A Systematic Review of Spatiotemporal Deep Learning Approaches for Drought Monitoring and Forecasting using Remote Sensing Data
Ritu Khandelwal, Hemlata Goyal, Rajveer Singh ShekhawatIntroduction:
Drought is a prolonged period of abnormally low precipitation resulting in water scarcity, reduced soil moisture, and deterioration of vegetation health. It is primarily caused by large-scale atmospheric circulation patterns. With increasing population pressure and expansion of agricultural activities, monitoring vegetation conditions has become essential for effective drought assessment and water resource management.
Methods:
This paper reviews existing literature on drought monitoring by first discussing fundamental concepts and types of droughts and commonly used drought indices and their limitations. Further, it explores the relationships between various drought factors through modelling approaches. Special emphasis is given to supervised Machine Learning (ML) and Deep Learning (DL) techniques used in previous studies for drought prediction and analysis.
Results:
The review highlights that while numerous models and indices have been developed, selecting the most appropriate model and input variables remains a significant challenge. Variations in climatic conditions, data availability, and regional characteristics affect model performance. ML approaches have shown promising results in improving drought prediction accuracy when combined with remote sensing data.
Discussion:
This Systematic review focuses on drought monitoring techniques based on vegetation indices derived from satellite data and the application of Machine Learning (ML) models. It critically evaluates comparative strengths, weaknesses, and optimal application contexts of models for drought assessment using Remote Sensing (RS).
Conclusion:
The study concludes that integrating vegetation indices with ML techniques can significantly enhance drought monitoring and forecasting. It also identifies research gaps and future opportunities in model optimization and parameter selection, encouraging further advancements in this field.