Multi-Objective Deep Reinforcement Learning-Based Spatial Optimization of Green Infrastructure in Rapidly Densifying Cities: A Case Study of Ghorahi, Dang, Nepal

Urban microclimatic degradation is compounded by the rate of rapid urban expansion and rapidly increasing urban density in rapidly densifying metropolitan areas, which increase the Urban Heat Island (UHI) effect and worsen the disruption of ecological connectivity. Two-dimensional (2D) spatial planning currently existing in the planning framework has not considered the vertical aspect of urban densification and the vertical dimension also influences the sub-optimal location of Green Infrastructure (GI) and Blue-Green Infrastructure (BGI). To address the above limitations this paper presents a novel three-dimensional (3D) spatial optimization algorithm, which integrates high-resolution remote sensing data with 3D morphology and microclimate models generated by UAVs and multi-objective Deep Reinforcement Learning (DRL). The methodology is based on the formulation of GI placement as a Markov Decision Process (MDP) so that three project aspects, namely 3D microclimatic cooling benefit, pedestrian heat stress and ecological network connectivity will be optimized without being subjected to socio-economic fairness conditions, without the risk of algorithmic bias and spatial displacement. The proposed 3D-aware DRL optimization has reduced heat stress by a mean of 2.8 °C within the urban testbed area of 5 km² in Ghorahi Sub-Metropolitan City, Dang, Nepal from conventional 2D heuristic of 1.2 °C compared to base condition and increased the Integral Index of Connectivity (IIC) from 0.14 to 0.45 compared to base condition. The framework further translates model outputs into actions by linking with urban governance tools using the municipal building bylaw and requirement of green space coverage provided in the Local Government Operation Act (2017) in the municipalities of Nepal through aligning outputs of scenarios with municipal bylaws. This work merges high-dimensional and AI-driven analytics and socio-technical urban governance and offers a scalable blueprint for urban densification that is robust, climate-resilient and equitable indeed.