Use artificial intelligence to combat climate change

Recent reports from the Intergovernmental Panel on Climate Change (IPCC) show that climate change is affecting every region of our planet and that some of the changes — such as rising temperatures and sea levels — can only be arrested, but are irreversible. Part of the problem stems from how complex an issue climate change is. It has scientific and economic elements and sociopolitical and ethical ones, and it requires cooperation on a scale hitherto unseen.

Halting climate change and responding to the effects of the damage already wrought requires two approaches. The first is mitigation, namely trying to remove carbon dioxide from the atmosphere while reducing emissions. The second is adapting to the existing, irreversible changes, and to those yet to come because of the trajectory we are on. 

AI will be critical to finding solutions

This is why any and all strategies to combat climate change or mitigate its effects are worth exploring, including artificial intelligence (AI), which will probably be critical in Africa’s climate action strategies.

AI enables computers to parse huge swathes of data, learn from disparate inputs, recognise patterns, and adapt decision-making and responses over time. 

AI solutions, like convolutional neural networks, are especially adept at tackling enormous datasets and turning them into predictive models, something meteorologists can use to analyse extreme weather events. Nvidia’s FourCastNet, for instance, models global weather patterns to predict outlier events, and can do with hitherto unseen accuracy and rapidity — it can create a seven-day forecast in a fraction of a second, which is five orders of magnitude (100 000 times) faster than numerical weather predictions.

Stanford researchers, meanwhile, have developed a machine-learning model that makes use of atmospheric patterns to predict the sort of outsized precipitation that causes flooding. The model can be used both for short-term prognostication (when flooding might happen), and for longer-term planning, like building flood-resistant infrastructure.

In much the same way AI has proven useful in diagnostic medicine, its ability to produce more accurate and effective weather forecasting ever is proving handy for myriad sectors. It can, for example, help farmers better time the sowing or harvesting of crops to sidestep unusual weather events. Similarly, it can help with forestry efforts by predicting the outbreaks of wildfires, or the proliferation of harmful insects. One such effort is IBM Deep Thunder, which can help predict the scale and path of tropical storms, or excessive precipitation and the effect it might have on infrastructure.

Many of these AI prediction services use a machine-learning technique called “random forests”, which relies on multiplier decision trees that are used in combination with enormous data sets to make predictions about the probability of various outcomes, and which can adapt predictions based on new data as it becomes available. As new data is added and the algorithms are refined, it’s possible these services could one day replace current prediction models completely.

AI can also aid in cutting carbon emissions by helping companies monitor their current emissions, predict their future ones, and find ways to reduce them by identifying areas where reductions can be made, say in parts of the production process, transportation, travel, or other business functions. For example, AI can analyse traffic conditions to plot more efficient routes, weather conditions to adjust climate control in buildings, or better manage product inventory to cut the need to move items between different locations.

Aeromon, a Finnish AI startup, tracks industrial emissions in real-time and can quantify those emissions while visualising their constituent parts. A company called Carbon Tracker, meanwhile, received a $1.7-million grant from Google to use its satellite-based technology to study the emissions of nearly 4 000 power plants, making it possible to hold them to account for their respective environmental impacts.

A steelmaker looking to reduce emissions turned to a company called BCG, which used AI to help it optimise its production processes. AI-powered process controls eliminated waste, reduced energy usage, and cut costs thanks to the use of thousands of sensors, the data from which was fed into the control system’s algorithms. The resultant 3% decrease in carbon emissions equates to 230 000 tonnes a year, and cost reductions of $40-million.

It’s not just manufacturing that’s a problem for climate change, though, online retail presents its own problems. Clothing retailer Moosejaw, managed to reduce “size sampling” (where customers buy multiple sizes of the same item and return those that don’t fit) by using AI to create a solution called True Fit that makes it more likely an item will actually fit a customer. Moosejaw used AI to calculate that almost 15% of returns were attributed to size sampling. 

True Fit lets users create a profile based on the size of other items they own, and cut size sampling by Moosejaw customers by 25% in a single year. Tech firm Optoro conducted research which found that consumers returning products in the United States alone could be responsible for 23 million tonnes of carbon emissions by 2025, so solutions like Moosejaws have the potential to play a significant role in reducing that problem, something customer education alone likely won’t solve.

AI for disaster risk management

Between 2005 and 2015, more than 1.5 billion people were affected by natural disasters. And in 2017 alone, natural disasters like earthquakes, floods, hurricanes and wildfires accounted for $306-billion in damages worldwide, which is almost double the figure ($188-billion) for 2016. Disaster risk reduction efforts can benefit from AI, which can assist with event detection and prediction, and also with real-time response systems and modelling outcomes as more data becomes available and a disaster unfolds.

When it comes to predictions and mapping of potential hazards, the quality of the source data is paramount. Fortunately, it continues to improve — satellite imagery continues to get better, sensors get smarter and more prolific, and the computers required to run modelling continue to become more potent. Services such as SkyAlert in Mexico, for instance, rely on smartphone apps combined with a network of sensors powered by Microsoft Azure to warn millions of people about earthquakes minutes before they happen.

Similarly, digital twin technology — which, with the help of AI creates virtual representations of physical spaces and is able to factor myriad variables into its models — can be used for public safety and emergency response initiatives, both preemptive and reactive.

Researchers at the Vrije University Institute for Environmental Studies in Amsterdam have found success using Twitter to help detect floods. The combined data from disaster response organisations, the Global Flood Detection System satellite flood signal, and location-specific Twitter mentions of flooding to understand the location, timing, severity, and effects of floods.

Streamlining energy production and use

AI can also encourage more ecologically sound energy production and use by, for instance, creating predictive models to account for shifts in demand on the grid and adjusting how much power from renewable resources is stored for later use, or allowing consumers using eco-friendly generation solutions to contribute surplus energy to the grid in times of need. 

