Climate and AI: landscape survey
As global temperatures and carbon emissions accelerate climate change, the need for innovations with adaptive and mitigation strategies is becoming more apparent. Global average temperatures have increased by 1.3 degrees Celsius since pre-industrial times, leaving many ecosystems vulnerable to collapse (California Air Resources Board). Furthermore, carbon emissions into the atmosphere have caused a myriad of environmental issues—ocean acidification, intense weather, and frequent droughts. As temperatures rise, these problems are further exacerbated and can create positive feedback loops, exponentially increasing climate related disaster risks. As a result, climate change response is needed more than ever to reduce crisis-inducing effects. With this, there are still efforts to help reduce the damage caused by climate change currently. Renewables are cheaper and more efficient. Legislative efforts in architecture, agriculture, land management, conservation, and transportation have created incentives for firms to begin working on innovative solutions (Energy Efficiency Office). As technology has developed, climate response has been met with software that provides a much needed role in climate response.
Instead of playing direct roles in climate response, software allows a platform for management processes to gain access to highly sophisticated data that can allow for informed decision making.
Unifying Climate Data
The Climate Model
Data Analysis Challenge
Fused
Wherobots
Detecting Hazards
Pano AI
AI Dash
SmokeD System
Strengthening Carbon Neutrality
Pachama
Treeconomy
Conclusion
Works Cited
Unifying Climate Data When tracking climate change data, there are a variety of different methodologies utilized to visualize and summarize points that can present compelling results. Simple statistics can utilize graphs to create connections between factors like emissions and vegetative mass, and maps can highlight different topographical features to determine mineral compounds. Alongside this, warm current patterns and hurricanes can be correlated together to form new connections with other data patterns. However, data is not unified—different entities utilize different methods of data collection to serve their tasks, involving different types of software and mapping platforms (CCAI). Much like the fragmented network of wildfire data tracking, the lack of compatibility between data sources can create massive issues when attempting to create useful correlations, increasing the cost and time of operation.
The Climate Model There is a major way climate scientists can get around this problem, which comes in the form of a climate model.
A climate model is a multidimensional data model that aims to combine key parts of climate visualization — sensor detail, observational data, and general laws of climate physics — together to form an overall picture of a regional or global climate.
These data-driven models are representations of the science community’s understanding of all the functions that drive our climate today, and creating them can help scientists form new connections and findings for climate change. However, climate models are still costly to synthesize, as modern software comes in a variety of different data types.
Data Analysis Challenge While AI can provide useful insights to help make important environmental decisions, the use case for combining geospatial data may only need a platform that can effectively plug in data from a wide variety of file types to purposely find new connections by itself. This can create new ways for firms and scientists to develop models that help find solutions to a myriad of issues.
Fused One such company focused on pairing together mapping data is Fused, a startup that utilizes User Defined Functions (UDFs) and numerous application integrations to build and render maps. Furthermore, Fused also holds a public community board where creators can share their projections and find new connections. One way in which Fused accomplishes its seamless capabilities is through its Python-based Integrated Development Environment (IDE) , which the company claims can ease up many of the inconsistencies and difficulties of other integrative mapping tools. According to Fused, issues with many mapping tools today come in many forms: price, complexity, cost, and scalability. In their troubles with integration, mapping tools can have steep learning curves, and the payoffs for their results can often take up power and memory, reducing the scope of a project. Larger scale tools can have high prices, but lack the data science needs of more specific projects. The Fused model claims to circumvent this issue entirely by their usage of serverless computing. According to the research group Climate Change AI (CCAI), machine learning networks can provide an in-route into speeding up the compiling and parsing of various forms of data found in mapping. Because data can be so fragmented, the cost and speed of an interpretative service outside of humans may become essential to creating climate models— rendering and displaying models end up being half the picture of making useful predictions for the future, and gleaning useful data out of them is the more important aspect. AI can start developing on this other half by taking geospatial data to make new inferences, giving researchers and scientists more information on unraveling the nuances of climate change.
