By Dominique Ritter | July 29, 2026
As climate change makes weather patterns increasingly unpredictable, traditional forecasting tools grounded purely in historical data are falling short. In response, Quebec City-based startup Geosapiens is developing advanced geospatial AI solutions to help the insurance industry evaluate climate risks, including wildfires and flooding, while giving Canadians better tools to protect themselves.
In a recent conversation, Chief Research Officer Chiranjib Chaudhuri explained how the company’s platform overcomes the limitations of standard machine learning to model extreme climate events accurately.
A major vulnerability of standard generative AI is "hallucination"—producing inaccurate outputs without any mechanism to measure underlying uncertainties. Geosapiens addresses this through geospatial AI, which actively quantifies data uncertainty rather than guessing outcomes blindly.
Furthermore, traditional deep learning models are strictly limited by the dataset used to train them. For instance, a model trained on 30 years of historical data can generally only predict up to a 30-year storm event. To bypass this barrier, Geosapiens utilizes distributional extrapolation. This technique estimates data parameters beyond standard sample ranges, allowing the platform to model rare occurrences—such as 100-year storms—even when historical records fall short.
The shift in climate predictability impacts different sectors in distinct ways:
Insurance Sector: Primarily focused on short-term horizons, insurers require high-precision annual models to accurately assess risk and determine upcoming premium costs.
Municipal Planning: Cities must plan decades ahead. Because climate change is shifting baseline risks—causing what was historically a 50-year storm to become the new 100-year standard—relying solely on past data for infrastructure planning is no longer viable.
According to Chaudhuri, effective models must understand the non-stationary nature of modern weather while respecting the physical laws governing weather variables. Incorporating observation data on a massive scale enables AI to recognize these shifting physical patterns, yielding far more reliable forecasts.
As deep learning models grow more complex, Chaudhuri warns against relying on "black box" systems where the inner decision-making process is hidden. Because weather modeling directly influences high-stakes decisions involving human lives and billions of dollars in infrastructure, AI architectures must be built specifically for human interpretability.
Ultimately, technology cannot replace human judgment. Chaudhuri stresses that maintaining a "human-in-the-loop" approach is essential for responsible AI deployment, ensuring scientists remain accountable for how algorithmic predictions are interpreted and applied in the real world.
For further insights, read about Geosapiens and their contributions at Geosapiens and explore their role within the Adaptech Accelerator.
This website uses cookies to save your preferences, and track popular pages. Cookies ensure we do not require visitors to register, login, or share any identity information.
