Conditional Evaluation of FY-3E WindRAD Data and ERA5 Reanalysis Fusion for Short-Term Typhoon Wind Vector Prediction

Typhoon wind prediction is important for disaster prevention. ERA5 reanalysis provides spatially continuous wind fields but may smooth local structures, whereas FY-3E WindRAD, a dual-frequency scatterometer, observes ocean surface winds through C-band and Ku-band measurements with sparse coverage and temporal mismatch. This study evaluates the conditional value of fusing FY-3E WindRAD data with ERA5 for 12-h short-term typhoon wind vector prediction using matched 2023 samples. We compared ERA5-only, naive, mask-aware, and gated fusion across coverage-sensitivity, band-comparison, and temporal-window experiments. Results show that FY-3E fusion provides conditional numerical benefits rather than consistent improvements. Under the band-specific evaluation, Ku-band gated fusion reduced vector root mean square error (RMSE) from 5.736 to 5.366 m/s and average wind speed error (AWSE) from 3.597 to 3.276 m/s, although these improvements were not statistically significant after multiple-comparison correction. Dual-band gated fusion did not outperform the ERA5-only baseline, and under the common-sample evaluation neither Ku-band nor Dual-band showed a clear advantage. The ±3 h window provided a more favorable balance between sample availability and temporal consistency than the wider ±6 h window. Overall, FY-3E WindRAD data provides complementary information, but its effectiveness depends on coverage, band selection, temporal matching, and fusion strategy.