Application of Deep Neural Network in Estimating Sea Surface Temperature from INSAT-3D Imager

Presented at the GHRSST XXIII international science team meeting, 27 June-1 July 2022, online and in-person (Barcelona). #GHRSST23

Short abstract

Sea surface temperature (SST) is one of the important essential climate variables. The continuous monitoring of the SST is required to study the Earth’s climate change. Additionally, SST information is necessary to understand many oceanic processes and meteorological events. In the present study, a machine learning technique based on deep neural network (DNN) is exploited to estimate the SST from infrared Imager on-board India’s geostationary satellite INSAT-3D. To establish the DNN, a matchup dataset is prepared by collocating the split-window observations of INSAT-3D Imager and in-situ measurement of SST for the years 2017-2020. Further, 70% of the matchup data is randomly selected for training the DNN, whereas, the rest 30% is used for testing. The assessment of the trained DNN is performed in terms of the standard statistical quality indicator viz., bias and root-mean-squared error (RMSE), etc.  A negligible bias with RMSE of ~0.5K is observed in both the training and testing datasets. To examine the robustness of the developed DNN, it is further applied on the independent dataset of January 2021 of INSAT-3D Imager and is validated against the in-situ SST measurements.  The validation shows the RMSE of ~0.6K in the estimated SST throughout the month.