A Novel Machine Learning Framework for Predicting Desertification Trends

This study explores the use of remote sensing and forecasting models to detect and monitor desertification changes in the Wardha region of Maharashtra. By analysing Landsat and MODIS satellite data, the approach highlights ecologically sensitive areas and tracks vegetation dynamics in semi-arid regions. Desertification threatens ecosystems, biodiversity, and livelihoods by degrading fertile land, reducing productivity, and escalating poverty and food insecurity. To benchmark and analyze vegetation growth trends, our study compares the performance of Temporal Convolutional Networks (TCN), Gated Recurrent Units (GRU), Long Short-Term Memory (LSTM) networks, and decision tree regressors, providing insights into the observed shifts. In addressing these challenges, a suite of sophisticated forecasting and change detection methods was stacked to support Sustainable Development Goals (SDGs), including Climate Action (SDG 13), Life on Land (SDG 15), and Zero Hunger (SDG 2).The NDVI (Normalized Difference Vegetation Index) data forecasted from MODIS for the years 2016-2024 was analyzed using the BFAST (Breaks For Additive Season and Trend) algorithm to detect any significant breakpoints in vegetation trends. To validate the BFAST results, the study incorporates a comparative analysis using Landsat satellite data from 2000 to 2015, which provided a historical baseline for vegetation trends in Wardha. The results revealed no substantial disruptions in the NDVI profile, indicating a stable degrading vegetation pattern throughout the forecast period. This apparent stability of ongoing desertification trends points to the insufficient effectiveness of current intervention strategies in halting land degradation. These findings underscore the need for more robust and targeted approaches to environmental conservation and sustainable land management.