One pretty obvious aspect of the climate emergency which may have flown under your radar is that human-driven global heating is disrupting traditional approaches to risk modelling around natural disasters since probabilistic models based on stuff that happened in the past start to come unstuck atop so much unprecedented change.
Step forward, a climate tech startup out of Barcelona, Spain that’s just raised a €13.25 million (~$14.4M) Series A funding round to grow usage of its risk modelling tools. The round was led by Kibo Ventures, with Microsoft Climate Innovation Fund, Nationwide Ventures, Faber Ventures, and CREAS Impacto also participating.
The startup is taking a data-heavy, physics-based approach to predicting climate-driven risks, such as wildfires, extreme weather and even volcanos, so it’s also modelling climate variability that’s being driven by climate change — applying high performance compute-driven risk modelling augmented with AI, including techniques like transfer learning so it can offer predictions in regions of the world where there’s a lack of high quality data to inform its models.
Mitiga’s customers, which number around 20 (recurring clients) at this stage a few years in, are businesses with asset management needs, such as energy manufacturers; or entities that need to collateralize risk, such as financial services companies, hedge funds, real estate companies and insurance firms. In the latter camp it names Axa Climate and Willis Towers Watson as among its customer roster. It also has customers in the humanitarian sector (such as the Red Cross).
Its pitch is that it’s able to more accurately model climate risks vs traditional risk modelling approaches that have relied upon “more stochastic and probabilistic approaches” — doing so in a way that’s price competitive vs traditional risk modelling despite the use of lots of high performance compute (thanks to optimized code and strategic use of cloud computing, as it tells it).
High performance computing is necessary to power such high resolution mathematical simulations of complex systems, with Mitiga’s algorithms crunching massive data-sets on physical and other conditions in order to get as close to predicting what’s going to happen as possible down to a scale of 30 meters (or even 10 meters in some locations).
CEO and co-founder, Dr Alejandro Martí, tells TechCrunch it’s loading more than 5 million data-points into its models every day. “Models are as good as the data you drive them with,” he emphasizes. “So you need to have massive amounts of data set.”
“Traditionally, risk modelling companies, whether they come from the insurance or the financial sector, they’ve been using more stochastic and probabilistic approaches to determine what the risk is. So that will be you look at stochastic analysis from the last 100 years. Then try to predict some trends. And then you apply those trends towards the future,” he goes on, explaining the drawbacks with traditional risk models in our rapidly warming (and dangerously changing) world.
“One of the things that we have seen in the past few years is that climate change is changing, especially, the tails of the distribution, the extreme events — so these massive wildfires that we’re seeing, the floods, the tsunamis, etc — we’re not seeing these kinds of events and the magnitude of these events represented in those long term distributions. So, obviously, this is the impact of climate change [which means there are] more events and the events and the magnitude of the events are harsher or bigger. So the traditional probabilistic models they are a little bit obsolete.”
Physics-based risk modelling means building mathematical simulations of complex systems by applying physical laws and principles to masses of data (on local conditions and dynamic variables) to perform predictive analytics of risks at a given location. This of course demands lots of high quality data. And, clearly, such data is not available everywhere in the world. Hence why Mitiga is using transfer learning to plug gaps so it’s able to sell risk modelling with a “global footprint”.
Martí says it’s using AI techniques to create data-sets for locations where there is a lack of high quality data to feed its models. He describes this as a process of transferring data “from data rich countries to data poor countries” — explaining it’s relying on “proxies”, such as from similar topography and environmental and/or metrological conditions, to create underlying data sets to build models for less “data rich” locations.
He caveats this by noting that the accuracy of the risk modelling does vary, depending on how much high quality (vs proxy) data is available for a particular location and risk context. “I always say when you speak about a global model it’s a lot of regional models that are customised to have a global footprint and that’s the best accuracy that you can have,” he tells us. “But there will be places where the uncertainty is high and then you just need to be honest. That the uncertainty is high — or, like, how you can mitigate that uncertainty.”
He also concedes that physics-based modelling is gaining ground with traditional players in the risk modelling business. So it’s not the only game in town. But while risk modelling startups have been growing in number in recent years, as entrepreneurs lock onto the risk-opportunity driven by climate change, on the competitive front Mitiga can claim a pioneering edge since it was spun out of Spain’s(also Barcelona-based and home to Europe’s ).
That happened back in early 2018 — so still relatively recently — but the team is well-versed in this kind of specialized, high performance computing-driven hard-math climate-risk modelling, with founding staff having been at it for some 10 years when they were working as researchers at the NSC.
Add to that almost half (40%) of Mitiga’s 30-strong team holds a PhD. (And Martí notes that the Series A funding is being earmarked for further expanding its talent pool to scale and accelerate the risk modelling capabilities.)
Martí himself, who holds one of the PhDs, cut his teeth in climate science working for the US government on then-emergent geospatial technology for around a decade, back in the noughties, including looking at the link between geospatial tech and modelling risk for climate. After that he came back to Europe to do his doctorate, at Exeter University in the UK, on a program managed by Cambridge University where he worked with the Met Office scientific developing climate models. So this is a climate tech startup built on a very solid foundation of deep science.
The team’s focus for the product at this stage is on modelling risks around so-called “secondary” perils — or what it describes as events which are “heavily impacted by climate change”.
