Climate change is the most pressing threat that the human species faces today. Artificial intelligence is the most powerful tool that humanity has at its disposal in the twenty-first century.
Can we deploy the second to combat the first?
A group of promising startups has emerged to do just that.
Both climate change and artificial intelligence are sprawling, cross-disciplinary fields. Both will transform literally every sector of the economy in the years ahead. There is therefore no single “silver bullet” application of AI to climate change. Instead, a wide range of machine learning use cases can help in the race to decarbonize our world.
Nearly every major activity that humanity engages in today contributes to our carbon footprint to some extent: building things, moving things, powering things, eating things, computing things. Of these, there is a subset of decarbonization initiatives to which artificial intelligence can productively be applied. And of these, there is a subset for which compelling startup business models exist.
Many billions of dollars of enterprise value will be created in the coming years as entrepreneurs chase down these opportunities.
And make no mistake: we have reached an important inflection point in the relationship between climate change and capitalism. Unprecedented amounts of capital have begun to flow toward climate-related initiatives. In the past few years, hundreds of the world’s largest companies have publicly committed to reaching net zero emissions and have begun to adapt their organizations’ operations accordingly. Last year, the largest asset manager in the world announced that it was making climate change central to its investment strategy. Governments’ central banks have even begun to take direct policy action as they come to view climate change as a systemic economic risk.
In short, tackling climate change is not only an urgent global imperative but a massive business opportunity. As investor Chamath Palihapitiya memorably predicted: “The world’s first trillionaire will be made in climate change.”MORE FOR YOUThese Are The Startups Applying AI To Transform HealthcareArtificial Intelligence And The End Of Work8 Leading Women In The Field Of AI
It is important to mention that, in addition to serving as a tool in the fight against climate change, machine learning does itself contribute to climate change given its voracious appetite for computing resources and thus electricity. This is an important ongoing challenge for the AI community, one that we have previously covered in this column. This article will focus on the ways that machine learning can help combat climate change.
Climate intelligence
One reason that companies are making climate change a strategic priority is that adverse weather events can and increasingly will wreak havoc on business’ core operations: disrupting supply chains, forcing mass evacuations due to wildfires, flooding facilities near coasts, halting outdoor activity due to extreme heat, making certain regions less habitable.
“People are beginning to understand that climate risk is not something in the distant future, it’s something that is happening today,” said MSCI’s global ESG head Oliver Marchand.
In order to prepare for and respond to these climate-driven disruptions, organizations need specific, actionable intelligence about the risks that they face. The rapidly changing environment means that simply looking to past weather patterns is not a reliable means of assessing future risk.
For instance, a bank might want to understand the expected impact of increased hurricane activity on property damage along the coast as it assesses its real estate loan portfolio. A government may seek to make targeted investments to bolster its country’s critical infrastructure in the face of more punishing weather conditions. An international hotel chain may find it worthwhile to better understand long-term weather patterns before deciding where to develop new properties.
If there is one thing at which modern machine learning excels, it is making predictions about complex systems based on lots of data.
A host of “climate intelligence” startups has emerged that offer predictive analytics platforms to enable organizations to better anticipate and prepare for extreme weather events. These startups generally combine machine learning with more traditional weather modeling techniques, leveraging a mix of publicly available, proprietary, and customer-specific data to train their models.
One of the oldest and most well-funded of these startups is One Concern, which announced a new $45 million funding round earlier this month. One Concern is developing a “digital twin” of the world’s natural and built environments in order to dynamically and hyperlocally model the effects of climate change, offering its customers what it terms “Resilience-as-a-Service.” The company has focused primarily on the Japanese market to date.
Jupiter Intelligence is another well-established player in this category, with an impressive set of blue-chip customers that includes NASA, BP, Liberty Mutual Insurance and the City of Miami.
