Climate-related financial losses hit $145 billion in the US in 2024, up 40% from the previous year. Most financial institutions still use static, rule-based risk models built for yesterday's weather. This talk shows how leading banks are using machine learning to turn climate risk into a competitive advantage.

The Problem:
Traditional models use linear projections and fixed thresholds that miss the exponential nature of climate impacts. Financial institutions face three risk types: physical (your data center underwater), transition (your clients' assets stranded), and liability (your disclosure challenged in court). Each needs a different analytical approach. The real challenge is processing 10TB of daily satellite imagery, a million IoT readings per second, and tens of thousands of news articles and company reports into something actionable.

What We Built:
I'll walk through a production ML pipeline running at major financial institutions. We use an ensemble approach: Random Forest handles missing data and achieves 95% accuracy on default prediction. XGBoost captures non-linear patterns and detects 90% of risk events. LSTM networks provide 6-month early warning signals. Dynamic Beta models update climate sensitivity every trading day.
I'll cover two case studies. JPMorgan Chase's Carbon Assessment Framework processes 500 petabytes annually across 50,000 corporate clients, creating $1.5B in value. BNP Paribas built satellite-based ESG monitoring that detects methane leaks from space, saving €50M through early detection.

Results That Matter:
25% fewer climate-related losses. 60% faster risk assessments. 40% lower compliance costs. 50,000 analyst hours saved annually. Over $100B in green finance opportunities identified.
Attendees will leave with a practical framework for evaluating ML-readiness in their own climate risk work. The session works for data scientists looking at deployment patterns, risk managers exploring AI, and business leaders weighing the ROI of climate analytics.