This means — not earthquakes or flooding (which the insurance industry classes as primary perils) — but aforementioned climate-linked risks such as wildfires, extreme weather and drought. The risk of volcanic eruptions is another on this focus list which stands out as a bit different. Albeit volcanic eruptions can certainly contribute to climate change (and therefore to climate risk) by spewing out emissions and aerosols which can increase heating. (Plus, per Martí, there is some live scientific debate about a possible feedback effect where global warming might be increasing volcanic activity. So, er, yikes!)
Despite secondary perils having a merely modestly scary-sounding label, Martí notes that the associated insurance industry loss ratio has already flipped, meaning secondary perils now (collectively) account for more than 50% of the insurance industry’s losses (which used to be the case for primary perils). Which suggests the risks they pose to human life are also on the rise. So they are probably in need of a rebranding.
Add to that, given these hazards are the ones really impacted by climate change, the dangers that they pose (and their ability to drive massive commercial losses) are only likely to grow in the coming years (unless or until humans actually manage to stop heating the climate). Hence why Mitiga reckoned it had spotted a risk-modelling opportunity-gap to lock onto.
Its marketing also talks up the opportunity for customers to act on the risk data to mitigate even worse climate harms by making proactive interventions aimed at stopping a potential hazard from turning into a full-blown natural disaster. Of course this doesn’t mean that data and fancy modelling can stop tornados or prevent the heavens from opening. Rather the idea is the tool can arm businesses with intel to proactively adapt and improve their resilience to likely risky events. Such as, for example, installing certain types of windows that can reduce the impact of extreme heat inside buildings, or adapting buildings and other physical installations to make them more resilient to water ingress.
In the coming years, many (if not all) businesses will need to consider how to adapt their assets and operations to the havoc being rained down by climate change. And, clearly, risk modelling schemes that can help enterprises prioritize what to tackle first is an elementary tool for them to reach for.
Add to that, incoming regulations in Europe (and elsewhere) requiring businesses to score climate-related risks to their assets will drive uptake of this sort of climate tech — likely pushing it far beyond the usual suspects (i.e. insurance firms) whose businesses give them a particular interest in risk modelling. And on this front Martí notes that Mitiga will shortly be launching what he refers to as a “global climate score” which is aimed at helping customers comply with climate risk regulations.
“The climate score is targeted not only to the insurance sector but any asset manager… so financial institutions, real estate, you know, hedge funds, etc,” he says, adding: “We’re having a lot of traction on that because, for example, these [EU Taxonomy-related] regulations went live in Europe in January 1 2023, and even though they have about a year or two to adapt obviously this is the next thing that everyone is going to have to comply with.”
Transparency around the risk predictions it provides to its customers is another element of differentiation he highlights vs traditional players.
“If you’re going to have to assume your risk, based on our models, it’s only fair that we tell you what is the uncertainty associated to the model. So that’s something that our clients appreciate,” he says. “In this sector there are a lot of black boxes, and a lot of decisions are made with those black boxes — which has a financial impact but it also has a social impact. So I would say that the combination of technology, transparency and know-how is what makes Mitiga a contender to challenge the traditional model providers.”
The startup is not expecting the (risk-averse) insurance industry to switch away from traditional risk modelling providers en masse and overnight. Rather it anticipates being able to build traction on the side — by offering more customers modular terms (vs traditional risk modelling players’ per-market-based licensing) — enabling clients to try the tech and “start de-riskifying their portfolios”. From this supplemental position it hopes to keep scaling the business (and “growing up” as a company), setting its sights on becoming “a true contender for them to consider as one of the main providers” down the line, as Martí puts it.
Commenting on its Series A funding round in a statement, Javier Torremocha, co-founder and managing partner at Kibo Ventures, said: “There is a lot of potential and resilience in climate technology. We have been impressed by what Alex and the team have built; a proprietary state-of-the-art technology with multiple applications. We are delighted to support Mitiga with its vision to become a category leader while helping to reduce climate change disasters.”
In another supporting statement, Brandon Middaugh, senior director at Microsoft Climate Innovation Fund, added: “The ability to predict and manage the effects of climate-related hazards is a critical need to adapt to a changing ecosystem. Mitiga‘s use of AI and high performance computing is a valuable tool to assess climate-related risk across a variety of hazards to mitigate threats and build a more resilient future.”
Given the current precipitously high levels of hype being attached to AI — which, just earlier this week, featured a turn in the global spotlight by OpenAI’s CEO Sam Altman (of ChatGPT fame) who suggested to a US senate committee the tech might one day help humanity fix climate change, even as he simultaneously talked up the vast potential for generative AI to power all sorts of societal harms — TechCrunch took the opportunity to ask for Martí’s long view on what AI might (realistically) be able to do vis-a-vis the climate crisis.
“There are things that AI can help and things that AI is not going to resolve,” he predicted. “You cannot have artificial intelligence resolve something that hasn’t happened and be right about it. Artificial intelligence builds, again, on the past, understanding the trends of the future. But it’s nothing about the problem itself. It’s about the trends.
“When you go into climatic scale, the noise of the climate models themselves, between years, is so high that you cannot resolve that [variability]. So AI again, continues to be a tool… that complements other things. At least in our space.”
That said, he wasn’t willing to look too far ahead in capability forecasting here — cautioning: “If we fast forward 10 years from now, it’s super exciting and scary at the same time what AI can do.”
NB: Mitiga Solutions is no relation to the which covered previously