Implementation of deep learning algorithms to model agricultural drought towards sustainable land management in Namibia's Omusati region

Abstract

Namibia's Omusati region, a semiarid agroecological zone, faces intensifying agricultural droughts driven by climate change, erratic rainfall, and land degradation. With 70 % of its population dependent on rain-fed subsistence farming, these droughts threaten food security, livelihoods, and ecological stability, highlighting the need for predictive tools to support sustainable land management. This study employs deep learning models to analyse remote sensing data from Moderate Resolution Imaging Spectroradiometer (MODIS), Climate Hazards Group InfraRed Precipitation with Station (CHIRPS), and Global Land Data Assimilation System (GLDAS) data to predict drought drivers, such as the normalized difference vegetation index (NDVI), evapotranspiration, land surface temperature, rainfall, soil moisture, and leaf area index. Trend analysis revealed a drought index increase from 0.3 (2003–2015) to over 0.7 by 2022, with tree cover decreasing to 0.08 % by 2023 and bare ground expanding, indicating severe ecological degradation. Projections indicate an 8 % increase in drought probability by 2024, endangering 81 % of farmland. Convolutional neural networks (CNNs), long short-term memory networks (LSTMs), identify the NDVI and land surface temperature as key predictors, whereas Convolutional LSTM (ConvLSTM) provides spatial insight into high-risk zones, although with lower accuracy. These findings can guide targeted strategies for adopting drought-resistant crops, water-efficient irrigation, and soil conservation in vulnerable areas. We recommend integrating AI-driven drought forecasts into Namibia’s climate and land policies through real-time monitoring, community-led adaptation and interagency coordination to increase on-ground resilience.