Wherobots A major aspect of creating models depicting climate depends on mapping efforts done by various government organizations and private firms. They can have different data types that contain these maps, and combining them without interpreting them properly together can render the data useless for research and decision making. This issue can be solved by AI in two ways: Create a new data infrastructure that utilizes past data as a foundation for the AI, which then learns and grows off the new data it collects This is the current method that Pano AI uses to improve its wildfire detection model Patch together various data sets from various data types and use AI to interpret the data The latter of these methods is the approach of Wherobots, a startup taking an AI based approach to reformatting geospatial data. Their model, much like Fused, aims to simplify and deal with the difficulties combining and interpreting mapping data, but aim to store and process data using a cloud based model. Many of these use cases stretch beyond the world of climate change, but utilize climate data to create predictions on a variety of economical factors. Some examples listed by the company website included tracking efficient truck driving routes and home insurance risks, but the potential uses for this model can stretch way beyond that. Geospatial mapping data can be used to create highly sophisticated climate models, and affects every aspect of both civil and scientific planning. Adaptive measures that minimize climate damage, speed up rescue response routes, prioritize risk-prone areas, and more utilize geospatial data to achieve the best method of action. Their AI interpreter can then further determine effective response by finding new connections between different geo-spatial data models.
Detecting Hazards One major way that AI can be used in adaptive and mitigative responses is to differentiate hazards and be able to accurately predict them. Utilizing large scale data as a base for training, AI can not only learn to detect if an event is a hazard, but also determine the severity and scale of response to deal with it (CCAI). This is especially useful in fast and aggressive weather events, including forest fires, tornadoes, and flash floods. Wildfires are an especially dangerous occurrence that have long lasting consequences on people’s living communities and the environment around them (Cal Fire). In the state of California, wildfires have easily bordered many metropolitan areas, often burning down smaller woodland cities and crippling power lines. Wildfires are now more common than ever. A study done on 7,000 large wildfires in the Rocky Mountain ranges showcases a 73% increase in wildfire burns between the years of 1984 and 2011, with an average increase of 18 fires per year (Funk and Saunders, 14). The reasoning for the rise in wildfires is attributed to an increased quantity of developing biomass formed after an era of low wildfires caused by agriculture. By eating away at brush and wildgrass, wildfires had initially lost their conduits to cattle, but accumulated biomass and destructive power line ruptures now increase their frequency and potential for destruction (Funk and Saunders, 15) . The present results are clear, and concerning— According to the California Fire Department, the most destructive wildfires have occurred in the last 25 years— the Camp Wildfire of 2018 had destroyed over 18,000 structures alone, becoming California’s most expensive and destructive wildfire ever. Due to the speed of wildfire growth, efforts to reduce wildfire damage have been primarily concerned with early detection and prevention. Large scale wildfires can spread fast and unpredictably, and if not dealt with in time, can quickly become uncontrollable. As a result, local authorities often invest in public safety messages and restrictions on outdoor fire use. Furthermore, in order to prevent large and devastating wildfires, authorities may also conduct controlled burns in order to reduce the total amount of biomass accumulated in fire-prone areas (Cal Fire). To detect early wildfires, fire departments currently employ large data sets tracking fire-prone areas and historical areas of spread to determine their scale of response (CCAI). Parsing through this large data set can be difficult, due to the overwhelming number of data sources paired with a lack of a unified data set. AI finds an important usecase here, as it can parse and interpret through data maps quickly and be able to give an effective response.
Pano AI Startups such as Pano AI aim to instead build an entirely new network of data collection systems specially designed for their AI model– one that they claim can accurately detect and bring focus to wildfires. Pano Rapid Detect, their AI based detection model, uses specially designed cameras that cover a 15 kilometer radius. Pano claims to have their product be able to rapidly zoom in and locate wildfires, while also providing contacts and information to local authorities by compiling all of this data and pairing them with satellite and geospatial imagery. After this, a fire correspondent will then determine and dictate the level of response needed and communicate this to local authorities. Furthermore, the system also provides authorities the ability to share information with citizens for quicker emergency responses. There are many benefits to Pano’s detection system; the ability to have native camera systems integrated with an AI system removes much of the data parsing required for other models to detect fires. Furthermore, a sophisticated camera system with significant reach allows it to have a moderately wide range and provide quick response. The major flaws with this system seem to rest with the scalability of this model, mainly involving the price and budgeting required to build such an expansive system. Furthermore, there are still concerns with the ability for the system to be able to differentiate between wildfires and other types of smoke. This issue is one that will continue to define the product’s effectiveness, as a misdiagnosis of the issue can cost time and money for responding forces.
