Estimation of ocean subsurface thermal structure from surface parameters : a neural network approach

Abstract

Satellite remote sensing provides diverse and useful ocean surface observations. It is of interest to determine if such surface observations can be used to infer information about the vertical structure of the ocean's interior, like that of temperature profiles. Earlier studies used either sea surface temperature or dynamic height/sea surface height to infer the subsurface temperature profiles. In this study we have used neural network approach to estimate the temperature structure from sea surface temperature, sea surface height, wind stress, net radiation, and net heat flux, available from an Arabian Sea mooring from October 1994 to October 1995, deployed by the Woods Hole Oceanographic Institution. On the average, 50% of the estimations are within an error of ±0.5°C and 90% within ±1.0°C. The average RMS error between the estimated temperature profiles and in situ observations is 0.584°C with a depth-wise average correlation coefficient of 0.92.