Climate change is a serious issue facing the world. The climate changes which are already affecting our planet can be seen in rising sea levels, melting ice caps and glaciers, more severe storms and hurricanes, more droughts, and wildfires increased precipitation in some areas of the world while other regions experience less rainfall. It’s important that we do what we can to reduce climate change risks by reducing greenhouse gas emissions into the atmosphere as well as adapting to climate impacts. Artificial intelligence (AI), machine learning (ML)/ deep learning (DL), advanced analytics have been widely used for decades across different industries such as finance, healthcare, etc., but their use cases related to climate change have not been explored much yet. In this blog post, we will list down different use cases in which AI & ML can be used to tackle climate change related problems.
Before getting ahead and reading the blog, let’s look at a video in relation to climate change and its impact:
What are some of the key AI / ML use cases related to climate change?
Here are key AI / ML / deep learning use cases of climate change:
Predicting extreme precipitation
Extreme precipitation is defined as rainfall that is greater than the 99th percentile of historical climate data. Extreme precipitation forecasting can be done with climate models and machine learning techniques. The climate models are getting better at predicting weather and climate patterns. However, they’re still not great at accurately forecasting extreme precipitation. The climate model takes in meteorological variables such as temperature, humidity, surface pressure to make a prediction about future extreme precipitation events. Meanwhile, the machine learning technique is fed historical data on daily weather patterns along with climate change projections of greenhouse gas emissions to predict how often an area will experience heavy rainfall or snowfall.
When reasoning from first principles, we can arrive at different causes (problems/opportunities) and KPIs related to extreme precipitation and track them in order to arrive at appropriate AI/ML solutions. The following are some details:
- As per Dr. David Novak of NOAA’s Weather Prediction Center, many events of extreme rainfall share characteristics, such as high moisture and an atmospheric disturbance (i.e., a winter storm, warm/cold front or tropical cyclone). The longer these conditions persist in the same place, the more extreme the rainfall you’re likely to have. And, it is known that warmer air can hold more moisture, so it stands to reason that a warmer environment overall is capable of generating more extreme rainfall. We will thus want to design and perform hypothesis testing around air moisture and warmer air to ascertain the relationship between extreme precipition, air moisture and warmer air. Accordingly, use the related features in creating predictive models for predicting extreme precipitation.
- As mentioned in the previous point, it is the warmer air that can hold air mosture which can possibly result in heavy rainfall after a point. So, what results in warmer air? One of the reason is presence of greenshouse gases in the atmosphere which absorb the infrared radiation from the surface, the atmosphere heats up. The goal is to determine what results in greenhouse gases, design and perform hypothesis testing around the reasons for greenhouse gases and appropriately use the information to build the predictive or AI/ML models.
Modeling carbon sequestration
It is a method to predict how much carbon will be removed from the atmosphere and stored. The earth’s climate has swung back and forth between glacial periods (ice ages) and interglacial periods (warm times). This is due to fluctuations in the earth’s orbit around the sun. The climate models can’t predict what the climate will be like in 100 or 1000 years because it’s very sensitive to small changes in temperature. Researchers are using machine learning algorithms to model carbon sequestration and its impact on climate change over time.
When reasoning from first principles, some of the following would need to be understood in order to build predictive models around carbon sequestration:
- What can be the indictors for the success of carbon sequestration? How to model the success of carbon sequestration methods?
- Carbon in what all forms can be processed using carbon sequestration method? How to prioritize the carbon forms?
- What can be different types of carbon sequestration methods and what will be KPIs for each one of them?
Estimate / Predict carbon dioxide emissions
The greenhouse gases CO2, CH4, and N2O are emitted during the combustion of fuels. CO2 accounts for the majority of the GHG emissions from stationary combustion sources. Fossil fuel combustion is a major source of CO2 emissions. Researchers are using climate models and machine learning algorithms to estimate how much CO2 can be emitted while still remaining within a given climate goal (i.e., staying under two degrees Celsius). This forms the key aspect of carbon accounting where the goal is to determine some of the following.
- Globally, what’s the carbon budget at any given point in time with the goal to avoid the carbon emissions beyond a limit resulting in rise of global temperature by 1.5 deg or so?
- What’s the carbon emissions by country, region, etc?
AI/ML models can be used for predicting different aspects of carbon accounting such as those mentioned above. The GHG emissions at any point of time can be determined using different methods such as the following:
- Direct measurement: Direct measurement of CO2 emissions is performed through the use of a Continuous Emissions Monitoring System (CEMS)
- Fuel input analysis: Fuel analysis is essentially a mass balance approach in which carbon content factors are applied to fuel input to determine emissions.
