Spatially Aware Machine Learning for Urban Green Space Accessibility and Landscape Planning Optimization

Abstract

Urban green spaces play a critical role in improving environmental quality, promoting public well-being, and supporting sustainable urban development. However, conventional landscape planning approaches often fail to adequately capture spatial inequalities, population distribution patterns, and complex interactions between urban infrastructure and green space accessibility. This study proposes a spatially aware machine learning framework for evaluating urban green space accessibility and optimizing landscape planning strategies. The proposed framework integrates geographic information systems, satellite-derived vegetation indices, population density information, and urban road network characteristics to construct a comprehensive representation of urban landscape environments. A graph-based spatial learning module is introduced to capture neighborhood connectivity and spatial dependencies, while a multi-objective optimization strategy balances green space accessibility, population coverage, and land-use constraints. The framework is designed for evaluation using approximately 12,000 urban spatial units across three metropolitan areas, integrating Sentinel-2 satellite imagery, OpenStreetMap road networks, and publicly available population datasets. Experimental comparisons will include Random Forest, XGBoost, spatial regression, and graph neural network models. Evaluation metrics will include accessibility prediction accuracy, mean absolute error, population coverage improvement, and spatial equity indicators. The proposed framework aims to support data-driven landscape planning and more equitable urban green infrastructure development.