As we continue to face serious environmental challenges, it is increasingly tempting to imagine a innovative relationship between AI and climate change. There are a growing number of results supporting the notion that society could derive significant benefits from the use of AI in tackling the impacts of climate change. AI’s contribution can lead to groundbreaking adaptation measures for resilience and disaster management. Here, we’ll explore how AI can reduce our vulnerability to climate impacts, and how it can support society in its recovery from climate-induced disruptions.
AI and Climate Change Ecology: monitoring environmental degradation
Tracking and understanding the changes in our ecosystems and biodiversity has traditionally been achieved through costly and time-consuming manual and on-the-ground observations. AI can offer better ways to provide large amounts of data as well as better insights. This could help transform the fields of ecology, wildlife biology, zoology, conservation biology, and animal behaviour.
Such technologies are being used in the Yunque National Forest in Puerto Rico for example, where the effects of Hurricane Maria on the distribution of tree species are being assessed through aerial imaging. Dr Uriarte, a Biologist from Columbia University and Dr Zheng, an associate director for education at Columbia’s Data Science Institute, created a machine learning algorithm that teaches computers to identify tree species in aerial photos taken by NASA. By imitating humans’ cognitive learning processes, the algorithm uses data to identify patterns, trends, and make predictions. It produced great results, and was able to identify tree species in an aerial photograph the size of a postage stamp in 0.6 seconds, in comparison to the 80 seconds it would take humans. Similarly, it took four hours for the algorithm to identify all the species of trees in a large part of the forest, something that took five people 167 days.
There are many other examples of AI being used to further research, such as the use of AI to combat poaching, automate animal identification, follow animal migration patterns, and automate inferences on fish abundance from reef 3D models. Bringing environmental scientists and AI together can provide larger pools of data that can be used as evidence to support better policy decisions and better assess the impact of ecological interventions. As Dr Uriarte says in an interview with Columbia University News “our collaboration has allowed me to ask questions I wouldn’t have been able to ask and get answers I wouldn’t have been able to get.”
AI and Climate Change Infrastructure: the need for resilient design
Extreme weather patterns are likely to put our built environment under increasing stress. In order to comply with several UN sustainable development goals, infrastructure will need to adapt to the uncertain changes that climate change will bring. For example, resilient transportation infrastructure will need to be built to deal with increased flooding. Studies have shown how using multiple data sources, such as flood hazards, traffic information, weather data, cellular network data, CCTV, as well as historical and proxy data, can help better identify vulnerable roads, predict future flood damage, as well as the effect floods on human mobility. Similar methods can be used for other climate-change related hazards such as droughts and wildfires.
Climate change will also exacerbate the inequality in the access to drinking water. Due to the uncertainty surrounding regional climate conditions, designs for water resource planning, such as dams, need to be adaptive. This type of flexible infrastructure can however be very expensive, and it is difficult for planners to know when to trigger adaptive actions. Advances in search algorithms, and the use of the Bayesian model have shown a lot of promise in identifying areas where flexible infrastructure will be reliable and effective, and other areas where it would be too costly, favouring more traditional infrastructure.
These uses cases will dramatically improve the ability to better assess and anticipate the human impacts of climate change and make better decisions about investments in infrastructure.
AI and Climate Change insecurity: protecting the social fabric
Climate change risks affecting a wide array of communities by causing food insecurity and mass migration, among other things. The effects of extreme weather on crop and food production are clear in many areas around the world, such as North America, West Africa, and East Asia. Data can be used to guide intervention by surveying the risks of food shortages and predicting long-term risk areas.
UN Global Pulse found high correlations between airtime credit purchases and the results of surveys that track the consumption of certain foods. They thus argue that mobile phone data could serve as a proxy indicator for food-spending. Social media data and credit card transitions could serve similar ends in training models to link this decontextualized data with the reality on the ground. Using semi-supervised learning processes with various sources of data can yield interesting results. As described in the ecology section, aerial imagery can be used to identify crop-diseases, allowing for communities to be alerted much more rapidly. AI can also potentially help minimise food waste in global and local supply chains by, among other things, predicting supply and demand and identifying efficient shipping routes.
AI can also predict large-scale immigration patterns by using proxies like social media and phone call records, or aerial imagery. This can be used to track migrants’ and refugees’ movements and enact more accurate rescue operations. One such initiative is the use of intelligent drone swarms for rescue operations in the Mediterranean. A research group at Stanford University have also used a machine learning algorithm to assign better placement locations for individual refugees, based on socioeconomic data from more than 30,000 refugees aged 18 to 64, resettled in the US from 2011 to 2016. Researchers believe that the algorithm has the potential to greatly increase the employment rate of refugees and help policymakers. However, systems that survey vulnerable groups can unfortunately be exploited. It is thus particularly important to involve policymakers, experts, and ethicists, and to ensure that the use of ethical frameworks and privacy considerations are included in the design of these algorithms.
AI and Climate Change Destruction: responding to a crisis
Through more accurate prediction and coordination, we can better prepare for the natural disasters and health crises that will arise from climate change. While the use of social media data is already prevalent, further machine learning tools can contribute to the disaster response and public health disciplines.
Extreme weather is likely to increase the chance of certain epidemics. Diseases forecasting systems have been built using web data, such as FluBreaks, based on Google Flu Trends. Moving beyond surveillance, research has also focused on improving diagnostics in places where equipment, expertise, and general logistics are lacking. A good example is the diagnosis and treatment of malaria, a disease that some believe will become more widespread with climate change. In the areas worst affected, reliable diagnoses are difficult to obtain and false positives or negatives can be fatal or lead to drug resistance. A great deal of attention is thus placed on other ways to effectively diagnose patients, such as through blood smear image analysis with mobile phones. By using the existing mobile phone hardware, the only new hardware necessary to combine them is an adapter to mount the phone onto the microscope eyepiece or trinocular tube. Researchers found that this method is a promising low-cost method for more accurate diagnosis. Similar work has been done in the development of a machine learning system (Ubenwa) that allows the diagnosis of Birth Asphyxia through an automated analysis of the infant’s cry.
When it comes to disaster prevention and response, the creation of maps from aerial photos, as well as data collection from social media have proven to be useful. Aerial photos can help create real-time accurate maps, that can assist in creating feasible evacuation plans, deliver relief, and can also help in assessing the damage caused, by comparing pre-and post-disaster photos quickly and accurately. Data from social media can also help inform relief efforts as people report their plight and their locations can be assessed.
AI and Climate Change: a work in progress
AI, like any technology, has the potential to benefit society and to contribute to the fight against climate change. It can enable better monitoring, accelerate the process of scientific discovery, improve efficiency, and provide innovative potential solutions to various social problems. While this use of AI might not be immediately profitable for the private sector, a serious effort should be made to develop these initiatives further on a not-for-profit basis. However, AI is no silver bullet. A lot of the research and innovations surrounding the use of AI are not scaled and may not be representative of global use cases. In order to apply them widely there would be a need for a lot more data, and collecting it can raise its own policy and ethical issues.
For a comprehensive analysis of the topic of AI and climate change, I direct you to the Climate Change AI Paper written by 22 leading AI specialists, including Andrew Ng, Yoshua Bengio, Demis Hassabis, and Jennifer Chayes.
Written by Zoe Caramitsou-Tzira