In fuel input analysis method, Based on the information available on the characteristics of the fuel being consumed, appropriate equations such as the following can be used to calculate emissions. The details can be read in this document. Emissions is calculated as the mass of CO2, CH4,or N2O emitted.
- Emissions = Mass or volume of fuel combusted (CO2, CH4 or N2O) X EF (Emssions factor per unit mass or volume)
- Emissions = Mass or volume of fuel combusted (CO2, CH4 or N2O) X EF (Emssions factor per unit mass or volume) X fuel heat content (Higher heating value – Units of energy per unit mass or volume)
- Emissions = Mass or volume of fuel combusted (CO2, CH4 or N2O) X Fuel carbon content (in units of mass of carbon per mass or volume of fuel) X 44 / 12
In all of the above formula, it can be noted that the following needs to be estimated or predicted and this is where statistical and machine learning methods can be used:
- Mass or volume of fuel combusted
- Emissions factor
- Fuel heat content
- Fuel carbon content
Predicting greenhouse gas concentrations
Greenhouse gases trap heat from escaping into space and are responsible for keeping the earth’s climate warm enough for life as we know it. Researchers use climate models to predict future concentrations of greenhouse gases. Machine learning techniques are used in climate model predictions so that they can be run faster and more efficiently.
The following picture represents the three key concentrations of green house gas (GHG) emissions including carbon dioxide (CO2), methane (CH4) and Nitrous oxide (N2O) as per this DowntoEarth.org page. Other gases primarily include hydrofluorocarbon, perflurocarbon and Sulfur hexafloride gases. Greenhouse gases all have different global warming potentials (GWP), which is a measurement of how much heat they can trap in the atmosphere.
The concentrations shown in the diagram needs to be estimated or predicted from time-to-time and this is where statistical and machine learning methods can be very helpful.
Forests play an important role in many forest degradation and climate change solutions approaches. They help regulate the Earth’s temperature, produce oxygen, and provide habitats for plants and animals. However, forests are being degraded at an alarming rate.
Forest degradation is a major problem that contributes to climate change. It occurs when forests are damaged or destroyed, typically as a result of human activity. Deforestation, forest fires, pests, and poor forest management are all major causes of forest degradation. The loss of trees and other vegetation can have a significant impact on the local climate. In addition to absorbing carbon dioxide, trees and plants help to regulate temperature and precipitation levels. When forests are degraded, these important functions are compromised, contributing to climate change. As a result, the amount of carbon dioxide being removed from the atmosphere is decreasing, and the Earth is becoming warmer. In addition, forest degradation can also cause soil erosion, which can lead to desertification and loss of biodiversity. Therefore, it is essential to address the problem of forest degradation in order to protect our environment.
The solution to forest degradation that has a great impact on climate change must be two-fold: stopping the destruction of forests and planting new trees. To stop forest degradation, we must first reduce our reliance on wood products. This can be done by investing in Forest management practices, using Forest certified products, and recycled paper products. We must also protect forests by creating protected areas, promoting forest conservation, and reducing emissions from deforestation. In addition, we must work to replant trees and restore forest ecosystems that have been degraded. Reforestation efforts will help to mitigate climate change, provide habitats for wildlife, and create jobs in forest-dependent communities. To summarize, here are some of the hypotheses which can help reduce forest degradation.
- Early detection of forest destruction
- Identification of lands where trees can be replanted
- Reduced dependence on wood products
Given the above solution approaches, it is important for environmental scientists to be able to detect forest degradation as quickly and easily as possible so that climate change impacts can be reduced by implementing some of the above mentioned solution approaches. Researchers are using machine learning algorithms and build models using the satellite data such as high-resolution images from the Landsat series of satellites in order to automate this process.
Identifying climate-vulnerable regions
Climate vulnerable regions are those which are especially prone to the effects of climate change. Such regions are typically low-lying coastal areas, small island states, and developing countries in the tropics. Climate change impacts these regions in a number of ways, including sea level rise, increased frequency and intensity of extreme weather events, changes in precipitation patterns, more extreme weather events, and reduced access to water. The identification of climate vulnerable regions is an important first step in tackling the problem of climate change. By identifying regions which are most at risk, it is possible to better target resources and support. However, climate change is a global problem, and its effects will be felt by all regions in one way or another. These impacts will exacerbate existing problems such as poverty, water shortages, and food insecurity.
Identifying climate vulnerable regions is no easy task, as climate change will affect countries in different ways and to varying degrees. However, by studying historical climate data and using climate models, we can begin to get a sense of which regions are likely to be most vulnerable. Once these regions have been identified, we can start to develop strategies for mitigating the impact of climate change on them. In doing so, we can help ensure that future generations inherit a world that is still habitable and prosperous. Climate scientists and researchers are using machine learning techniques on climate model outputs in order to identify climate-vulnerable regions that may be subject to extreme heat, drought or flooding due to climate change.