AI Dash Where Pano AI can have issues with overreach and scalability, other solutions can be presented in the form of satellite imagery and predictive services. Utilizing AI to compile and interpret imagery, AI Dash utilizes satellite imagery to determine response with exponentially wider coverage. The benefits of this model are clear: AI Dash’s Climate Risk Intelligence System (CRIS) creates a framework for not only wildfire and power outage detection, but also for cleanup and recovery efforts. CRIS claims to increase the speed and lower the cost of response by 30% through their use of historical data patterns for disaster spread, and their tracking methods to predict power outages. There are two key design details which AI Dash claims its CRIS model performs better at than other wildfire tracking services, one of which relies on the usage of satellites. 80% percent of power outages are caused by vegetation damage, and the ability to detect potentially harmful vegetative mass is impeded by the lack of visibility at ground level. A satellite can obviously circumvent this issue entirely, while also adding faster detection and AI-based interpretation on top of that. However, the main issues with this system are based on the efficacy of satellite detection. For example, while thickets of vegetation can provide visibility issues in level vegetation, they can also obscure a top down view— small burns are hard to detect under thickets. However, cameras can then fill in this role to create more detailed and responsive points of view to detect fires even at night-time.
SmokeD System is a hardware and software startup focused on detecting wildfires utilizing AI and cameras. Much like Pano, the camera system employed by SmokeD utilizes a surveillance range to then use its AI model to detect and interpret wildfires. According to their product lines, SmokeD has specially created cameras that utilize 24/7 surveillance in a 10 mile radius. The system claims to detect fires within ten minutes, a span of time which the company claims can prevent most wildfires from occurring. This fast detection time utilizes AI interpretation to determine and optimize the level of response. This response time can be optimized as well depending on the terrain of the local area–SmokeD claims that this helps to determine response type and scale. Due to this model being AI powered, the company claims that this response time and optimization will continue to improve as the AI model is trained more and more on the field. SmokeD also offers two separate workstations, one for average consumers and the other for major companies. Both track camera views and provide insights into these networks: SmokeD Web is browser based, and covers the widest range of images and data from every detector, while SmokeD Alerts is a mobile application that provides notifications for wildfire risk in their network.
Strengthening Carbon Neutrality Over the last 50 years, governments and firms have slowly begun working and incentivizing industries to become more climate sustainable. One of the major ideas that have come out from governments to incentivize this push comes in the concept of carbon neutrality— a concept that involves balancing the carbon a country or firm releases into the atmosphere is entirely balanced out by carbon absorbing measures. Carbon neutrality is often paired with other environmental standards like the ESG framework. Due to ESG being a favored investment asset, the need for carbon neutrality becomes more popular among firms (Witzel et al, 3). There are two ways companies can achieve carbon neutrality. One major way is to integrate sustainable procedures in each and every aspect of the workplace— recycling materials, effective waste management, lowering power usage, and managing food waste. Furthermore, companies can also ensure carbon neutrality in their supply chain by financing groups that assure a carbon neutral sourcing. This can include creating livable workplace conditions, sourcing renewable energy sources, and providing guarantees on safety and compensation. The second method to achieve carbon neutrality comes in the form of carbon offsetting, where a firm can neutralize their carbon consumption by helping fund restorative ‘carbon’ projects or purchasing carbon credits (Witzel et al, 4). The Carbon Markets The carbon market was fully actualized with the 1997 Kyoto Protocol, when it was taken from a regional US market and taken to the global stage. Instead of using fiat currency however, the carbon credit market would run on carbon credits— representing one certified ton of CO2 (The Ultimate Guide to Understanding Carbon Credits). This meant that any restorative project that could take carbon away from the atmosphere could earn carbon credits, which could be sold for currency or bought to be able to emit CO2 without tax burdens . There are many ways of earning carbon credits, and carbon projects aim to create large swathes of credits that can be traded on two types of markets: compliance and voluntary markets. Compliance markets involve credit markets that involve trades meant to follow regulatory standards, while voluntary markets are often trades meant for public health, goodwill, and an investment in the future. These two markets are very important due to how they affect the climate change industry: electric car markets, renewable energy, agriculture, biofuels, reforestation efforts and much more run on the incentives provided by carbon credits (The Ultimate Guide). With the future bringing a greater need for climate response solutions, carbon credits may become more important to incentivise growth.
