Modeling Flood and Drought Risk Perception Gaps in India

Understanding Risk Perception Gaps

• ⚠️ India faces escalating climate-related risks including heat waves, drought, flooding, air pollution, and energy disruptions, often in densely populated areas.
• 🧠 There's a significant gap in understanding how individuals and communities perceive these risks compared to expert assessments.
• 🎯 Risk perceptions drive real-world outcomes and preparedness, making it crucial to bridge this gap for effective policy and emergency alert systems.

Data and Modeling Approach

• 📊 The project utilized national survey data from the Yale Program on Climate Change Communication, combined with climate exposure metrics and socio-demographic variables.
• 🤖 Machine learning and AI models (Random Forest, XG Boost, Logistic Regression) were employed to predict individual-level perceived risk for floods and droughts.
• 🛰️ A novel data source, Alpha Earth embeddings (Google's machine learning processed satellite images), was integrated to capture complex environmental features traditional datasets might miss.
• ✅ Model performance was evaluated using accuracy, precision, recall, F1 score, and ROC curves.

Flood Perception Insights

• 🌳 Random Forest models demonstrated the highest performance in predicting flood perception, particularly when incorporating Alpha Earth embeddings.
• 💡 Alpha Earth embeddings significantly improved model accuracy, suggesting they capture crucial, unexplained terrain characteristics relevant to flood risk perception.
• 💬 Feature importance varied across models, with sentiment analysis, urban/rural status, gender, and specific Alpha Earth bands emerging as key predictors.

Drought Perception Findings

• 📈 For drought perception, Random Forest and Gradient Boosting models showed superior performance.
• 🌍 Alpha Earth embeddings were also highly relevant in predicting drought perception, especially within machine learning models.
• 🔍 Drought models generally performed better than flood models, potentially due to challenges in accurately mapping floods in urban or vegetated areas and the differing ways people experience these disasters.

Implications and Future Directions

• 🚀 The team developed a flexible modeling framework that can be adapted to other countries and various hazards.
• 🗺️ Ongoing work includes refining models to create district-level risk perception maps and quantifying the gaps between perceived and expert-assessed risks.
• 📚 Two manuscripts are currently in preparation, detailing the findings on flood and drought risk in India.
• 🌐 All project data, notebooks, and data cards are publicly available on the iGuide platform for further exploration and collaboration.