From Mapping to Decision Making: A hybrid Rule-Based and Machine Learning Framework for Spatial Land-Use Zoning

Fatma Esen¹Enes Karadeniz^2*Fatih Sunbul³Asli Deniz Adiguzel¹

• ¹Bitlis Eren Universitesi, Bitlis, Türkiye
• ²İnönü University, Malatya, Türkiye
• ³Bakircay Universitesi, İzmir, Türkiye

Rapid land use change in coastal and peri-urban regions is increasingly associated with biodiversity loss, ecosystem service decline, and growing spatial tensions between conservation priorities and socio-economic development. This study introduces the Dual-Logic Spatial Zoning Model (DLSZM), a hybrid rule-based and machine learning framework designed to move beyond mapping-based assessments toward more operational land use decision support. DLSZM links socio-ecological indicators to explicit land use regimes by integrating three thematic components, Ecosystem Services and Biodiversity (ESB), Human Use and Benefits (HUB), and Stress and Vulnerability (SV), and translating them into four planning zones: Strict Conservation, Managed Use, Development Guidance, and Restoration. In the expert-derived pathway, fuzzy triangular judgments and Log-LMAW weighting are used to establish a transparent rule-based zoning logic, while the machine learning pathway applies the same predictors as continuous raster inputs to train a CatBoost model that learns and generalises this logic with SHAP-based explainability. Applied to the Antalya region in Türkiye, the two pathways show substantial spatial agreement at the regional scale, particularly within dominant Managed Use areas, whereas observed divergences are spatially structured and tend to cluster along coastal belts and transitional landscapes characterised by overlapping ecological value, accessibility, and cumulative pressure gradients. Compared to the expert-derived outcomes, the machine learning (ML)- based zoning allocates a larger proportion of both natural and artificial land cover to the Restoration category, indicating a higher sensitivity to interacting pressure and vulnerability signals. Feature importance and SHAP analyses provide insights that suggest ESB-related variables are more closely associated with conservation-focused zoning outcomes. Conversely, variables linked to HUB and SV tend to play a more significant role in shaping decisions relating to Development Guidance and Restoration. Overall, these findings indicate that the DLSZM is a transparent and reproducible zoning framework that makes it easier to identify where rule-based expert reasoning aligns with, or differs from, data-driven model outputs. In this sense, the framework provides a practical and flexible foundation for supporting adaptive land use management in dynamic coastal areas.