The Flaw with the Carbon Market However, there are still significant issues with carbon markets and their implementation in the global sphere. A 2024 article by the Center for Strategic and International studies hammers down a significant flaw within the methodology of carbon offsetting. While carbon restoration projects yield credits based on the restorative project, there is no realistic way to determine whether or not a carbon project will actually yield its worth of carbon credits. The article cites a 2023 study done investigating the credit certification company Verra, which found that 90% of the carbon credits gained through reforestation were found to have very little effect in reducing emissions. This lack of efficacy can create distrust in the voluntary market, which solely operates on the goodwill of its participants. In addition to the lack of efficacy, there is also a significant lack of direction given to credit buyers in effectively using carbon credits to manage their business. In a market report by non-profit Ecosystem Marketplace, directors Stephen Donofrio and Alex Procton stressed the importance of “a standard approach for voluntary carbon buyers’ ambition and action criteria,” which they found was essential to the rate of voluntary investment in carbon credits. However, the advent of AI in recent times has allowed some shake-ups in the carbon infrastructure industry. Advanced data interpretation and decision making has made massive developments for companies to create more sophisticated and future proof carbon projects–ones that could reinsure confidence in the industry and introduce new players to the carbon market.
Pachama One such company focused on this development is Pachama, a San Francisco carbon development company focused on funding and managing carbon projects using AI. Aiming to create a framework where carbon projects are managed in order to yield maximum credits, Pachama claims to have improved the quality and quantity of the carbon credits created using their tools. Having over 150 forest carbon projects certified by carbon registries, the company claims that their processing plan was key to their success. In their approach to ensure the quality of their projects, Pachama holds a 4 step code of quality control: ensuring net additional carbon credit yield, determining quality yields, planning longevity, and protecting local ecosystems. For each of these qualifiers, Pachama provides a detailed process on how to achieve it. For one, carbon project emissions are treated conservatively to maintain expectations and increase the legitimacy of each carbon credit. Alongside this, carbon projects that are presented have to be more conservative with their yield predictions than the models Pachama predicts with their remote sensing models—Pachama achieves this by setting up what they label as a baseline model. This baseline model assumes the carbon emission reductions that would have occurred without the existence of a carbon project, which then makes it easy to determine the worth of a proposed project. One essential and overlooked aspect of a carbon project is the overall benefit a potential project could provide to the land it is attempting to restore. There are a multitude of social and ecological factors that play into the efficacy of a carbon project, one that go beyond credit yield. These factors are important because of how they can affect the social response to a specific carbon project— being unable to follow these can lead to funding and responsibility issues. As a result, Pachama will also conduct project research only if they are certified around local communities and considering trackers of biodiversity and ecological impact. Changes such as planting a wide variety of native flora and keeping uniform with their density and symbiosis in natural areas can help the newly planted areas fit better with their environment. Making sure to not impede on local communities and their expansion also helps retain trust and prevents ethical concerns. Many of these concerns can be potentially solved by using data, tracking, and legal management, all of which Pachama’s AI and Satellite model can help verify before greenlighting a project.