Another climate intelligence competitor to keep an eye on is London-based Cervest, which raised a $30 million Series A last month from Marc Benioff, Chris Sacca and others. Cervest’s approach is to make its climate intelligence platform available to all with a freemium business model, a strategy it may hope leads to network effects.
Other startups in this space include Climavision, Gro Intelligence, ClimateAI and Terrafuse AI.
A key challenge that venture-backed companies in this category face is building offerings that are truly productized and scalable. Every prospective customer will have a unique set of climate questions and objectives based on its particular situation, geographical footprint, physical assets, business priorities, and so forth. This inevitably pulls these startups toward providing bespoke, professional-services-heavy solutions for each customer.
Consultancies can be profitable businesses but rarely achieve outsize, venture-scale outcomes. Time will tell whether one or more startups in this space succeed in getting enough leverage out of software and machine learning to build a scalable, category-defining technology company in climate intelligence.
Climate insurance
While the preceding category of startups aims to help organizations better understand and predict climate risks, another set of competitors is using AI to help organizations protect themselves financially from those risks. They are doing so by innovating in one of the world’s oldest businesses: insurance.
Legacy insurance companies struggle to effectively assess and price the novel financial risks posed by climate change. As a result, coverage can be prohibitively expensive or altogether unavailable for many types of climate risk. According to Aon, there is a whopping $171 billion gap in climate insurance globally.
A group of new entrants is leveraging alternative data sources, real-time analytics and AI to price risk more accurately and create novel insurance products for the era of climate change.
A key industry innovation at the heart of many of these startups’ strategies is known as parametric insurance. Unlike traditional insurance, which compensates policyholders for actual losses incurred, parametric insurance automatically pays out a certain amount upon the occurrence of a predefined event like a fire or storm. Parametric insurance eliminates the need for insurance agents to assess and verify individual policyholder losses, instead simply guaranteeing payment when certain parameters (hence the name) are met: for instance, a certain number of inches of rain, a certain heat index, a hurricane of a certain intensity, a certain reduction in foot traffic.
The parametric model makes insurance more automated, data-driven and transparent, with faster and more certain payouts. It requires sophisticated data and analytics capabilities to execute effectively.
Paris-based Descartes Underwriting, founded in 2018, offers parametric insurance for a wide range of climate-related risks including floods, droughts, supply chain disruptions, renewables yield, construction interruptions and more. Descartes uses machine learning to underwrite and monitor its parametric policies, ingesting data in real-time from a variety of sources including satellite imagery, stationary sensors, Internet-of-Things devices, radar and sonar.
Another player offering parametric insurance for climate risk is Arbol, which uses smart contracts on the Ethereum blockchain to codify its insurance policies. This enables it to automatically pay out claims in two weeks or less.
One of the buzzier startups in this category is Kettle, which offers reinsurance for climate change risk with an initial focus on wildfires. Using a proprietary neural network architecture, Kettle claims that it can make hyperaccurate, hyperlocal predictions about wildfire risk, including successfully predicting the locations of the 14 largest wildfires in California in 2020. Kettle’s AI-powered risk modeling enables it to offer more competitive pricing to its insurance customers, delivering 25%+ higher returns than traditional reinsurance players.
Other startups to watch in this category include Understory and Cloud to Street.
Climate change will reshape property and casualty insurance in the years ahead. This upheaval will create massive opportunities for data-driven, AI-powered challengers to grab market share in what has traditionally been a staid industry. Expect some big venture-backed winners in climate insurance.
Carbon offsets
In just the past few years, hundreds of the world’s largest companies representing trillions of dollars of market capitalization—from Amazon to Unilever, from Starbucks to Ford—have publicly committed to achieving net zero emissions on a specified timeline, often by 2040 or 2050.
Any organization seeking to reduce its carbon footprint has two basic ways to do so: directly eliminating emissions from its operations and buying carbon offsets. The latter method, while controversial, is poised to play a central role in the years ahead in the global fight against climate change.
