<p>Satellite data have become an integral part of managing natural resources and understanding and monitoring various environmental processes at diverse scales. But the problem is with their spatial and temporal resolutions. Some of the important phenomena, like drought, forest fire, and plant health, require thermal data, but these data are available at coarser resolutions, though at high temporal resolutions. Recently, a lot of work has been done on deriving high spatial thermal data using statistical and downscaling methods through multi-sensor fusion. However, despite these advancements, there is still a critical need to systematically assess and unify recent artificial intelligence (AI)-based downscaling approaches to enhance spatial accuracy, computational efficiency, and generalization across varying environmental conditions. This study aims to review the problem, issues, existing techniques, and progress in the domain of downscaling. Recent availability of machine learning models like ANN, CNN, and RNN appears to be promising as they are efficient in finding patterns from the non-linearly correlated data. These tools are powerful for applications in the field of agriculture, forestry, disaster response, and environmental monitoring. In this regard, the current work aims to provide the reader with detailed information about the downscaling problem, prerequisites, accuracy, uncertainty, and processing requirements for various sensors. Integrating physical models with AI frameworks, embedding edge computing capabilities for increasing the real-time processing speed, and providing benchmark datasets are some of the interesting possibilities being addressed in this study.</p>