A Unified Framework for Spatio-Temporal Land-Use Change Modelling

Thesis Abstract

Contextual pressures from urbanization, agricultural intensification, and climate variability drive dynamic land-use changes that are spatially heterogeneous and temporally evolving, challenging conventional modelling approaches. The study addresses the need for a unified framework that integrates spatio-temporal processes, multi-source data, and theory-driven mechanisms to improve prediction, interpretation, and policy relevance of land-use change. The aim is to develop a cohesive modelling framework that fuses spatial econometrics, cellular automata, and machine learning within a social-ecological theory lens to capture both proximal drivers and distal influences on land-use transitions over decadal horizons. Specific objectives are to (i) identify and harmonize key drivers across biophysical, socio-economic, and institutional domains; (ii) formulate a hybrid modelling architecture that combines cellular automata, spatial autoregressive processes, and temporal deep learning components; (iii) calibrate and validate the framework using multi-temporal land-cover datasets and census-derived socio-economic indicators; (iv) evaluate predictive performance against benchmark models and perform scenario analysis under policy and climate stressors; and (v) provide actionable insights for land management and regional planning. Methodologically, the research adopts a mixed-methods design under a positivist-constructivist philosophical stance to build generalizable yet context-sensitive knowledge. The population comprises metropolitan and peri-urban regions with diverse land-use trajectories across three continents, selected to ensure variability in governance structures and biophysical conditions. A stratified random sample of 24 municipalities per region (72 in total) yields a robust dataset for model development and cross-region validation. Data collection draws from (i) high-resolution remote sensing-derived land-use maps for the years 2000, 2010, 2020, and 2030 projections; (ii) census and administrative records for socio-economic indicators (population density, income, agricultural intensity, infrastructure access); (iii) climate datasets (precipitation, temperature) and soil quality indices; and (iv) policy and governance variables (zoning regulations, protected areas, incentive schemes). Instruments include a validated land-use classification schema, standardized GIS templates, and structured coding schemes for policy variables. Data quality is ensured through cross-validation of remotely sensed classifications (overall accuracy >85%), temporal harmonization, and imputation of missing values with multiple imputation techniques. Analytical methods combine (i) descriptive spatial-temporal analysis to characterize change regimes; (ii) a unified model architecture integrating cellular automata for neighborhood effects, spatial autoregressive models for spatial dependence, and long short-term memory networks for temporal dynamics; (iii) structural equation modelling to identify latent drivers and test theoretical pathways anchored in urban morphology and land-use transition theories; (iv) feature selection via elastic net regularization to handle high-dimensional driver sets; (v) rigorous out-of-sample validation using a rolling-origin approach and region-based cross-validation; and (vi) scenario analysis employing policy levers (zoning liberalization, green infrastructure subsidies) and climate projections to assess potential futures. Hypothesis tests will examine the significance of biophysical, economic, and governance factors, and the added predictive value of the hybrid framework relative to standard CA-Markov and purely statistical models. Expected findings indicate that the unified framework yields superior predictive performance (out-of-sample R2 improvements of 12–18% and reductions in mean absolute error by 8–14%) and reveals distinct spatio-temporal regimes driven by governance quality, market access, and climate variability. The framework is anticipated to uncover non-linear interactions between proximity to urban centers, agricultural profitability, and policy incentives that precipitate abrupt land conversions, with temporal lags varying by region. The study contributes to knowledge by providing a scalable, theory-grounded modelling approach that reconciles mechanistic and data-driven perspectives, extends the applicability of spatio-temporal modelling to land-use planning, and delivers a reusable framework and software prototype for practitioners. The main conclusion is that a truly unified spatio-temporal modelling framework—grounded in urban morphology theory and socio-ecological systems theory, operationalized through an integrated hybrid architecture, and validated across heterogeneous regional contexts—offers robust, interpretable, and policy-relevant insights for anticipating land-use change under multiple futures. Recommendations include the adoption of the framework in regional planning agencies to inform zoning and conservation strategies, investment in harmonized multi-source data infrastructures, and further refinement of scenario-based decision support tools to incorporate emerging governance regimes and climate-resilient pathways.