Predicting wildfire risk globally
The wild fire risk is a problem that has been identified with climate change. A wild fire is an uncontrolled fire that burns in an area of land that is not prepared or managed for fire. They typically occur in late summer or early fall, when the weather is hot and dry. In recent years, there has been an increase in wild fire activity across the globe, and scientists believe that climate change is a contributing factor. Wild fires can have devastating consequences, destroying homes and natural habitats. They also release large amounts of carbon dioxide into the atmosphere, further worsening climate change.
Climate change is making wild fires more likely and more severe, and we need to take action to prevent them. One way to do this is to reduce the amount of greenhouse gases we emit into the atmosphere. This can be done by using less fossil fuels, planting trees, and protecting natural areas. We also need to be prepared for wild fires by creating defensible spaces around our homes and communities, and by having an emergency plan in place. By taking these steps, we can minimize the risk of wild fires and protect our families and property from harm.
As wild fire season lengthens and becomes more destructive, the need to predict wild fire risk has never been greater. Machine learning is a powerful tool that can be used to identify patterns and make predictions. By analyzing data from past wild fires, machine learning can help to identify areas at risk and the conditions that lead to wild fires. This information can then be used to create predictive models that can help decision-makers take action to prevent wild fires.
Predicting climate change impacts on crop yields
Crop yields forecasting is the process of making predictions about future crop production. It is the scientific study of past crop yields and climatic patterns in order to make predictions about future crop yields. The practice can help farmers to plan for droughts, heatwaves, and other extreme weather events that are likely to become more common as the world climate continues to change. Crop yields forecasting is a complex task that takes into account a variety of factors, including solar radiation, temperature, precipitation, and soil moisture levels. However, the accuracy of forecasts can vary depending on the stability of the climate in a particular region. For example, crop yields in the Sahel region of Africa are highly variable from year to year due to the unpredictable nature of the rains. As a result, forecasting crop yields in this region is particularly challenging. Nevertheless, crop yields forecasting is a valuable tool that can help farmers to adapt to a changing climate.
Predicting sea ice loss due to climate warming
There is a strong scientific consensus that human activities such as burning fossil fuels and deforestation are leading to climate change. One of the most visible effects of climate change is the loss of sea ice. Sea ice is formed when seawater freezes, and it plays an important role in regulating the Earth’s climate. Sea ice reflects sunlight, helping to keep the planet cool. When sea ice melts, it reveals dark ocean water, which absorbs more heat and causes the Earth to warm even further. Climate models predict that sea ice will continue to decline in the coming years, amplifying the effects of climate change and resulting in even more extreme weather patterns. Therefore, it is essential that we take action to reduce greenhouse gas emissions and slow the pace of climate change.
It is important for us to be able to predict how climate change will affect sea ice loss in both the Arctic and Antarctic. Researchers have used machine learning algorithms to model climate warming over time while also estimating what that means for sea ice levels across the planet by 2040.
Improving estimates of carbon emissions
Carbon dioxide (CO₂) is a heat-trapping greenhouse gas (GHG) that is emitted into the atmosphere through human activities such as driving cars, burning coal or natural gas for energy, and deforestation. Researchers have developed a climate-carbon cycle model using machine learning algorithms to improve estimates of carbon emissions by 50 percent over traditional methods.
Detecting climate change-induced drought
Climate change-induced drought is a major cause of concern. Drought can lead to climate change refugees and also impact agriculture production leading to famines. There is an urgent need of detecting climate change-induced drought so that appropriate measures are taken up right away in areas, which may be affected soon by climate-induced droughts. Various deep learning techniques can be used to detect climate change-induced drought. Image recognition algorithms can be used in order to detect climate change-induced drought. For instance, deep convolutional neural networks (CNN) are known for their accurate performance and would thus be able to help people identify climate change-induced changes of regions rather quickly.
Predicting sea level rise
Machine learning algorithms are being employed to study climate data from satellites measuring sea level rise — which will help scientists determine how quickly seas could swell in the future. Researchers from the University of Colorado, Boulder, and Old Dominion University are using machine learning algorithms to make climate change predictions by studying sea level rise. The study seeks to understand how climate changes affect global sea levels in order for scientists to better predict future climate patterns that may pose a threat to coastal cities around the world.
Climate change action plan
Climate change action plans can be a part of overall government policy or local neighborhood levels initiatives like green roofs and urban heat island reduction. Deep learning techniques can be used for building data models on various factors such as city infrastructure and vegetation etc.