Multi-Parameter Neural Network for Altimeter Tropical Cyclone Wind Speed Estimation

Abstract

The ability of satellite altimeter to estimate wind speed in tropical cyclone condition has been investigated. In the extreme condition with higher spatio-temporal variation, the ocean-atmosphere interaction is very complex and makes the existing algorithm become an ill-posed solution. In such condition, the developed algorithm from single frequency backscatter and significant waves height were insufficient. Besides, wind speed estimates become saturated at high regimes and the reflected backscatter was contaminated by rain. Therefore, other simultaneously observed parameters are needed to comprehensively account for this condition and is expected to improve the accuracy of wind speed retrieval. Aside from altimeter instrument, the microwave radiometer onboard Jason-2 concurrently records the brightness temperature and the rain information. To accommodate related multiple parameters for wind speed derivation, the neural network approach is proposed. Its unique advantage is relationship among multi-parameters can be easily established without prior knowledge on their physical attributes. Therefore, this study intended to determine the multi-parameter neural network (MPNN) model in estimating altimeter wind speed during the tropical cyclone condition. The results proved that the MPNN technique has potential in reducing the root mean square error by 30% in comparison between tropical cyclone wind speed estimate by the existing algorithm.