Krupa P. Nadgir, M. Nithyashree, Prithiviraj N, Manasvini G. Padmasali, Deeksha Hegde
UHI LENS is an innovative solution designed to mitigate Urban Heat Islands (UHIs) caused by the temperature difference between urban and surrounding rural areas.
The system combines satellite imagery, socio-economic features, machine learning models, and LLM-based recommendations to analyze, detect, and propose mitigation strategies against UHIs.
📄 Publication Link: IEEE Xplore 🎥 MVP Demo Video: Watch here
The system integrates multiple models, each designed for a specific purpose:
- UHI Index Prediction – DL model to predict Urban Heat Island Index (UHII) from satellite imagery.
- Semantic Segmentation Model – DL model to predict Land Use Land Cover (LULC) masks of the ROI.
- UHI Classification – Random Forest Classifier to determine whether UHI sources are man-made (controllable) or natural (uncontrollable).
- 12-Month UHI Regression Model – Ensemble model (RFR, SVR, LGBM) to forecast monthly UHII variations.
- LLM (Llama3) – Provides region-specific mitigation strategies based on local socio-economic and environmental conditions.
The platform is built with a modern and interactive web interface powered by:
- HTML, CSS, JavaScript
- Tailwind CSS
- ArcGIS Maps SDK for JavaScript
- Plotly
- Flask
- Input: User enters a location name.
- Data Fetching:
- Geocoding API – Fetch latitude & longitude of ROI.
- Earth Engine API – Authenticate and capture the latest satellite imagery.
- Preprocessing: Cloud masking, atmospheric correction, color correction, and semantic segmentation → Extract LULC Mask.
- Feature Extraction:
- 15 Sentinel Features
- 9 Socio-Urban Features
- LULC Features
(All features are viewable and analyzable separately.)
- Model Predictions: Predict LULC, project UHII, classify cause, forecast 12-month UHII trends, and provide LLM recommendations.
All models and APIs are integrated within the system and deployed for end-to-end UHI assessment.