A Genetic Algorithm–Assisted Deep Neural Network Model for Merging Microwave and Infrared Daily Sea Surface Temperature Products

Abstract

Sea surface temperature (SST) is an important factor in the global ocean–atmosphere system, being vital in a variety of climate analyses and air–sea interaction research studies. However, estimating daily SST with both high precision and high spatial completeness remains a challenge. This article attempts to solve this problem by merging two complementary daily SST products, that is, the 25 km-resolution Advanced Microwave Scanning Radiometer for EOS (AMSR-E) SST and 4 km-resolution Moderate Resolution Imaging Spectroradiometer (MODIS) SST, using a genetic algorithm–assisted deep neural network model (GA-DNNM). The merged SST with a spatial resolution of 4 km and a temporal resolution of 1 day is achieved. Experiments in the Asia and Indo-Pacific Ocean (AIPO) region in 2005 were conducted to demonstrate the feasibility and advantages of the proposed method. Results showed that the spatial coverages of the original MODIS SST and AMSR-E SST are ranging from 25.0 to 48.1%, and 31.5 to 47.6%, respectively, while the merged SST achieves a spatial coverage ranging from 56.1 to 73.1%, with improvements ranging from 50.2 to 131.7% relative to the original MODIS SST. Comparisons with drifting buoy observations indicate that the merged SST is accurate, with an average bias of 0.006°C and an average RMSE of 0.502°C, in places where the MODIS SST data are missing before being merged in the AIPO area, and with an average bias of −0.082 °C, and an average RMSE of 0.603°C for the merged SST in the whole study area.