Ankur Garg, Aatmaj Salunke

This research investigates the efficacy of deep learning techniques in estimating Land Surface Temperature (LST) and Emissivity from Landsat satellite imagery across seven distinct geographical regions. Utilizing the Single Channel Method for LST estimation and an NDVI-based approach for Emissivity estimation, our study spans the years 2018 to 2023, ensuring data integrity with cloud cover below 10%. We meticulously calibrated radiometric values and curated combined datasets for training Pix2Pix models, subsequently evaluating their performance using robust metrics. Our findings demonstrate the effectiveness of this approach in accurately predicting LST and Emissivity, even on unseen data, with adept handling of boundary null values during image stitching. The results showcase the potential of deep learning models in remote sensing applications, contributing
to improved land surface monitoring and environmental assessments. This research underscores the importance of integrating advanced computational techniques with earth observation data for enhanced insights into climate dynamics and land surface processes.