Predictive AI Models for Disaster Management and Climate Resilience

Aim

Workshop Objectives

Workshop Structure

📅 Day 1 — Forecasting the Extreme (Predictive Analytics)

• Focus: Mastering the temporal aspect of natural disasters
• Physics-Informed Machine Learning (PIML): Integrating physical constraints (fluid dynamics, thermodynamics) into deep learning loss functions to model rare climate events
• Multivariate Time-Series for Early Warning: Implementing Transformer and LSTM architectures for sub-seasonal to seasonal (S2S) forecasting of droughts and heatwaves
• Industry Application: Transitioning from standard error metrics to decision-based risk modeling for insurance and agricultural sectors
• 🚀 Colab Hands-on: [DeepFlood-Predictor] – Build a flood forecasting model using NASA’s Global Flood Awareness System (GloFAS) open data

📅 Day 2 — Real-time Intelligence & Spatial Risk

• Focus: Leveraging computer vision and GIS for immediate climate response
• Multimodal Data Fusion: Synchronizing Synthetic Aperture Radar (SAR) and optical satellite imagery to “see” through cloud cover during active storms
• Automated Change Detection: Using Siamese Networks and Vision Transformers to identify post-disaster infrastructure damage (bridges, roads, buildings)
• Digital Twin Integration: Syncing AI-generated risk maps with ArcGIS/QGIS for urban climate adaptation planning
• 🚀 Colab Hands-on: [Sentinel-SAR-Detect] – Implement a change detection pipeline to identify deforestation or flooded urban zones using SAR imagery

📅 Day 3 — Deployment, Ethics & Resilient Infrastructure

• Focus: Moving from research notebooks to mission-critical environments
• Edge AI for Remote Sensing: Model quantization and deployment on drones or IoT sensors for offline-first wildfire detection
• Explainable AI (XAI) for Stakeholders: Using SHAP and LIME to provide interpretable justifications for AI-driven evacuation orders
• Algorithmic Fairness in Climate Action: Addressing data scarcity in “Data Deserts” to protect vulnerable populations equitably
• 🚀 Colab Hands-on: [Quantize-&-Explain] – Compress a Vision Transformer (ViT) using TensorFlow Lite and generate a SHAP heatmap to explain its risk classification

Who Should Enrol?

Important Dates

Workshop Outcomes