California-based AutoGrid uses AI to help its customers adjust their consumption to match their usage patterns. Google’s DeepMind, meanwhile, is able to predict wind turbine power generation based on historical data, which is especially valuable for a power source that’s notoriously inconsistent. In a similar vein, IBM developed a solution for the US department of energy’s SunShot Initiative that uses historical weather data, real-time information from local weather stations, networks of sensors and cameras, and even satellite imagery, to improve the accuracy of “solar forecasting” — predicting how sunny it’s going to be and how much power can thus be generated from solar panels rather than taken from the power grid.

The energy sector is responsible for 73.2% of global greenhouse gas emissions, so improving its efficiency could have significant benefits. More accurate forecasting is just one way people are trying to do precisely that. Xcel Energy, a Colorado-based business, uses AI-based data mining methods gleaned from the National Center for Atmospheric Research to create highly accurate weather reports.

General Electric uses AI to predict when its wind turbines or hydro-power generators are likely to need maintenance, reducing downtime and improving reliability. PowerScout, in Oakland, California, uses AI in conjunction with industry data to predict the likelihood of a household choosing to invest in solar generation systems, which allows sales teams to concentrate their efforts on the strongest leads.

Informing and implementing strategies for a sustainable future

Vehicle emissions are a major contributor to climate change, and although there is an increased move towards electric vehicles in the consumer space, in freight that remains some way off. And with ever more retailers relying on delivery to fulfil orders, we can’t afford to wait for a shift to more eco-friendly solutions. But AI can help with logistics and drive efficiencies, reducing emissions in the process.

From traffic monitoring using cameras, sensors, or historic information to route optimisation and timing, AI is being used in fleet management solutions like Fleetilla and Vitreo to reduce emissions while still enabling efficient and responsive supply chains. Key to these solutions is the ability to predict environmental factors or adjust to them in real-time, and it’s also helpful to be able to optimise vehicle maintenance, or the maintenance of any machine in the value chain, for that matter.

A study, titled How AI Can Enable a Sustainable Future, suggests that AI aimed at environmental protection could contribute up to $5.2-trillion to the global economy by 2030. An example of one such solution is the field of “predictive maintenance”, which uses permanent, automated monitoring of machines to reduce downtime, reduce the excessive energy use that can happen when machines become inefficient, and make it less likely machinery will result in higher emissions due to malfunctions or inefficiency.

The role of AI in climate infomatics and forecasting

Climate informatics is a relatively new research discipline that seeks to facilitate collaboration between climate scientists and data scientists, both of whom have seen huge progress in their respective fields in recent decades, but who don’t always turn those from data points into actionable insights. Given the pressing demands for action that climate change brings, bridging the chasm between them is imperative.

The idea of climate informatics was introduced in 2012 by Claire Monteleoni, an associate professor at George Washington University. She continues to use machine learning to try and illuminate the climate change crisis and to promote collaboration between machine learning and AI researchers and climate scientists, who have increasingly powerful physics-driven climate models at their disposal which can be used to generate simulations that provide information not only about the past but also about the likely future.

Because climate and weather are so hard to predict, anything that can make them more so is valuable, especially at a time when edge-case climate calamities are increasingly common. Machine-learning or AI-powered solutions also tend to consistently outperform traditional ones, making the investment in improving them even easier to justify. For instance, deep learning is being used to predict urban air pollution using only satellite data, and convolutional neural networks are being used to distinguish the various micro-climates of some urban areas using ground-level images.

Many researchers have used deep-learning and random forest models to assess climate data and assess the impact of extreme events or evaluate the creditworthiness of borrowers exposed to physical climate risks. M Enslin (2022) found that only a small number of events had a statistically significant impact on the default rates, default predictors, and property valuations, which probably would not have been the assumption had AI not been used to analyse the data.

Accelerating climate change adaptation using AI

A 2021 report by the international NGO CDP found that climate change is set to cost businesses $1.3-trillion by 2026. AI can help those businesses anticipate and prepare for the worst using the best information available to them. It can help them monitor and assess their exposure to climate risk, and adjust their responses accordingly as circumstances change — whether internal or external.

Companies like Climate AI use a mix of readily available information and proprietary machine learning to help companies gauge the risk climate change poses for them and their unique circumstances, and make suggestions on how to mitigate or prepare for it. Another start-up, One Concern, is developing a digital twin of the whole world to try and deliver hyperlocal information to customers, which it’s positioning as “resilience as a service.”

Unsurprisingly, AI is also being harnessed to assess insurance risk. A start-up called Kettle is using AI to make predictions about wildfires in California and using that data to offer insurance companies more competitively priced reinsurance. At the same time, IBM’s Environmental Intelligence Suite allows businesses to measure the impact of their activities on the environment and aids them in beginning to mitigate their own carbon footprints.

AI can also enable environmental, social, and governance assessments and disclosures, which are essential in a growing number of markets where measuring and reporting carbon emissions is no longer optional for companies. Often as important as measurement and reporting is how that data is governed, and here AI can help again.

Because measurement and demonstrable reductions are growing in importance, there’s also a growing risk that companies will try to misrepresent or “greenwash” their metrics. Here, too, AI is extremely helpful. Imperial College’s business school is using a combination of AI and statistical tools to proactively detect greenwashing or false reporting while also trying to offer insights into the effectiveness of, or best practices for, net-zero efforts.

It’s not yet clear precisely all of the roles AI will play in combating, mitigating or reacting to climate change, but it’s increasingly clear it’s going to be both substantial and crucial.

Dr Mark Nasila is chief analytics officer for First National Bank’s chief risk office

The views expressed are those of the author and do not necessarily reflect the official policy or position of the Mail & Guardian.