GeoAI-based Urban Environmental Forecasting: A Remote Sensing Driven Hybrid Deep Learning and Machine Learning Framework

Mirza Md Tasnim Mukarram, Quazi Umme Rukiya, Ibrahim Demir

This study presents a hybrid GeoAI forecasting framework for long-term environmental monitoring in Dhaka, Bangladesh, one of the most densely populated and environmentally degraded megacities in the Global South. Using a 25-year record (2000–2024) of multi-source satellite and climate data, we modeled monthly trends in five key variables: Normalized Difference Vegetation Index (NDVI), Bare Soil Index (BSI), precipitation, and maximum and minimum air temperatures. Five supervised learning algorithms were trained and validated using univariate time-series features: Random Forest (RF), Support Vector Regression (SVR), Dense Neural Network (DNN), Long Short-Term Memory (LSTM), and a hybrid LSTM–XGBoost ensemble. Among these, the hybrid model demonstrated superior performance with R² values exceeding 0.99 for NDVI and BSI and ≥0.95 for temperature variables, while maintaining robust generalization for seasonal precipitation anomalies. Google Earth Engine is used to generate forecasts for 2030, yielding spatially explicit raster predictions for all variables. Model outputs indicated relatively stable conditions between 2024 and 2030, with localized environmental stress persisting in peri-urban and low-vegetation zones. To validate model interpretability and operational relevance, a two-stage participatory process was conducted with 28 urban stakeholders through structured interviews and a validation workshop. Survey results indicated that 72% of participants rated the forecasts as “useful” or “very useful” for urban planning, while 64% found the NDVI and temperature maps “understandable” without specialized training. Thematic analysis highlighted accessibility, local specificity, and trust in AI forecasts as key factors influencing user acceptance. These findings support the scalability of the proposed GeoAI framework and its alignment with planning priorities in climate-stressed urban environments of the Global South.