While still relatively small, the carbon offsets market has been expanding rapidly in recent years: from $34 million of offsets issued in 2016, to $73 million in 2018, to $181 million in 2020. Some experts believe the market is on the brink of explosive growth, with companies around the world preparing to go on an offsets buying spree. Influential climate financier and policymaker Mark Carney has stated that carbon offsets could be a $100 billion market by 2030.
The concept of a carbon offset is straightforward in principle: one party pays for another party, anywhere in the world, to eliminate an agreed-upon quantity of greenhouse gases from the atmosphere through emissions reduction or recapture. Common examples of offset projects include planting trees (which soak up carbon dioxide) and financing renewable energy infrastructure like wind turbines.
In theory, if an organization generates one ton of carbon emissions and then funds an offset project that eliminates one ton of carbon emissions, that organization has had zero overall adverse impact on climate change.
That’s in theory. In practice, carbon offsets are the source of controversy and operational complexity.
Most fundamentally, some observers object to carbon offsets on the grounds that they artificially absolve emitters of their climate sins, allowing them to throw money at the problem while avoiding the hard work of adjusting their own activities to reduce emissions. There is some truth to this perspective: ideally, carbon offsets should be used only as a final step to get to carbon neutrality, after an organization has reduced its own emissions as much as it can. Offsets by themselves will never get us to a zero-carbon world.
Yet in the immediate term, carbon offsets are an important tool that leverages the power of the market to reduce emissions. Especially given that certain essential human activities—air travel and heavy industry, for instance—are for basic technological reasons unlikely to be carbon-free any time soon, offsets are the only way for most organizations to get all the way to net zero emissions.
A more tactical problem with carbon offsets is that they are difficult to operationalize at scale. Coordination problems loom large given that the buyer and the seller can be on opposite sides of the world. Verifying the legitimacy of a carbon offset project—for example, that a tree has actually been planted, that it would not otherwise have been planted (“additionality”), that it stays alive and continues to grow over time (“permanence”)—is operationally daunting. These challenges have limited the size of the offsets market to date.
To tackle these challenges, an exciting new group of startups is applying software and machine learning to streamline, digitize and automate the carbon offsets market. These companies believe they can unleash massive pent-up demand in offsets and serve as the backbone of a new multi-billion-dollar industry.
Pachama and NCX (formerly known as SilviaTerra) are two promising companies building AI-powered carbon offset marketplaces with a focus on forestation. Both companies apply computer vision to aerial imagery and other sensor data to automatically estimate the carbon stored in forest trees and to continuously monitor the integrity of carbon offset projects on their platforms.
One area in which the two companies differ is their approach to the supply side of the marketplace. While Pachama selects a curated set of forestation projects from which users on its platform can buy offsets, NCX’s approach is more radically democratized: any individual landowner, no matter how small, can join the platform and sell carbon credits in exchange for a commitment to preserve trees.
“As climate change becomes a central topic on the agenda of corporations and governments, it’s absolutely necessary that we improve the state of carbon offsets markets, ensuring integrity, transparency and accountability,” said Pachama CEO Diego Saez Gil. “If we get it right we could help finance the restoration of tens of millions of hectares of forests, removing gigatonnes of CO2 from the atmosphere.”
Another company to watch in this category is Patch, which recently raised a $4.5 million seed round from Andreessen Horowitz. Patch’s platform abstracts away the complexity of managing carbon offsets, making offset projects accessible via an API and a few lines of code. Behind the scenes, the company vets and partners with a handful of high-quality offset organizations. With its API-first approach, Patch CEO Brennan Spellacy describes the company’s vision as “Plaid for decarbonization.”
One major question that these companies will face as they pursue commercial scale is how motivated organizations truly are to spend money on carbon offsets. Skeptics argue that, while there is no shortage of corporate rhetoric around long-term net zero emissions goals, many companies will hesitate to voluntarily commit significant amounts of capital to offset their emissions.
The bet that these startups (and their investors) are making is that organizations will take their net zero commitments seriously and will invest real dollars, sooner rather than later, to make progress toward those commitments.