Treeconomy Like Pachama, Treeconomy is a startup that relies on AI models to create sophisticated predictions for carbon projects. However, the company’s Sherwood developing hub aims for a more specific purpose: streamlining and tracking carbon projects using satellite and AI models. According to Treeconomy, their company model can reportedly give results within three business days, a feature that can potentially provide an edge to other development models. The Sherwood model works in three major phases: pre-development, credit verification, and project check in. For each phase, Treeconomy breaks down each of the processes into much smaller and practical terms. The pre-development stage consists of the proposal being submitted through their hub, after which the sender will receive metrics on the eligibility of the site, satellite map data, and the various ways in which those items were delivered and sourced from. After this, assessments provided by the hub will help to prepare the potentially gained credits for certification. Finally, the development hub will also help the project gain early stage finance. These assessments, much like Pachama, also measure both the feasibility and undertaking of a project by comparing its own predictions with the ones of the project proposal. Unlike Pachama’s singular deep assessment model however, Treeconomy aims to give two separate evaluations: rapid and comprehensive carbon assessments. Rapid carbon assessments provide fast information on the amount of biomass and carbon stock on the project site, while comprehensive carbon assessments create detailed maps and extensive metrics that the company claims can give information as specific as tree canopy metrics. There are also a multitude of other assessments, such as longevity, health monitoring, hazard detection, and certification of carbon credits. The final piece of Treeconomy’s product is to provide an open funding platform for carbon projects to receive funding and to also perform the sale of carbon credits. This project also contains qualifiers for capable projects—carbon benefit, ecological advancement, and community support are essential requirements for projects to receive funding. Speculation With the interest in the carbon market steadily growing, companies are providing new solutions to ever growing problems. Funding for carbon projects can be hard to accrue, and services like Pachama and Treeconomy can prove to be potential routes for new players to not only gain confidence in carbon credits but also learn how to earn them. With new solutions also come new challenges to overcome. For one, while companies like Pachama and Treeconomy can provide ample funding and future-proofing for their models, there are simply too many factors that can negatively affect the yield of carbon credits in the future . Carbon credits are only recorded based on the potential value of their yield, and while proper planning can help achieve that yield, it can only be calculated after a significant growth period (Dawes, CSIS). Reforestation projects can take decades to deliver results, and have to constantly have their value reassessed to predict the amount of credits gained. Furthermore, even with the important qualifications required before approving a carbon project, certification boards can often misrepresent results. In fact, Pachama utilizes the certification board Verra for their board certification, the same organization that was reported in the 2023 study to have overreported 94% of their carbon credit values, rendering them effectively worthless. According to the Guardian, only 5.5 million of the 94.9 million reported carbon credits by Verra actually reported real emissions claims— a devastatingly low number that calls into question . With Verra being the largest certifying board for carbon credits, this significantly de-legitimizes the voluntary carbon market. With this also comes uncertainty about how new AI-based development platforms can account for and perform important data-based tasks. Carbon projects can have many factors to manage, and while AI claims to speed up the interpretative times and provide results, the results of these projects may only come to fruition in the later future. However, as the industry begins to become more important in this new age of global warming, there is still potential for growth and corrections to past mistakes.
Conclusion Artificial intelligence and its relatively recent uptick has allowed new startups to take important foundational software and begin applying it to more niche uses, such as climate software. Because of how large and wide-reaching an issue it is, there have been many different solutions that aim to solve a variety of issues—these include large scale economics problems, climate modeling, and also hazard detection concerned with fast scale events like wildfires. New companies such as Pano and SmokeD use AI-integrated camera systems to help set up close-up fire detection, while Pachama and Treeconomy utilize AI to help extract the highest amount of carbon credits from projects. These startups provide an interesting glimpse into the world of AI and its uses in climate change. Many startups focus on using AI as a tool for cleaning out inefficiencies with interpreting data, with different ranges of usage: Wherobots uses it as a tool to interpret connections within compiled mapping data, while Pachama scales AI interpretation to gather information of various ecological and topographical factors. AI is also used as a 24/7 detection service that can point and locate specific details in record times— a tool used by Pano AI and SmokeD Systems to market their product. Startups like the aforementioned ones are marketed based on their AI features as well, as it helps define themselves from previous products. AI is a key developmental advantage that these companies claim helps benefit the world of climate software, and this is a claim supported by research organizations like CCAI.
Regardless, AI has proven to be a major advance for climate software development. Many concerns with the climate are defined by the lack of speed, accuracy, and scale of responses to a variety of problems, and many of these responses are based on the inefficiencies of human computation. The diversity of usage of artificial intelligence in this sphere supports this concept, and much like the investors that helped fund the infrastructure for today, it may be worthwhile to continue investing in more sophisticated technology for tomorrow.
This article was co-authored with Aditya Shukla, a venture fellow in the summer of 2024.
Companies mentioned: Fused, Wherobots, PanoAI, AIDash, SmokeD, Pachama, Treeconomy
Works Cited
“2024 State of the Voluntary Carbon Market (SOVCM).” Ecosystem Marketplace, 18 Aug. 2024, www.ecosystemmarketplace.com/publications/2024-state-of-the-voluntary-carbon-markets-sovcm/. Abhishyant Kidangoor, Sandy Watt, et al.
