Page 262 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

effects, creating challenges for urban comfort, heat risk mitigation, and sustainable design.
This use case aims to develop a scalable AI solution using Vision Transformers to predict
microclimate parameters—temperature, humidity, wind speed, and solar irradiance, at high
spatial and temporal resolution.

The model is trained on:
• Satellite imagery from Mapbox API
• Street-level images taken near weather stations
• Real-time weather station data from a pilot campus

Unlike conventional forecasting models, Vision Transformers (ViT) leverage self-attention
mechanisms to learn spatial-temporal relationships from both image and numeric data.
This enables precise mapping without dense sensor networks or computationally expensive
simulations.

Expected Benefits:

• More informed urban planning and public space design
• Better outdoor thermal comfort for pedestrians and cyclists
• Improved climate resilience and resource-efficient agriculture
• Data-driven decision support for campus and urban development

Use Case Status: Operational prototype; academic deployment on a college campus

2�2 Benefits of use case

• Supports thermal comfort and reduces heat-related risks for vulnerable urban populations
• Promotes AI-driven climate monitoring innovations and resilient infrastructure planning
• Informs climate-sensitive urban design and mitigates urban heat islands
• Enables efficient irrigation and resource use in agriculture through fine-grained
predictions

2�3 Future Work

• Data Expansion: Increase training data with seasonal and multi-location inputs
• Model Enhancement: Integrate multi-modal transformer models for better prediction
• Real-Time Optimization: Streamline model for deployment on edge or real-time systems
• Collaborations: Explore partnerships with urban authorities and agritech firms for
validation and scale-up
• Infrastructure Needs: Additional sensors, high-res imagery access, and GPU upgrades

3 Use Case Requirements

• REQ-01: Access to satellite imagery APIs (e.g., Mapbox Static Tiles) for geo-tagged visuals
• REQ-02: Distributed weather stations to collect real-time climate parameters
• REQ-03: Camera-enabled image collection system near sensor locations
• REQ-04: GPU-enabled computing infrastructure for model training and deployment

226