Understanding long-term land surface dynamics is critical for assessing the resilience of agricultural landscapes under increasing climate variability. This study examines vegetation trends, bare soil exposure, and irrigation expansion in Hamilton County from 1936 to 2021 using remote sensing and deep learning-based segmentation models. The methodology integrates historical aerial imagery, the Palmer Drought Severity Index (PDSI), and deep learning models-specifically the Segment Anything Model (SAM) and U-Net CNN-to classify dryland and irrigated farming systems with segmentation accuracy between 84 % and 97.3 %. Results indicate that extreme droughts in the 1930s, 1950s, 2000s, and 2010s increased bare soil exposure from 26- 30 % to 40-47 %, with intensified aeolian activity and land degradation. However, post 1930s, 1950s, and 2000s droughts, there was significant vegetation recovery with 50-70 % of the land surfaces nearly (70-75 %) or fully vegetated which highlights the resilience of semi-arid agroecosystems. The expansion of central pivot irrigation systems (CPIS) post 1963 decreased drought impact by maintaining crop productivity, but the area under CPIS stabilized post 2002 possibly due to groundwater depletion and underscores the limits of irrigation-dependent agriculture to mitigate drought effects. Hamilton County in the 1930s had <40 % of land surfaces under cultivation, with 62 % of the land area nearly or fully denuded, a potent source for dust emissivity, and mostly from uncultivated surfaces. This analysis indicates that climate, as synthesized by Summer PDSI, is the dominant factor for landscape denudation for dust generation, rather than purely human agency, which underscores the sensitivity of former Dust Bowl Drought areas to future hydroclimatic deficits and excessive heat.