Moreover, the entire complexion of this market could change if governments start establishing more rigorous mandates around carbon emission limits. This has already begun to happen in some jurisdictions. Such regulation could transform carbon offsets from a nice-to-have into a legal necessity for thousands of organizations around the world.
Carbon accounting
In order to reduce or offset its carbon footprint, an organization must first understand what its carbon footprint is. This is a challenging, messy, data-intensive process.
A company’s overall carbon footprint can be broken down into three categories: direct emissions from the company’s own operations (known as Scope 1 emissions), emissions required to generate the electricity that the company uses (Scope 2 emissions), and—most challenging to measure—emissions that go into the production and consumption of the company’s products across its value chain, from upstream suppliers to downstream customers (Scope 3 emissions).
Two illustrative examples of Scope 3 emissions: Sweetgreen includes in its Scope 3 emissions the methane emitted by cows that produce the cheese in its salads, while Square counts the carbon output from the electricity that individual merchants use to power their Square Registers.
Over the past year, a crowd of startups has emerged to provide tools to help organizations measure and track their carbon emissions, from Scope 1 to Scope 3.
These startups’ product visions go beyond simply helping companies quantify emissions. Once a company has a comprehensive view of its carbon footprint, it can then develop and execute on a data-driven plan to reduce its emissions: for instance, switching to renewable electricity sources, adapting its real estate footprint, pushing its suppliers to adopt more low-carbon practices, providing employees with information to make more sustainable daily decisions, or (as discussed above) buying carbon offsets.
The most high-profile company building an enterprise carbon accounting platform is Watershed, recently founded by a team of Stripe veterans. Watershed has attracted buzz in no small part because it is backed by Sequoia’s Michael Moritz and Kleiner Perkins’ John Doerr; the last early-stage startup these two iconic venture capitalists teamed up on was Google in 1999.
“Climate is a data problem,” said Watershed cofounder Taylor Francis. “Organizations make decisions every day with carbon impact—which supplier to procure from, how to transport a good—yet decision makers are blind to the carbon scores of their choices. At Watershed, we are building tools to unlock this insight and turn raw business data—from utility bills, purchase orders, transit logs, and so on—into concrete actions that reduce carbon emissions.”
Other competitors in this fast-developing space include Emitwise, SINAI Technologies, Persefoni and CarbonChain.
A fundamental challenge that these startups face is one of data wrangling and data quality. Scope 3 emissions in particular can prove prohibitively difficult to collect reliable data on.
For example, imagine an organization with a complex supply chain stretching across several foreign countries, including various intermediary suppliers who are unwilling to share detailed information about their operations and energy use. Developing a detailed picture of this company’s Scope 3 emissions would be difficult indeed. As the age-old saying about data goes: garbage in, garbage out.
Devising an accurate, repeatable, scalable means of collecting carbon emissions data from around the globe and across a product’s lifecycle will be a key unlock for startups in this category.
It is worth noting that some startups in this category are developing carbon offset marketplaces as a natural extension of their carbon accounting platforms. This will bring them into direct competition with the offset startups discussed above as these companies’ product visions converge.
And as with the previous section, an existential question facing the startups in this category will be how much appetite customers have to voluntarily part with hard-earned cash in order to build out robust decarbonization programs.
Buildings
Buildings produce close to one-fifth of the world’s total carbon emissions. Making our buildings more efficient is therefore essential to the fight against climate change.
The good news is that much low-hanging fruit exists, because buildings today are massively underoptimized for energy efficiency: the carbon footprint of many existing buildings can be reduced by as much as 90% via retrofit strategies.
The single largest opportunity for emissions reductions in buildings is in heating, ventilation, and air conditioning (HVAC). Heating and cooling buildings is incredibly energy-intensive today, accounting for roughly half of all the energy that buildings consume.
