Can Environmental Science and AI help achieve Sustainability?

Published in

Warwick Artificial Intelligence

6 min readJan 17, 2022

A talk by Kevin Donkers from the Met Office

See more from Kevin here: linkedin.com/in/kaedonkers

On 14th October, Warwick AI (https://warwick.ai) hosted a talk by Kevin Donkers on ‘Environmental Intelligence: Data, AI and Environmental Challenges’. Kevin Donkers is currently working at the Met Office Informatics Lab and doing a PhD in Environmental Intelligence at the University of Exeter.

The following article gives an insight into understanding the interface of Environmental & Sustainability science, Data Science and AI.

The Challenge

Kevin began by stressing the urgency of addressing Climate Change and Global Warming [1,2]. Humans are causing global warming by excessive burning of fossil fuels, livestock farming, use of fertilisers and deforestation, all of which release greenhouse gases that trap the Sun’s heat [3]. According to the IPCC 6th Assessment Report (2021) [4], humanity’s effect on the climate is indisputable and there exists a definite link with dangerous extreme weather events. More positively, we still have a chance to limit the warming, currently at 1.2 degrees, to 1.5 degrees, but this is becoming more of a challenge the longer we ignore it. Under the Paris Agreement of 2015 [5], the 1.5 degrees limit became an international goal because scientists agree at this level too many natural systems begin crossing points of no return.

Kevin then drew our attention to the Ecological Crisis describing how we are losing species and wild habitats at increasing rates which in return is impacting availability of pollinating insects and food, quality of water, quality of air and so on. Finally, Kevin emphasised the impact on human health, shocking us with the fact air pollution is the fifth highest cause of health-related deaths and explaining that zoonotic diseases such as Covid-19 are increasing in likelihood due to increased encroachment into wild spaces. On top of this, the increase in extreme temperature itself is dangerous to our health. The collapse of our environment and ecosystems is clearly being caused by society and economies, as well as having a direct negative impact on economies and human health. Being able to understand the complex interaction between all these factors is an important step forward we need to take.

The Opportunity

The first opportunity Kevin highlighted is the recent explosion in environmental data. According to a recent Forbes study, humans generate 2.5 quintillion bytes of data every day [6]. Much of this is from remote sensing whereby aerial sensors detect objects on the earth’s land surface or atmosphere [7], cheap sensor output from the increasingly large Internet of Things [8], numerical model outputs and proxies such as mobile phones. Most of this data is unstructured which is a problem machine learning can help solve.

Another opportunity is availability of cloud computing which gives more people access to computing power and removes the need for government funding. In addition, countless new analytical techniques are now available such as Bayesian statistics, deep neural networks (DNNs), gaussian processes and agent-based modelling. Old-style data science was to manually crunch numbers but now with data on the scale of Exabytes, it is only plausible to use automated methods like machine learning, especially deep learning. Tasks such as image classification, Natural Language Processing, and many other “unusual” data are processed significantly well by deep learning without any requirement of time-consuming feature engineering [9]. For example, the mortality rate due to air quality predicted using DNNs had better accuracy than the standard one hidden layer Neural Networks [10]. The combination of huge volumes of environmental data, increasingly powerful computing power and new analytical tools and methods, means the Met Office and many other organisations can now better extract useful information about the environment. This is called Environmental Intelligence.

The Response

At the Met Office Informatics Lab [11,12], Kevin is part of a data science and AI Agile research team looking at weather and Climate Science. If you are interested in joining, perhaps as an internship, get in touch via kevin.donkers@informatics.co.uk. You may also be interested in contacting Kevin about the 4-year PhD programme he is on, delivered by the UKRI Centre for Doctoral Training in Environmental Intelligence [13].

The programme aims to provide interdisciplinary doctoral training throughout the 4 years in data science & AI, environmental science and data governance, ethics and social impact. You produce your own research proposal and find a supervisor by the end of year 1 and then carry out the PhD research for 3 years, a challenging but rewarding process.

Kevin is also part of the Joint Centre for Excellence in Environmental Intelligence [14], recently launched in December 2020. This is a partnership between the Met Office and University of Exeter which focuses on promoting collaboration between industry, academia and the public sector for the use of environmental intelligence. It has three areas of focus: research, developing infrastructure and equipping people with skills in environmental intelligence, with several interesting ongoing projects relating to these areas.

Projects

One of the main projects is called CLIMAR. This is a mathematical framework for quantifying and visualising the risks associated with climate change to aid decision makers, using the disaster risk model where Risk = Hazard + Exposure + Vulnerability.

Relationship between the physical climate system, hazard, exposure and vulnerability producing risk. Developed from IPCC (2014).

As an example, CLIMAR has been used by Bristol City Council to assess the risk of urban heat, cities warming up considerably more than rural areas due to the large number of hard surfaces in them. Under the model hazard = UKCP18 projections of weather, exposure = building characteristics, vulnerability = demographic data. The results highlight which areas of the city are more at risk of excess deaths, indicating inequalities between population groups and the efficacy of heat mitigation strategies [15].

Another project is Precipitation Nowcasting, using deep generative techniques to better predict rainfall from radar data. This was a public, academic and private collaboration between the Informatics lab, University of Exeter and Deep Mind. Using radar images is considered the best way for forecasting rainfall. Data is observed from radar stations every 5 minutes and then injected into live running weather forecasting models; a process called data simulation. The issue is this new data introduces a ‘statistical shock’ due to differences between the old and new states, so there is poor reliability for predictions made within the next 10 minutes or 2 hours. The solution found uses ML to generate plausible futures based on what is happening currently and other information such as the orography (precipitation produced when moist air is lifted as it moves over a mountain range). It repeatedly trains using a discriminator to distinguish real data from data generated by the generator until it gives very plausible images for up to the next 20 minutes. When weather forecasters were asked which method they would prefer to use in forecasting, they said that the new deep learning technique was more useful in making plausible predictions showing just how much of a success this paper was [16].

There are many more projects to explore from looking at the effect of climate change on agriculture to bridging satellite data with surface sensors to creating a community platform for Big Data geoscience. Kevin has shown us that Environmental Intelligence is a very exciting and diverse emerging field of research.

A recording of his talk can be found on Youtube -

References/Resources

Can Environmental Science and AI help achieve Sustainability?

A talk by Kevin Donkers from the Met Office

Written by Emilia Matthews