Answers are grounded in this paper and cite their sources.
Global navigation satellite system reflectometry offers a promising means for ocean rainfall remote sensing. Based on a spatiotemporally matched dataset constructed from cyclone global navigation satellite system forward scattering observations, GPM IMERG rainfall rate products, and ERA5 wind speed data, this study systematically evaluates five machine learning models (BPNN, Random Forest, XGBoost, LightGBM, and CatBoost) coupled with the Optuna hyperparameter optimization framework and two data balancing strategies for rainfall identification. Using the optimal model identified, a rainfall intensity retrieval model is further developed and evaluated. The results show that for rainfall identification, the Optuna optimized XGBoost model with random undersampling achieves the best performance, attaining an accuracy of 95%, a Recall of 87%, an area under the receiver operating characteristic curve of 0.98, and a false positive rate of 3%, substantially outperforming existing convolutional neural network models under moderate to high wind speeds; for intensity retrieval, the XGBoost regression model achieves a root-mean-square error (RMSE) of 0.628 mm/h, mean absolute error (MAE) of 0.280 mm/h, and Bias of 0.010 mm/h, reducing RMSE by 22% and MAE by 10% compared with prior RF-based approaches. In addition, the proposed models demonstrate satisfactory spatiotemporal generalization across global oceanic regions. Limitations include degraded performance under light rainfall, extreme wind speeds, and heavy precipitation. Future efforts will focus on targeted model enhancements to address these scenarios.
No takes yet. Share an insight, caveat, or question.
Deng et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: