Page 262 - AI for Good Innovate for Impact
P. 262

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
   257   258   259   260   261   262   263   264   265   266   267