A Spatial-Temporal Framework for Integrated Land-Use Change Modeling

Thesis Abstract

Urban areas worldwide face rapid land-use transformations driven by population growth, economic development, and climate pressures, which create complex spatio-temporal dynamics that standard single-temporal analyses fail to capture. This study addresses the need for an integrated, spatial-temporal modeling framework capable of quantifying and predicting land-use change (LUC) by synthesizing heterogeneous data sources and theoretical lenses. The aim is to develop a cohesive framework that combines process-based and data-driven approaches to improve forecasting accuracy and policy relevance for land management. The specific objectives are (i) to formulate a spatial-temporal conceptual model that links drivers, land-cover transitions, and resulting ecosystem services; (ii) to integrate multi-temporal remote sensing, census, and ancillary geospatial datasets within a unified modeling architecture; (iii) to operationalize the framework through a hybrid modeling workflow that couples cellular automata with machine learning predictors and a dynamic spatial econometric component; (iv) to evaluate model performance against independent validation data and compare with conventional single-method approaches; and (v) to demonstrate scenario-based applications for urban growth, agricultural rebound, and conservation planning under climate-adaptive policies. The methodological design adopts a mixed-methods, sequential explanatory strategy anchored in a spatial-temporal system theory and supported by the location- and scale-aware insights of the Push-Pull-Moose framework and the Theory of Planned Behavior for human-environment interactions. The population comprises metropolitan-urban fringe regions in a mid-latitude country characterized by rapid development, with a sampling frame drawn from five case study cities each covering approximately 1,500 square kilometers. A stratified random sample of 200 spatial units per city (total N ? 1,000) is selected to ensure representation across urban cores, peri-urban zones, and agricultural mosaics. Data collection instruments include multi-temporal satellite imagery (Landsat 5/8 and Sentinel-2 for the years 2000–2024 at 30 m to 10 m resolution), historical land-use inventories, census-derived socio-economic indicators, infrastructure and policy dataset layers, and high-resolution orthophotos for validation. Primary data are complemented by semi-structured interviews with 40 urban planners, 20 regional policymakers, and 20 landowners to capture governance regimes, incentives, and behavioral drivers. Methodologically, the study advances a hybrid modeling pipeline (i) a cellular automata–neural network (CA–NN) hybrid to capture local spatial diffusion and nonlinear temporal dynamics; (ii) a dynamic spatial panel model (spatial lag and spatial error components) to quantify driving forces and spatial dependency across time; (iii) a random forest or gradient boosting component to estimate non-linear, high-dimensional predictor interactions from socio-economic and accessibility variables; and (iv) a Bayesian updating mechanism to assimilate new data and quantify predictive uncertainty. Model specification is guided by a conceptual framework that links biophysical constraints, economic motivations, and policy interventions to LUC outcomes, with explicit equations formalizing transition probabilities, driver weights, and spatial spillover effects. Validation employs cross-validation across time slices, with metrics including Kappa, F1-score, RMSE, and DPPL for probabilistic forecasts. Sensitivity analyses examine the influence of data quality, temporal resolution, and policy scenarios. Expected findings indicate that the integrated framework outperforms single-method approaches in predictive accuracy (improved Kappa by 10–15%, RMSE reduction of 12–18%), while revealing distinct spatio-temporal patterns of urban expansion, agricultural consolidation, and woodland regeneration under varying policy regimes. The framework will identify key drivers and thresholds, such as accessibility improvements and zoning reforms, that precipitate transition tipping points, and will quantify spatial spillovers affecting adjacent land uses. The study contributes to knowledge by delivering a replicable, theory-grounded modeling toolkit that integrates processual and data-driven perspectives for LUC, explicitly incorporating temporal dynamics and governance contexts. It provides a scalable approach adaptable to other regions and supports scenario analysis under climate-adaptive urban planning, agricultural policy reforms, and conservation prioritization. The main conclusion is that an explicit spatial-temporal integration of drivers, land-use transitions, and governance mechanisms provides superior predictive power and actionable insights for land management. Recommendations include embedding the framework within regional planning workflows, expanding data fusion capabilities to incorporate crowd-sourced land-use observations, and extending the modeling horizon to 20–30 years for long-range policy assessment, along with developing open-access software modules to facilitate adoption by researchers and practitioners.