The QMEE CDT Project proposal database

Welcome to the QMEE CDT Project proposal database. This is a live list of projects proposals put forward by PIs across the CDT partner institutions

PIs/Supervisors will continue to add projects to this list over the next few months, so do keep checking back! You can search the projects using the box below: simply enter some text and press Search to do a text search across all the database fields. If you want to search more finely, the search tool also allows you to search on particular details of the project descriptions: you will see these finer search options appear if you click on the search box.

Click on the view button next to a project to get the full proposal description. If you want to download project details, either for all projects, or for a subset you have searched for, then click on the 'Download details' button.

Deep learning in remotely sensed image understanding for biodiversity and ecosystem services
Forecasting the impacts of climate change and habitat destruction on biodiversity and ecosystem services represents one of the greatest challenges in ecology. One promising practical tool is agent-based modelling, which aims to determine population response to environmental change as the emergent property of simple rules that govern an individual’s physiology, life-history and behaviour across complex landscapes. These ecological models are used to both evaluate risk and help develop management strategies to reduce vulnerability but they often require detailed spatial and temporal information on the microclimate and resources that are difficult to obtain. Recent technological advances in remote sensing, specifically the use of unmanned aerial vehicles (UAVs), offer the possibility not only to map the microclimate in detail but also monitor changes in plant traits. The use of this technology to create ‘energy landscape’ maps for species has potentially profound implications for modelling food webs and ecosystem services. Technological advances are now enabling high spatial resolution (< 1m) hyperspectral and thermal infrared (IR) imaging from UAVs, covering the visible, near infrared and shortwave infrared region. Recent work has revealed strong relationships between hyperspectral reflectance and various leaf traits from grassland plants and tropical trees, a number of which are known to determine the quality of food for insect herbivores (e.g. leaf water and dry matter content, C/N ratio). In the study of butterflies, there is both a large body of experimental work to parametrize the individual energy budgets of species in respect to both temperature and food plant quality and, critically, long-term abundance records (UKBMS) to validate models using Approximate Bayesian Computation methods. This studentship will investigate the potential of using remote sensing to map both the microclimate and the spatial and temporal variability of the butterfly’s host plants, in terms of their cover density and quality. In order to establish quantitative relationships between information captured by hyperspectral and infrared images and the reality of the butterfly’s host plants, computer vision and machine learning technologies will be applied to systematically analyse the captured images with regards to the plant composition, traits and condition. Convolution Neural Network (CNN), as a deep learning mechanism, will be used in the learning process to understand remotely sensed information. The process requires training data to support. This can be done by using the empirical PLSR model that has already been tested in principle for tropical forest canopies [1], and the radiative transfer PROSAIL model [2] parameterised for grasslands. To obtain reliable training data, the student will also need to monitor a series of small (1 m2) experimental plots of single and mixed plant species seeded at Sonning Farm. The study will produce a food density and quality map which, along with IR surface temperature, would provide an ‘energy map’ to parametrize a butterfly agent-based model (ABM) under development. [1] Asner et al., New Phytologist 204:127-39 (2014) [2] Jacquemoud et al., Remote Sensing of Environment 113:S56–S66 (2009)
Hong Wei
Richard Walters
France Gerard, Climate Systems Group, Centre for Ecology and Hydrology, Wallingford
Computing, Quantitative data analysis
Hong Wei
There are two aspects regarding quantitative skills. For computing: to analyse a large number of hyperspectral and infrared images and to establish quantitative relationships between information contained in the images and the plant density and quality. For ecosystems: to produce an agent-based model in which the plant condition indicates the butterfly population quantitatively.
Deep learning used in automated understanding information captured by hyperspectral and infrared images is a new area attractive researchers. Building information fusion into the understanding will better make use of the data sources. Quantitative analysis of the relationship between the images and plant conditions spatially and temporally is a novel approach.
Recent studies make us believe that there are strong relationships between hyperspectral reflectance and various leaf traits from grassland plants and tropical trees, a number of which are known to determine the quality of food for insect herbivores. Use of visible, near infrared and shortwave infrared images processed by the quantitative methods, the above ecological theory is being addressed.
The real world application to be address is, by using the quantitative methods to process image data from remote sensing, the ecological forecasting of agricultural landscapes can be achieved in an advanced way which could be carried out much faster, cost effective and potentially more accurate in a large scale. This newly developed method can be used to replace the current low efficient methods.
A key challenge of this project is to determine how the variability in plant composition, traits and condition is best captured in field by hyperspectral reflectance/surface temperature data collected within the field of view (FOV) of an UAV. This approach will ultimately be used to assess the effects of predicted drought stress and landscape management which has a significant impact to the field.
This project uses quantitative methods to process C/N ratio, leaf water content and other relevant traits from the remote sensing images and explore how best to capture this information in natural mixed species plots. As a multi-disciplinary project this proposal has potentially profound implications for our understanding of food web dynamics and ecosystem services in a changing climate.
Population ecology, Ecosystem-scale processes and land use
Follow training opportunities will be offered to the successful candidate: 1) Basic statistical, mathematical and analytical skills; 2) Technical skills through the collection of plant spectra, plant traits and samples for nutrient analysis; 3) Programming skills through using R for statistical analysis and Matlab for PLSR analysis and 4) Deep learning in remotely sensed data understanding.
University of Reading
2017-10-01 20:02:15