Year: 2023
Primary Supervisor: Prof Joseph Holden <J.Holden@leeds.ac.uk>
Institution: University of Leeds (School of Geography)
Academic Supervisors: Dr Encarni Medina-Lopez <Encarni.Medina-Lopez@ed.ac.uk> (University of Edinburgh, School of Engineering), Dr Susanna Ebmeier <S.K.Ebmeier@leeds.ac.uk> (University of Leeds, School of Earth and Environment)
Research Themes: Artificial Intelligence, Climate, Earth Observation, ecosysetms, water
Project Partners: Wetlands International
Research Keywords: Artificial Intelligence, Atmosphere, Earth Observation, ecosystems, machine learning, Soil
Supervisors: Pro Joseph Holden (School of Geography, University of Leeds); Dr Susi Ebmeier (School of Earth & Environment, University of Leeds), Dr Encarni Medina-Lopez (School of Geosciences, University of Edinburgh). External partner: Wetlands International. Student based at: School of Geography, University of Leeds
Background
Wetlands provide a range of vital ecosystem services and may store around half of all terrestrial soil carbon. However, wetlands have been threatened by agricultural expansion, urbanisation, pollution and climate change. This puts at risk wetland carbon stocks, which can be rapidly lost to the atmosphere when wetlands degrade1,2. There are no globally consistent, high resolution wetland mapping products that can distinguish between wetland types and that can map change over time. Such a tool would allow vastly enhanced estimates of wetland carbon stocks and changes in net carbon loss/gain over time. These data would be crucial for underpinning IPCC assessments and provide huge improvements to the IPCC wetland supplement. They are also essential data to inform global efforts in wetland conservation and restoration.
Aim
This project will develop a novel, automated, high resolution wetland change map at global scale. In doing so it will investigate recent changes in wetland extent and patchiness, wetland carbon stocks, and examine drivers of change in wetland extent.
Objectives
- To use Sentinel 1 and 2 datasets combined with machine learning techniques to accurately map and distinguish wetland types at high resolution globally.
- To develop an automatically-updating EO-driven wetland change detection tool that produces maps and statistical outputs related to key variables.
- To use the change map tool along with other data products to determine potential changes in wetland carbon stocks and other key ecosystem services.
Methods
The project will be supervised by a multi-disciplinary (ecology/hydrology/EO/machine learning) and diverse team (female/male, early career/senior) with a pedigree of high quality outputs, impacts and learned society awards, providing the student with an excellent support environment to develop their project. The CASE partner would also provide an excellent applied learning environment to ensure the project provides outputs that can be applied by the global policy and practice community. For Objective 1 – new high resolution global wetland map: the PhD researcher would utilise global Sentinel 1 and Sentinel 2 imagery combined with ground truth data from the project partner, Wetlands International, and machine learning algorithms to produce a new 10 m resolution global wetland map. This will be a novel product but build on existing research at the University of Leeds which has examined wetland distributions in Africa3,4. There will be significant challenge in distinguishing between wetland types across different regions. AI feature recognition techniques will be used alongside learning algorithms to classify and validate wetland types. To deliver Objective 2 an automation process developed via machine learning will be developed for the new global wetland map to obtain and process Sentinel imagery in real time to show change over time in wetland extent and types. This can be used to track natural seasonal changes as well as anthropogenically-driven change in wetlands. Researchers, NGOs and policy communities will then have access to a vital tool to aid decision-making and provide alerts to wetland changes. Spatial statistical tools will seek to answer questions about changing wetland patch distributions in space/time/form and relationships to other datasets (e.g. driven by the CASE partner’s needs), such as climatological, topographical, population, or protected status data. For Objective 3 the researcher will apply models of carbon fluxes and stocks based on global empirical assessments (e.g. ref. 2), and use the outputs from objectives 1 and 2 to determine potential changes in global carbon stocks and fluxes caused by changes in wetland extent in different regions for 2015-2025. Water and nutrient flux models could also be included in this assessment of change.
References: 1. Evans et al 2021 Nature 593, 548–552. 2. Zhou et al 2022 Nature Geoscience, 15, 627-632. 3. Garba et al 2023 Wetlands Ecology and Management in press. 4. Garcin et al 2022, Nature, https://doi.org/10.1038/s41586-022-05389-3