HVAC is a complex, well-defined, data-rich, multivariable system: in other words, it is ideally suited to optimization via machine learning (in particular reinforcement learning).
In 2016, DeepMind pioneered the application of machine learning to HVAC systems with a widely-discussed study that sought to improve energy efficiency in Google’s massive data centers. Using deep learning to optimize the data centers’ cooling systems, DeepMind was able to reduce the facilities’ overall energy consumption by up to 40%. Considering that data centers are responsible for 2% of total greenhouse gas emissions, this was a landmark result.
One of Google’s data center operators shared an example of how DeepMind’s system worked: “It was amazing to see the AI learn to take advantage of winter conditions and produce colder than normal water, which reduces the energy required for cooling within the data centre. Rules don’t get better over time, but AI does.”
A handful of startups is seeking to develop HVAC optimization solutions for buildings.
Montreal-based BrainBox AI claims that its software can reduce a building’s carbon footprint by 20% to 40% within a few months by making precise, localized, real-time microadjustments to the building’s heating and cooling settings. BrainBox’s technology, which does not require the deployment of sensors, is live today in dozens of residences, hotels, airports, nursing facilities and grocery stores, impacting over 100 million square feet of real estate in total.
Bill Gates-backed startup 75F also uses machine learning to monitor, automate and optimize HVAC systems in buildings. Other competitors in this space include Nomad Go, which has developed a computer vision-based solution to streamline buildings’ energy use.
Importantly, these companies’ offerings do not just reduce carbon emissions—they also cut costs for building operators. BrainBox claims that its technology saves customers up to 25% on their energy bills while also improving building occupant comfort by 60%. This sort of incentive alignment is crucial for the broad adoption of decarbonization solutions.
Ultimately, energy consumption in buildings will never be fully sustainable until the grid’s electricity itself comes from zero-carbon sources. But in the nearer term, machine learning offers us actionable opportunities to reduce buildings’ carbon footprints.
Precision agriculture
Agriculture is a major driver of climate change, accounting for between 10% and 15% of the world’s greenhouse gas emissions.
Modern agriculture is resource-intensive and wasteful. For instance, over 200 million tons of fertilizer are used for farming every year, millions of which are wasted due to imprecise and excessive application. This is a major problem for climate change: fertilizer alone is responsible for 2.5% of all greenhouse gas emissions. Worse, the greenhouse gas that fertilizers generate, nitrous oxide, is especially harmful: pound for pound, it warms the atmosphere about 300 times more than does carbon dioxide.
A massive opportunity exists to apply digital technologies to make agriculture more efficient, reducing its carbon footprint while increasing food yields.
Precision agriculture is the practice of optimizing crop inputs (e.g., fertilizer, water, pesticides) on a targeted, localized basis, sometimes even plant by plant, rather than indiscriminately treating all plants the same across entire fields and farms. It will play a central role in making agriculture more sustainable in the years ahead.
According to the World Economic Forum, if 15% to 25% of farms adopted precision agriculture techniques, greenhouse gas emissions could be reduced by 10% and water use could be reduced by 20%, all while increasing farming yields by 15%.
AI startups are playing a key role in making precision agriculture a reality on farms around the world.
One approach is to apply computer vision to aerial imagery in order to give farmers real-time insights about how to optimally deploy resources on their farms: where to apply more or less fertilizer, where to fix leaking irrigation pipes, and so forth. The appeal of this business model is that it is software-based and thus capital-efficient and scalable. Startups pursuing this strategy include Ceres, Hummingbird Technologies, Gamaya and Prospera (which was acquired for $300 million just last month).
Semios and Arable are two well-funded precision agriculture startups that use stationary on-the-ground hardware sensors to enable more precise crop management. Semios says it has installed over 2 million sensors on farms to date, from which it collects over 500 million data points every day.
