Page 261 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

Use Case 19: Spatio-Temporal Microclimate Prediction using Deep

Learning Change 4.2-Climate

Organization: Sri Sivasubramaniya Nadar College of Engineering

Country: India

Contact Person(s):

Syed Azim – syed2110753@ ssn .edu .in
Srikanth S – srikanth2110893@ ssn .edu .in
Prasanna Karthikeyan – prasanna2110778@ ssn .edu .in
Kaythry P – kaythryp@ ssn .edu .in

1 Use Case Summary Table

Item Details
Category Urban Health & Comfort

Problem Addressed Limited availability of localized microclimate data in urban areas due
to sparse weather stations and high cost of dense sensor networks.

Key Aspects of Solution - Vision Transformer model to predict microclimate parameters
- Combines satellite/street images with weather station data
- High-resolution weather maps without dense sensor deployment

Technology Keywords Vision Transformer (ViT), Microclimate Prediction, Satellite Imagery,
Spatio-Temporal Modeling
Data Availability Public: Satellite imagery from Mapbox Static Tiles API
Private: On-ground weather data & street-level images

Metadata (Type of Data) Visual: Satellite and street-level imagery
Numerical: Time-series weather data (temperature, humidity, wind
speed, irradiance)

Model Training and Vision Transformer (ViT) trained with multi-source visual and numeric
Fine-Tuning data to capture spatial and temporal patterns

Testbeds or Pilot ScienceDirect Publication[7]
Deployments

Code Repositories GitHub: Microclimate Prediction[8]

2 Use Case Description

2�1 Description

Urban environments experience significant microclimate variability caused by buildings,
vegetation, and human activity. Traditional weather systems fail to capture these localized

225