A neural network to estimate the latent heat flux over oceans from satellite observations

Abstract A new method based on a neural network is proposed to infer the mesoscale instantaneous latent heat flux from satellite observations. The new method is compared to the bulk method on a global dataset. The rms error on the estimation of the flux is 35 W m -2 for the network and is 55 W m -2 for the bulk method. The estimation error is more variable at mid-latitudes than in the tropics and near the equator. An analysis of the error as a function of environmental conditions shows that the accuracy of satellite derived fluxes is better in low humidity and intermediate to strong wind conditions than in low wind or large humidity conditions. The two methods are validated using three regional datasets. The rms errors on these datasets range from 25 W m -2 to 45 W m -2, which is consistent with the results obtained at global scale.