AI Enhanced Urban Microclimate Mapping Using Deep Learning Algorithms Dr. Heltin Genitha C(a*), Aashika Perpetual G(b), Mohanapriya K(c)
a*- Professor, Department of Information Technology,
St Joseph^s College of Engineering.
b, c- Student, Department of Information Technology,
St Joseph^s College of Engineering.
Abstract
Urban microclimates-localized variations in temperature, humidity, and air quality-are strongly influenced by factors such as building density, traffic, vegetation, and land use. Traditional climate monitoring methods lack the spatial granularity needed to capture these variations, limiting their usefulness for urban planning and environmental management.
This paper presents an AI-driven framework for dynamic urban microclimate mapping using satellite-based remote sensing data. The system integrates multi-temporal imagery from the Sentinel and Landsat datasets with advanced machine learning models to analyze both spatial and temporal environmental patterns. A ResNet-based Convolutional Neural Network (CNN) is employed for spatial feature extraction from satellite imagery, while a Long Short-Term Memory (LSTM) network models temporal trends across different urban zones.
The framework was implemented and evaluated on satellite data from a Chennai metropolitan area, India demonstrating enhanced spatial resolution and predictive accuracy in detecting heat islands and pollution zones. The resulting heatmaps and environmental visualizations provide valuable insights for urban planners and policymakers, supporting data-driven decisions for climate-resilient infrastructure and sustainable development.
Keywords: Urban Microclimate, Remote Sensing, Sentinel, Landsat, Deep Learning, Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Heat Island Detection.
