy frequency by 16%/6%/16%/17% across seasons, proving that traditional methods overestimate IMERG-gauge deviation occurrence by 6%–17%. This establishes gauge density as critical accuracy determinant, provides robust error-quantification framework, and reveals that terrain-complexity misinterpretations arise when disregarding representativeness errors, with implications for global satellite precipitation validation.