In July 2021, an extreme precipitation event occurred in Henan, China, causing tremendous damage and deaths; so, it is very important to study the observation technology of extreme precipitation. Surface rain gauge precipitation observations have high accuracy but low resolution and coverage. Satellite remote sensing has high spatial resolution and wide coverage, but has large precipitation accuracy and distribution errors. Therefore, how to merge the above two kinds of precipitation observations effectively to obtain heavy precipitation products with more accurate geographic distributions has become an important but difficult scientific problem. In this paper, a new information fusion method for improving the position accuracy of satellite precipitation estimations is used based on the idea of registration and warping in image processing. The key point is constructing a loss function that includes a term for measuring two information field differences and a term for a warping field constraint. By minimizing the loss function, the purpose of position error correction of quantitative precipitation estimation from FY-4A and Integrated Multisatellite Retrievals of GPM are achieved, respectively, using observations from surface rain gauge stations. The errors of different satellite precipitation products relative to ground stations are compared and analyzed before and after position correction, using the '720' extreme precipitation in Henan, China, as an example. The experimental results show that the final run has the best performance and FY-4A has the worse performance. After position corrections, the precipitation products of the three satellites are improved, among which FY-4A has the largest improvement, IMERG final run has the smallest improvement, and IMERG late run has the best performance and the smallest error. Their mean absolute errors are reduced by 23%, 14%, and 16%, respectively, and their correlation coefficients with rain gauge stations are improved by 63%, 9%, and 16%, respectively. The error decomposition model is used to examine the contributions of each error component to the total error. The results show that the new method improves the precipitation products of GPM primarily in terms of hit bias. However, it does not significantly reduce the hit bias of precipitation products of FY-4A while it reduces the total error by reducing the number of false alarms.
Keywords: data fusion; extreme precipitation; image registration; position correction; satellite precipitation estimation.