AI and Data Science in Environmental Research

News about natural disasters is dominated by unsettling images of melting glaciers, deforestation, forest fires, extreme floods and other potentially destructive events that are becoming more frequent due to climate change. These news stories raise awareness of and concern for our environment, creating a sense of urgency for the public. Research-based evidence and scientific data along with public-led advocacy, have been compelling government and policymakers to take immediate action on sustainability crises that have existed for decades.

In environmental research, images used for social awareness are only a small part of the climate change discourse and a fraction of data is gathered daily from different sources. An ocean of scientific data serves as an invaluable source of information that not only shapes our understanding of climate catastrophe but also guides data-based monitoring, forecasting and simulations of different conditions.

Insights from data analysis and predictive models help us understand the earth’s processes and systems better and prepare to respond to climate change. Stronger mitigation measures for reducing greenhouse gas emissions, and changes in electricity networks, transportation, buildings, industry and land use are required. Research-based mitigation strategies should guide the qualitative and quantitative increase in energy efficiency, enhancing the capacity of carbon sinks (such as forests and oceans) and sustaining natural systems. The science shows that even the most effective climate change mitigation will not be enough to entirely prevent further climate change impacts, thus making the need for adaptation unavoidable [Nelson, Adger and Brown, 2007]. Climate change adaptation aims to build resilience by reducing harms caused by extreme weather events and other natural disasters by building the capacity to prepare for, recover and adapt to the changing climate. This necessitates cross-disciplinary research that combines knowledge of our environmental systems utilising heterogeneous cross-sector data through the development of climate models and predictions.

In this report, we discuss AI in Environment and Sustainability and provide examples from ASG under the following four categories:

• Monitoring the environment: AI tools, data models and workflows have advanced environmental monitoring capability by combining earth observations from satellites with those from the ground sensors, bringing together data of different quality, timescale and resolutions, with novel mechanisms to reduce uncertainty.
• Forecasting environmental change: The application of deep learning on accumulated data has helped forecast environmental changes accurately and use energy systems effectively, creating possibilities to mitigate the extent of climate change.
• Simulating the human cost of climate change: Beyond giving deeper insights into environmental changes, multi-domain data have been used to develop effective AI approaches to build climate interventions, simulate their impact on populations and inform effective policies for implementation.
• Adapting to climate change: Bringing combined knowledge from other areas of AI, Environment and Sustainability projects have undertaken a broad role in responding to the environmental crisis through technology and policies for mitigation and adaptation.

To demonstrate the successful implementation of AI-based solutions in these areas, among many exciting projects, we highlight Scivision which applies computer vision models and datasets at various scales – from microscopic to satellite; IceNet, a deep learning-based AI predictive tool that produces accurate Arctic sea ice forecasts, outperforming physics-based models in forecasting Arctic sea ice change; and DyME-CHH microsimulation-based toolkits that assist local authorities in creating action plans to address issues ranging from pandemic to climate crisis.

Following the standard of “as open as possible, as closed as necessary”, researchers have involved different stakeholders to build user-friendly open-source computing platforms, provide open access to resources for understanding AI technology as well as open up underlying data and software. Further practices have been systematically applied for sharing outputs and knowledge to enhance accessibility, interpretability and reuse of different components from the projects. Considerable attention has been given to building broadly applicable, reliable and explainable models that take the uncertainties and limitations of curated data into account. Development and dissemination of reusable tools, methods and data have ensured that solutions from environmental research can be deployed across a wide variety of applications in different contexts, compounding the long-term impact of the work.

Building a resilient future will require increasing public awareness of climate change, providing effective interventions and accelerating the adoption of modern solutions in different sectors. By offering data-intensive insights and sustainable solutions for different climate change scenarios, the AI/ML work discussed in this paper has informed policies and decision-making at all levels, from individuals and organisations to government and society. Impactful bodies of work include a solar energy generation forecasting project that used citizen science approaches to crowdsource solar panel location information in the UK and forecast the production of solar-based renewable energy in order to minimise the circulation of fossil fuel-based electricity production. Another project called Synthetic Population Catalyst (SPC) integrates population data to create an open dataset of synthetic populations that can be used in complex simulations to help inform policymakers. Tools like EnergyFlex have allowed local authorities to explore inequalities in energy efficiency and target homes in need of retrofits as well as predict which residents may require support for fuel poverty.

The Turing’s leadership on the applications of AI in environmental research have built stronger evidence for innovative interventions, taken alongside nature-based measures for restoring natural ecosystems and making our society more resilient to climate change.