Biodiversity, conservation and design thinking: an AI-driven, geodesign framework for determining habitat quality

Abstract Species Distribution Modeling (SDM) is central to biodiversity assessment, yet its diagnostic outputs rarely translate into spatially explicit conservation strategies. We present an AI-driven Geodesign framework that embeds Maximum Entropy (MaxEnt) modeling in an iterative refinement cycle to produce phased conservation priorities. Applied to Illinois, we compiled 826,989 occurrence records for 10,029 species across 11 taxonomic classes (fungi, invertebrates, vertebrates, and plants) modeling class-level habitat suitability against 27 environmental predictors at 600 m resolution. Iterative refinement improved mean apparent Area Under the Curve (AUC) from 0.684 to 0.764 (mean True Skill Statistic 0.439). Comparison with 542 Critical Trends Assessment Program (CTAP) field sites showed statistically significant but modest concordance between the modeled plant component and observed plant richness (Spearman ρ = 0.275, p < 0.001). Protected area covers only 4.42% of state lands, misaligned with modeled conservation priorities along the Mississippi River corridor and Shawnee Hills. Four conservation tiers quantify protection gaps: the most conservative (95th percentile, 7.25% of area) requires 1.6-fold expansion, while meeting 30×30 goals (75th percentile, 36.25%) demands an additional 4.64 million hectares. A Biodiversity Efficiency Index showed deciduous forests support 120-fold greater composite habitat suitability per hectare than barren land. The framework provides a reproducible workflow with quantitative benchmarks aligned with 30×30 and sustainable development goals.