A graph clustering neural network framework for mapping urban functional zones in a fast-evolving urban landscape of Bangladesh

Rapid urbanization and uncoordinated land development in Bangladesh's cities, particularly Khulna City Corporation (KCC), require new approaches to spatial planning and functional zoning. This study aims to identify and classify urban functional zones in KCC using a data-driven framework. It combines various geospatial datasets including urban spatial network structure, digital elevation model (DEM), road network, land use land cover (LULC), nighttime light data, and points of interest (POI) with an advanced Graph Clustering Neural Network (GCNN) model. Five functional zones are defined including Industrial and Eco-Agricultural Zone (C0), Residential Zone (C1), Commercial and Residential Zone (C2), Mixed Functional Zone (C3), and Business-Service Zone (C4). The model effectively captured complex spatial relationships and achieved great clustering accuracy, confirmed by satellite imagery and land use patterns. Results have shown that C1 and C2 are the most urbanized, with respectively 49.92% and 45.52% of their areas covered by built-up zones. In contrast, C0 and C4 respectively serve essential industrial, ecological, and administrative support. Residential zones (C1) within the service clusters largely lack essential services, especially health facilities, but are largely present in C4. This study identifies important variations in access to services, green space, and infrastructure across KCC. Overcoming limitations in traditional clustering methods by using graph-based spatial learning has provided a strong method for urban functional zoning. Recommendations that accord with data-driven urban planning, improving infrastructure, and sustainable development strategies are thus applicable in rapidly urbanizing areas across Bangladesh.