Enhancing the accuracy of Jason-3 PWV products over coastal areas using the back propagation neural network

Abstract

The performance of microwave radiometers aboard altimetric satellites in measuring water vapor degrades significantly over coastal areas due to the mixing of land within its footprint. In this study, we propose using the back propagation neural network (BPNN) models to enhance the accuracy of Jason-3 precipitable water vapor (PWV) over coastal areas. PWV data from 2076 globally distributed coastal and island Global Navigation Satellite System (GNSS) stations and 237 radiosonde stations are used as the reference. Specifically, the GNSS PWV data in 2016 and 2017 are used to train the BPNN models, while the GNSS and radiosonde PWV observations from January 2018 to June 2023 are used to test the performances of the BPNN models proposed. Our results show that the proposed BPNN PWV models can considerably enhance the accuracy of Jason-3 PWV recorded in the coastal areas (within 25 km of land). Evaluated by the GNSS PWV, BPNN models can reduce the root mean square error (RMSE) of Jason-3 PWV in the coastal areas from 4.2 to 2.7 kg/m2 (35.7% of RMSE reduction). Assessed by the radiosonde PWV, the results indicate that the RMSE of Jason-3 PWV in the coastal areas is decreased from 5.0 to 3.6 kg/m2 (28.0% of RMSE reduction) after using the proposed BPNN models.