Nature recovery at scale using state-of-the-art Artificial Intelligence and Machine Learning techniques.
We will develop AI algorithms to monitor, predict, simulate and contrast the baseline change in nature with the impact of interventions, including nature-based solutions and financial investments, over large areas and over long time periods and will reduce the need for costly human monitoring on the ground. We will advance state-of-the-art AI approaches to combine different sources of data, including drones, satellite, survey data, reports and social media, that are robust to a range of environmental scenarios, data noise and model reliability, providing estimates with appropriate levels of uncertainty.

More information (PDF)

Mining the efficacy of Nature-based Solutions

Our AI researchers are working closely with the Nature-based Solutions Initiative (Department of Biology) to mine the evidence base for the effectiveness of nature-based solutions to climate change mitigation and adaptation. This work uses state-of-the-art transformer-based natural language processing to track the rapidly evolving field via published scientific articles or web-based text reports. Our AI algorithms will identify effective ways of working with natural ecosystems within the published literature, track sentiment towards restoration initiatives and filter key scientific reports.  Outputs will form the basis of guidance and tools for decision-makers and land managers.  Currently, it is hard for decision makers to access the best evidence, partly because that evidence is scattered among 1000s of journals and across several disciplines. Manual systematic reviews are extremely time-consuming and, as a result, poor decisions are being made that affect our futures. Deployment of AI approaches to speed up this process is urgently needed.

Related theme(s): Ecology

Mapping nature recovery at scale

For the first time we have the potential to map and model ecological connectivity across whole countries, map different farming approaches or infrastructure in fine detail and track the connectivity of biodiversity associated with different farming landscapes. Our AI team is developing state-of-the-art AI approaches to combine different sources of data, including drones, satellite, survey data and social media, that are robust to a range of environmental scenarios, data noise and model reliability.  We are harnessing big data and the ‘deep’ AI revolution to develop innovative technological approaches to assess nature recovery in both fine detail and at large spatial scales. Nature recovery needs to be delivered, monitored and evaluated in ways that incorporate both fine local context and broad spatial scale. Satellite remote sensing offers the scale, but until recently has usually been relatively crude and coarse resolution. Satellites and airborne platforms with hyperspectral, radar, lidar and other imagery are providing greater spatial resolution with regular repeat times to generate archives of nature loss and recovery. For example, our commercial satellite partners, Planet and ICEye, can deliver <50 cm resolution optical and radar images daily, allowing unprecedented potential for characterization and tracking of ecosystems in fine detail over seasons and years. This enables identification of tree species or meadow plant diversity through phenology and texture. Public social media posts provide a wealth of ‘ground validation’ but require automated text and image processing. In concert with this data richness, there is immense potential to use AI/machine learning to fuse, interpret and correlate data to work at both fine spatial resolution and large spatial scale in the midst of significant complexity. Developing such models and approaches is still challenging but would be transformative.

View project

Related theme(s): Scale

Robust ESG data for biodiversity

Financial institutions are increasingly aware of and interested in biodiversity- and nature- risks and opportunities, but such attempts have often been hindered by incomplete, incomparable and unreliable environment, social and governance (ESG) disclosure and scores. Importantly, existing ESG disclosure and scores tend to put great emphasis on emissions and climate-change, but fall short of comprehensively incorporating the location-specificity and breadth of nature and biodiversity concerns, especially in addressing nature recovery and entire-ecosystem uplift. Unlike emissions reduction, the assessment of such outcomes is multifaceted and are highly location specific. This in turn requires both geographically expansive but granular data to aid meaningful, impactful financial decision-making to ensure financial flows are aligned with nature outcomes, as well as to support nature-positive investments [link to WP 5, ‘Finance’]. Our AI team, in collaboration with the Oxford University Sustainable Finance group, the UNEP World Conservation Monitoring Centre, the Centre for the Environment and Hydrology (Wallingford) and the Satellite Applications Catapult are exploring innovative solutions to overcome existing challenges of ESG disclosure and analytics for biodiversity by tapping into the space of ‘discoverable data’ (e.g. satellite data, disclosures, traditional and social media, Internet of Things, in situ sensors). Our work explores how machine learning can deliver a data solution to comparably, comprehensibly and credibly measure and present granular, asset-level environmental risks and impact data, as well as co-benefits, to complement or fill gaps in existing nature disclosure datasets. This work will reduce investment risks faced by the finance sector, a key ask of our finance industry collaborators (inc. Barclays, Lombard-Odier) as well as scale up financial flow towards nature recovery.

Related theme(s): Finance