We present a new approach, based on a multi-parameter back-propagation neural network (BPNN) model, to simultaneously retrieve sea surface wind speed, sea surface temperature, near-surface air temperature, and dewpoint temperature over the global oceans from the Advanced Microwave Scanning Radiometer 2 (AMSR2) onboard the Global Change Observation Mission 1st-Water (GCOM-W1). The model is trained and validated with the collocations of AMSR2 multi-channel (6.9–36.5 GHz) brightness temperatures, under both clear and cloudy conditions, and National Data Buoy Center (NDBC) and Tropical Atmosphere Ocean Project (TAO) buoy measurements along with ECMWF ERA5 reanalysis data. The root-mean-square (rms) errors of BPNN-retrieved sea surface wind speed, sea surface temperature, near-surface air temperature, and dewpoint temperature are 1.13 m/s, 1.02 °C, 1.20 °C, and 1.57 °C, respectively. The first three retrieved geophysical parameters and the estimated relative humidity from near-surface air temperature and dewpoint temperature are used to compute the sensible heat flux (SHF) and latent heat flux (LHF), using an improved bulk flux parametrization. The rms errors of the estimated SHF and LHF from BPNN-derived air–sea interface variables, and those from buoy and reanalysis data, are 18.13 W/m2 and 39.56 W/m2. We also compare SHF and LHF estimates with the Yongxing air–sea flux tower measurements in the northern South China Sea. The estimated SHF and LHF in summer and autumn periods are closer to observations than in winter and spring. The proposed method has potential to derive instantaneous air–sea interface atmospheric and oceanic parameters as well as surface sensible and latent heat fluxes from AMSR2 along-track wide swath observations.
Read full abstract