The Center for Data Innovation recently spoke with Fernando Yu, co-founder of Suyana, a Massachusetts-based company that is using satellite imagery and machine learning models to provide climate risk insurance. Yu explained how Suyana’s models identify climate patterns, assess the likelihood and severity of damaging events, and translate those findings into insurance products suited to specific areas.
David Kertai: What challenge is Suyana solving in the climate-based insurance industry?
Fernando Yu: Traditionally, when farmers or businesses seek climate insurance for risks, such as droughts or flooding, they work with insurers who rely on loss-adjusters, field inspectors, and reporting agencies to evaluate disaster exposure. These teams conduct manual site visits, review historical claims, and rely on sparse weather‑station data. Collecting this information takes time, is costly, and often leaves critical gaps, so insurers price policies using broad regional averages instead of the real conditions on specific plots of land. In turn, claims move slowly, costs rise, and risk is often misrepresented.
Suyana takes a different approach by combining high‑resolution satellite imagery from providers such as Planet Labs with on‑ground data like soil moisture, precipitation, sea‑level measurements, and global climate‑model outputs. Our machine‑learning models merge these inputs to generate hyper‑specific risk assessments at a 1 km by 1 km resolution, a roughly 400‑fold improvement over the 20 km by 20 km grids common today. This level of detail lets us price climate risk at the scale of individual plots rather than entire regions, resulting in more accurate coverage.
Kertai: How do your models turn climate data into risk predictions and insurance triggers?
Yu: We use machine-learning models to predict the likelihood and severity of specific climate hazards at a given location, including droughts, floods, and coastal storm surges. The models estimate how often damaging conditions are likely to occur and how intense they may be for each place on the map.
To do this, we first group locations that behave similarly from a climate perspective—based on factors like rainfall patterns, soil characteristics, temperature variability, and coastal exposure. This step, known as clustering, helps us identify areas that tend to experience stress in similar ways. Within each group, we then build parametric models that track specific environmental indicators, such as soil moisture or wave height, and estimate how frequently those indicators cross damage-relevant thresholds. These predictions are what ultimately determine insurance pricing and payouts for individual plots of land.
Kertai: How do you build location-specific climate risk models when some places have little historical climate data?
Yu: In climate insurance, we call these data gaps—missing or unreliable local measurements, such as rainfall, soil moisture, or storm exposure, especially in regions without dense weather-station networks. These gaps make it hard to estimate risk accurately at the farm or community level.
We close such data gaps through satellite coverage, transfer learning, and local validation. Global satellite archives give us decades of consistent records on soil moisture, vegetation health, and precipitation, even in areas without weather stations.
We also use transfer learning, which allows a model trained in one region to apply its learned patterns to a similar region. For example, we identify comparable agro‑climatic zones and adapt validated models from places like Brazil to regions in Bolivia or Paraguay, then calibrate them for local conditions.
Finally, we invest heavily in ground‑truth validation. We work with development organizations to compare satellite‑derived indicators with real field conditions. During a drought in Bolivia, for instance, we collaborated with farmers and agronomists to confirm that our indicators matched what they observed on the ground. This human‑in‑the‑loop approach strengthens model accuracy and ensures we design the right insurance products for users.
Kertai: How do you maintain model reliability in the face of climate change?
Yu: Climate change is altering historical weather patterns, so our models must evolve continuously. We place greater weight on recent satellite and on‑ground observations to capture emerging trends rather than relying solely on historical data.
We also integrate forward‑looking outputs from global climate models and incorporate the latest climate‑science research into our updates. This helps us assess growing risks, such as rising storm‑surge levels or longer droughts driven by shifting rainfall patterns. We pair these updates with conservative pricing strategies to account for uncertainty and prioritize long‑term resilience.
Kertai: Can you share how Suyana is beginning to deploy its products in real-world markets?
Yu: In Bolivia, we’re working with four of the country’s five largest agricultural banks and its largest grain wholesaler to provide drought insurance covering 400,000 hectares. The 2023–24 growing season brought Bolivia’s worst drought in 30 years, devastating uninsured farmers and driving major loan defaults. Our embedded model bundles insurance directly with agricultural credit, so coverage comes automatically with the loan a farmer takes out for seeds, fertilizer, or equipment. This removes the need to search for a separate provider or navigate complex paperwork, barriers that often prevent smallholders from accessing protection.
In Peru, we’re piloting storm‑surge insurance for fishermen who currently rely on slow and inconsistent government subsidies when extreme weather shuts down ports. When ports close, fishermen lose their only source of income, and subsidies often take weeks or months to arrive. Our parametric product tracks wave heights and triggers payouts automatically when conditions force closures, offering speed and reliability the subsidy system can’t match.
