Crop planting pattern optimization is a critical strategy for alleviating agricultural water stress in arid and semiarid regions. However, conventional models based on coarse spatial units often overlook essential environmental heterogeneity, limiting the precision and feasibility of optimization results. While recent advances in remote sensing enable high-resolution spatial data, the influence of spatial resolution on optimization performance remains poorly understood. Here, we combine multi-year, high-resolution remote sensing data with spatial linear programming to evaluate how spatial resolution affects crop reallocation strategies in the Hetao Irrigation District, China. Optimization is performed at county, 3000 m, and 300 m scales, aiming to maximize economic benefits or minimize crop water consumption under realistic constraints including land and water availability. Despite consistent overall trends across different optimization scales-reducing maize area and increasing sunflower cultivation-finer spatial models yield substantially greater benefit increments. At 300 m resolution, spatially coherent re-allocations aligned with landscape heterogeneity achieve up to 205% higher economic benefit increments and 283% more water savings than county-level models. Distinct trade-offs between objectives shape land-use adjustments, i.e., economic maximization favors conservative maize reductions by retaining high-value fields, while water-saving goals drive broader maize replacement with sunflower to enhance efficiency. These findings highlight the pivotal role of spatial granularity in agricultural optimization and offer actionable insights for designing adaptive, efficient cropping strategies in water-scarce regions.