Key points are not available for this paper at this time.
The intricate interplay between land use, climate dynamics, and other contributing factors significantly influences the occurrence of extreme events such as droughts, floods, and heatwaves. Modeling this complex system in a high-dimensional space poses a formidable challenge, given incomplete understanding and limited availability of data. This study explores the application of deep learning approaches, specifically leveraging transformer architectures, to capture long-range dependencies in spatiotemporal data. These mechanisms are then employed to encapsulate the complex interactions between land use, climate, and other factors influencing extreme events. The proposed approach incorporates attention mechanisms, enhancing interpretability by highlighting crucial spatial and temporal features essential for forecasting. To evaluate the effectiveness of this methodology, a case study was conducted on the Godavari River Basin in India. Utilizing vegetation indices as a representation of crop type and land use, alongside climate data spanning from 2000 to 2020, the results provide valuable insights into the driving factors behind land use change and climate extremes in the region. The study not only demonstrates predictive capabilities of the proposed approach but also offers insights into the intricate relationships within the land-atmosphere feedback system. The extracted information is useful for making informed decisions related to land management, climate adaptation, and disaster risk reduction.
Subhadarsini et al. (Fri,) studied this question.