“New Tree Tech: AI, Drones, Satellites and Sensors Give Reforestation a Boost.” Mongabay Environmental News, 31 Aug. 2023, news.mongabay.com/2023/07/new-tree-tech-ai-drones-satellites-and-sensors-give-reforestation-a-boost/.
“Advanced Wildfire + Bushfire Detection Technology: Pano Ai.” Advanced Wildfire + Bushfire Detection Technology | Pano AI, www.pano.ai/. Accessed 30 Aug. 2024.
AiDash. “Why Utility Executives Rely on Satellite-Powered Vegetation Management.” AiDash, 5 Jan. 2024, www.aidash.com/utility-executives-satellite-powered-vegetation-management/.
Boukabara, Sid-Ahmed, et al. “Outlook for Exploiting Artificial Intelligence in the Earth and Environmental Sciences.” AMETSOC, American Meteorological Society, 26 May 2021, journals.ametsoc.org/view/journals/bams/102/5/BAMS-D-20-0031.1.xml?rskey=gOE0RA&result=9&tab_body=pdf.
“Build with Data, Instantly.” Fused, www.fused.io/. Accessed 30 Aug. 2024.
“California Air Resources Board.” Cap-and-Trade Program | California Air Resources Board, ww2.arb.ca.gov/our-work/programs/cap-and-trade-program. Accessed 30 Aug. 2024.
“California: Let’s Stop Making Wildfire History.” The Nature Conservancy, 15 Sept. 2018, www.nature.org/en-us/about-us/where-we-work/united-states/california/stories-in-california/californias-wildfire-future/. Carbon Credits.
“The Ultimate Guide to Understanding Carbon Credits.” Carbon Credits, 2 July 2024, carboncredits.com/the-ultimate-guide-to-understanding-carbon-credits/. Davidson, Nikki.
“AI: A Wildland Firefighter’s New Best Friend?” GovTech, GovTech, 7 May 2024, www.govtech.com/artificial-intelligence/ai-a-wildland-firefighters-new-best-friend. Dawes, Allegra.
“What’s Plaguing Voluntary Carbon Markets?” CSIS, www.csis.org/analysis/whats-plaguing-voluntary-carbon-markets. Accessed 30 Aug. 2024.
Franke, Ulrike, and Paola Sartori. European Council on Foreign Relations, 2019, Europe and the Ai Revolution, http://www.jstor.org/stable/resrep21907. Accessed 31 Aug. 2024.
Greenfield, Patrick. “Revealed: More than 90% of Rainforest Carbon Offsets by Biggest Certifier Are Worthless, Analysis Shows.”
The Guardian, Guardian News and Media, 18 Jan. 2023, www.theguardian.com/environment/2023/jan/18/revealed-forest-carbon-offsets-biggest-provider-worthless-verra-aoe.
Krishnan, Aarti, and Simon Maxwell. Odi, 2020, Counting Carbon in Global Trade: Why Imported Emissions Challenge the Climate Regime and What Might Be Done about It, http://www.jstor.org/stable/resrep49989. Accessed 31 Aug. 2024.
“Let’s Create a Greener, Cleaner, Safer Planet from Space.” AiDash, 30 July 2024, www.aidash.com/.
Oxtoby, David W. “Biologists and Carbon Neutrality.” BioScience, vol. 58, no. 5, 2008, pp. 382–383, https://doi.org/10.1641/b580502.
Rolnick, David, et al. “Tackling Climate Change with Machine Learning.” ACM Computing Surveys, 7 Feb. 2022, dl.acm.org/doi/10.1145/3485128.
“Smoked Wildfire Software and Applications.” SmokeD, 2 Aug. 2024, smokedsystem.com/software/. “Unlock Nature’s Value.”
Treeconomy, www.treeconomy.co/. Accessed 30 Aug. 2024.
Walker, Tenika Versey. “Climate Tech Startups Integrate Nvidia AI for Sustainability Applications.” NVIDIA Blog, 22 Apr. 2024, blogs.nvidia.com/blog/earth-day-2024-climate-tech-ai-startups/.
“Wherobots.” Wherobots Documentation, 17 May 2024, docs.wherobots.com/latest/. Witzel, Mardi, and Niraj Bhargava. Centre for International Governance Innovation, 2023, pp. 3–4,
What Is ESG?, http://www.jstor.org/stable/resrep52982.7. Accessed 31 Aug. 2024.
