A Hybrid Deep Learning and Remote Sensing Framework for Climate-Smart Livestock Grazing in Semi-Arid Nigeria

The agricultural sector in Nigeria plays a vital role in ensuring food security and providing jobs, particularly in rural areas like Adamawa State, where many families depend on livestock for their livelihoods. Today, however, farmers are struggling more than ever to sustain livestock production because of climate change, which brings unpredictable rainfall, degraded land, and shrinking grazing areas. This study develops a practical data-driven framework that utilizes hybrid deep learning and remote sensing to support climate-smart livestock grazing in Adamawa State. Information from multiple sources (Sentinel‑2 and Landsat satellite images, climate data, and field observations) was integrated to assess how suitable different regions are for grazing. A hybrid ConvLSTM (Convolutional Neural Network and Long Short‑Term Memory) model was used to capture how vegetation and environmental conditions change over space and time, and to classify grazing suitability into three levels (low, moderate, and high). The model performs excellently by achieving 90.1% overall accuracy with balanced recall, precision, and F1‑scores. We then translated these results into GIS-based grazing suitability maps for Fufore and Mubi South Local Government Areas, which revealed realistic environmental conditions that are consistent with field observations. The resulting prototype offers a practical decision-support tool that can aid farmers and herders in using resources more efficiently, which will improve grazing management and reduce climate-related risks, thereby contributing to more sustainable livestock production in semi-arid regions.