Assessing the impacts of urbanization and climate variability on cropland dynamics and food security in Punjab, Pakistan (2000-2025): a multi-scale remote sensing and machine learning approach.

Punjab, Pakistan's primary agricultural province, faces mounting pressure from urbanization, groundwater depletion, and climate variability, threatening cropland sustainability and food security. This study quantifies spatiotemporal cropland and built-up area dynamics at the district scale, detects long-term trends in eight climatic, hydrological, and vegetation variables, and attributes cropland change to its dominant drivers across Punjab's 36 districts over 2000-2025. Multi-resolution land cover datasets were integrated with ERA5-Land and GLDAS-2.2 climate reanalyses and processed in Google Earth Engine. District-scale Mann-Kendall trend analysis and Random Forest (RF) regression were applied to 288 district-year observations. Punjab lost 15,302 km2 of cropland (- 12.30%) between 2000 and 2025, while built-up area expanded by 180.31%, more than twice the global average rate. The provincial-scale Mann-Kendall test for cropland returned a non-significant trend (p = 0.387), reflecting high inter-annual variability that masks the spatially concentrated, structurally driven signal at the district level. All 36 districts recorded urban expansion, and cropland loss was concentrated in 13 peri-urban districts. Groundwater storage declined significantly by 12.16% (p = 0.035), while actual evapotranspiration nearly doubled (+ 91.02%, p = 0.002), indicating a tightening agricultural water balance. The RF model explained 92.96% of absolute cropland variance (cross-validated R2 = 0.59 on the change target), with change in built-up area and year emerged as co-dominant predictors (MDI = 0.212 each), followed by precipitation variability (0.170). These findings provide a district-scale, spatially explicit evidence base for land governance, groundwater regulation, and food security policy in Punjab and analogous South Asian agricultural systems.