A final set of competitors, including Bay Area-based startups FarmWise and Bear Flag Robotics, is deploying physical robots on farms in order to more directly carry out AI-guided precision farming. FarmWise’s initial focus is on weeding, with a longer-term vision to use its robots to enable precision agriculture across a broad spectrum of farming activities. Bear Flag Robotics has developed an autonomous tractor service.
It is important to note that many of these startups do not explicitly position themselves as “climate technology” companies. Precision agriculture technologies get market adoption first and foremost because they make farms more productive and efficient, saving costs and boosting output. The fact that they also drive decarbonization in one of the world’s largest carbon-emitting sectors is a fortuitous side benefit. As with HVAC optimization, discussed above, startups that align emissions reductions with economic value creation for customers are best positioned to succeed.
Expect to see multiple large companies built in precision agriculture in the years ahead.
Renewables and the grid
Producing zero-carbon electricity is at the very heart of the fight against climate change. As Bill Gates put it in his most recent book: “If a genie offered me one wish, a single breakthrough in just one activity that drives climate change, I’d pick making electricity: It’s going to play a big role in decarbonizing other parts of the physical economy.”
The fundamental breakthroughs needed to make abundant zero-carbon electricity a reality—better energy storage, next-generation nuclear fission, viable nuclear fusion—are first and foremost physical engineering challenges. Today’s artificial intelligence cannot serve as a silver bullet to produce these basic advances in physics and chemistry.
Yet there are various ways in which machine learning can help improve today’s electricity systems and move us toward a renewables-centric future.
The power grid is one of the most complex systems that humans have ever built. Because electricity cannot be efficiently stored at scale, the grid must continuously balance supply and demand in real-time. Machine learning can help automate and optimize this complex system, enabling grid operators to more accurately forecast electricity flows and to eliminate inefficiencies that lead to greater carbon emissions.
Winnipeg-based Invenia and Y Combinator alum Gaiascope are two interesting startups applying AI to predict and optimize electricity grid dynamics.
In a similar vein, Gridware uses edge AI and telemetry sensors to predict and detect faults in the grid’s physical infrastructure in real-time, reducing the risk of fires and other systemic failures.
Another player applying AI to help decarbonize the grid is Raptor Maps, a computer vision-based platform that facilitates the deployment and management of solar energy assets using data from drones, satellites and on-the-ground sensors.
Artificial intelligence can also contribute to the proliferation of renewable energy through improved materials science. For instance, researchers are applying machine learning to help discover new compounds that can harness the sun’s power more effectively than today’s photovoltaics.
One promising early-stage startup using AI to enable the discovery of new materials is Israel-based Materials Zone.
Fires
One of the most visible and destructive reminders of our warming planet is the alarming increase in wildfires in recent years, from California to Australia.
2020 was the worst wildfire year in recorded history in California, eclipsing the previous record set in 2018. Over 4% of the state burned, resulting in more than $12 billion in losses.
As climate change makes the planet hotter and drier, we face the prospect of devastating wildfires becoming the new normal, year in and year out. It is imperative that we do what we can to mitigate this climate risk. AI can help.
Pano AI and Fion Technologies are two young startups that are applying machine learning to bolster firefighting efforts. Their solutions use computer vision to identify where fire risk is greatest ahead of time (so that fire professionals can take preventive measures), to detect fires immediately when they start (so that firefighters can swiftly respond before the fire gets large) and to predict how fires will spread (in order to aid firefighting efforts in real-time).
Firemaps, meanwhile, is building a product to help individual households take steps to protect themselves from fires. Based on a customer’s home address, Firemaps analyzes satellite imagery and other fire risk data to automatically generate a personalized fire defense plan for the property, including actions like brush removal, fuel breaks, and home hardening.
Conclusion
Artificial intelligence is a general-purpose technology with infinite potential use cases. There is perhaps no AI application that matters more for humanity than decarbonizing the atmosphere and slowing climate change. The opportunity for economic and societal value creation is virtually unbounded.
It is hard to imagine a more worthy field for AI entrepreneurs, researchers and operators to devote themselves to in the decades ahead.