Answers are grounded in this paper and cite their sources.
The China–France Oceanography Satellite scatterometer (CSCAT) provides up to 16 multiangular normalized radar cross-section (NRCS) slices, which serve as rich information sources for high-quality ocean surface wind retrieval. However, rainfall can alter backscattering characteristics and consequently degrade the accuracy of wind retrieval. Therefore, accurate rainfall identification is of critical importance. In this study, various machine learning models integrated with different intelligent optimization algorithms are developed for CSCAT rainfall identification. Through systematic comparative evaluation, the optimal models are identified. Furthermore, the impacts of the number of valid NRCS slices, observational position within the swath, and polarization on the performance of the optimal models are investigated. In addition, the effects of classification probability threshold, rain-rate threshold, and CSCAT measurement instability on performance are discussed to support practical applicability. The results demonstrate that the eXtreme Gradient Boosting model optimized by Ant Colony Optimization (ACO-XGB) and the light gradient boosting machine model optimized by particle swarm optimization (PSO-LGBM) achieve better performance among all candidate models. Specifically, ACO-XGB exhibits superior discriminative capability (AUC = 0.9382) and a lower false positive rate (FPR = 0.0368), whereas PSO-LGBM shows better performance in reducing the false negative rate (FNR = 0.1973). By fully exploiting the multiangular and dual-polarization observations of CSCAT, the proposed models achieve effective rainfall identification and maintain robust performance under varying observational conditions, rendering them suitable for operational applications.
No takes yet. Share an insight, caveat, or question.
Song et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: