Reconstruction of MODIS Daily High Resolution Sea Surface Temperature over South Eastern Arabian Sea for Near Real Time Applications

Sea Surface Temperature (SST) derived from infrared radiometer based satellites often suffer hindrance due to cloud cover, resulting in huge data loss and poor data quality. Arabian sea, located in the tropics, is one of the cloudiest regions in the globe. The intense cloud cover in the region restricts the availability of high resolution SST data for near real time regional oceanographic applications. In order to overcome this data gap, the current study tests the Data INterpolating Empirical Orthogonal Functions (DINEOF) for the first time in South Eastern Arabian Sea to reconstruct daily high resolution SST. A decade long analysis has been conducted using level 2 Moderate Resolution Imaging Spectroradiometer (MODIS) Aqua datasets of 1 km resolution by adopting monovariate approach. The reconstructed SST dataset have been validated with respect to in situ reference data collected by Centre ERS d'Archivage et de Traitement (CERSAT) for the same period. Further, the DINEOF based reconstructed images were compared with cloud free SST estimates generated by support vector regression (SVR) based algorithm developed by Sunder et al., 2019.

The results of DINEOF approach were found to have coefficient of determination of 0.62 and an RMSE of 0.79⁰C for the study period. From the initial results, it is observed that SVR based SST estimates are better than DINEOF based SST estimates, in terms of correlation and RMSE. Detailed analysis is being carried out to find the ideal number of images required to achieve high accuracy through DINEOF with least error for near real time applications.

Reference

Sunder, Swathy, Ramsankaran, RAAJ, Ramakrishnan, Balaji (2019). “Machine Learning Techniques for Regional Scale Estimation of High-Resolution Cloud-Free Daily Sea Surface Temperatures from MODIS Data”. ISPRS Journal of Photogrammetry and Remote Sensing (Under review)