When Sarah Russell started working in climate adaptation and resiliency for X, Google’s moonshot factory, she didn’t see it as straying too far from her training as a doctor.
“Working on climate resilience is working on the health of humans, and human resilience,” she told Latitude Media.
The combination of Russell’s background in healthcare entrepreneurship and Google’s vast data capabilities led her to found Bellwether, which officially launched last April. The software platform applies artificial intelligence to huge quantities of data about the physical world to help people prepare for, predict, and recover from natural disasters.
“The core technology works with the behemoth of Earth data — petabytes and petabytes of Earth observation data going back 20 years — that we organize in a data pipeline that we then apply machine learning to, in order to ask a question or make a prediction,” Russell said.
There are three products built on Bellwether’s core technology, each focused on a different stage of the evolution of a natural disaster: a prediction tool focused on risk maps for wildfires, a rapid response tool that quickly assesses a scene after a natural disaster, and an insurance tool that helps insurers evaluate the overall risk across a portfolio of assets.
Using Bellwether Predict, for example, one could estimate the probability of a building being in a catastrophic wildfire one or five years into the future, with an accuracy that Russell says is around 90%. The tool also allows users to “quantify the impact of an adaptation effort on the overall risk.”
While Bellwether only made its debut earlier this year, the tool is already in use. An unnamed Western state used the platform to plan how to best use its wildfire mitigation budget. Bellwether ran an analysis of all the trees and forests under the state’s control, and helped the state select the areas where vegetation thinning or a controlled burn would have the highest return on investment.
And Bellwether’s rapid response software helped advance recovery efforts in the aftermath of Hurricane Helene, which struck the East Coast earlier this year. Russell said the tool processed 75,000 images captured by Civil Air Patrol planes, matching them to their locations and assessing the damage inflicted to infrastructure — which is something that was previously done entirely by humans.
Geo-foundation models
Something Russell is excited about in the coming years are geo-foundation models: large-scale deep learning models trained on geospatial data.
“They’re the geo version of large language models,” she said. “They’re around three years behind [LLMs], but they’re emerging now….That work is in its early prototype stages across the Google ecosystem, and I think it has the potential to be transformative.”
Those models will replace the need for Bellwether to process and organize geophysical data quickly, Russell added — as well as “make it possible for us to build these products more quickly, more accurately, across more geographies, in ways that compel people to act,” she said.
Bellwether’s main goal is to have an output that helps people drive investments in adaptation and resilience, a space that remains underfunded despite its urgency and potential.
Over the past year, Russell says she’s seen an uptick in attention, especially on the customer side. The risk-transfer space, in particular, is becoming an important end-user of adaptation solutions.
“Insurance is going to play a huge role in adaptation,” Russell said. “Their business model is based on underwriting risk, and selling and providing [them with] more granular data is a really good idea…Understanding how risk is correlated across a portfolio and how additional properties either lower or raise risk is hugely important to the insurance industry.”
Additional promising customers for Bellwether are the public sector and utilities, which are keenly aware of the risks of wildfires, and have been contributing to surge in firetech startups.
Bellwether is already commercially active as a part of Google’s X, but it’s yet to be seen how it will evolve, Russell said.
“Alphabet relies on X to generate products that might land back in the Alphabet ecosystem, might become another bet, or might spin out,” she said. “We could land in one of these three places.”
