Siting Solar without Displacing Pastoralists: A Conformal-Calibrated Many-Objective Framework for Kazakhstan

Presented at the IEEE Global Humanitarian Technology Conference (GHTC) 2026, University of Colorado Boulder, 7 October 2026. This is the author's accepted manuscript. Land allocation for Kazakhstan has a fundamental problem: solar energy, cereal cropping and, most importantly, pastoralists all compete for the same land use. Yet current land allocation frameworks have ignored the often invisible nature of pastoralists, resulting in displacement of herders when their presence is not considered. Current frameworks also fail to quantify their own uncertainty, creating a lack of trust for policymakers in the predictions that are made. This study proposes, to the best of our knowledge, a many-objective land allocation framework (NSGA-III) that considers all three factors on equal ground, with calibrated uncertainty through conformal prediction. We model each competing land use with three separate suitability maps, implement distribution-free conformal prediction intervals for each hexagon with a 90% coverage guarantee, and propagate uncertainty through the optimizer by implementing random draws from within the conformal intervals. This results in a trade-off frontier across the three land uses. We draw an indeterminacy layer of decision-confidence, flagging the locations where land use is uncertain, and re-run the pipeline under two emissions scenarios. Under multiple seeds, country-wide allocation stays constant (CV ≤ 0.06), while per-cell allocation is distinct (Jaccard ≈ 0.18), suggesting that the framework is reliable for national-scale planning but fails when used for selecting individual sites. Under climate change, cereal cropping value rises sharply (+172% under SSP5-8.5) and the solar centroid shifts north (+29 km), highlighting the implications of future climate scenarios for policymakers. This framework acts as a decision-support layer for the Kazakh government when allocating land plots, ensuring transparency in the decision process and flagging where real human decisions are required.