Machine learning approaches to improve retrieval of shelf sea algal biomass from ocean colour remote sensing.
Algal primary production in the marine environment constitutes approximately half of the global total, is therefore a very important element of the global carbon cycle and is the primary source of nutrition for the vast majority of life in the ocean. Various pieces of environmental legislation (EU MSFD, Water Directive etc) require governments to monitor the ecological state of national waters, with algal biomass being used as a proxy for eutrophication. There is also growing interest in monitoring for the presence of harmful algal blooms due to their potential impact on both aquaculture and public health more generally. Ocean colour remote sensing has radically transformed our ability to observe the growth and decay of algal blooms across the globe. However, the performance of standard algorithms for monitoring algal biomass is notoriously variable, with significantly lower performance in optically complex shelf seas. The aim of this project is to use state of the art machine learning approaches to improve understanding of local variability in the optical properties of natural waters and hence to inform interpretation of both historical ocean colour imagery and existing databases of in situ measurements of chlorophyll concentration. This will facilitate construction of a new, water-type specific approach to estimation of algal biomass for Scottish marine waters that will be integrated with regional hydrodynamic and ecosystem models to provide Marine Scotland and other Scottish public bodies with new tools for monitoring and predicting ecosystem status.
Aims & Objectives
In this project, we will use existing state of the art machine learning approaches to categorise optical signals from ocean colour imagery, to develop a database of optical water types. This will inform evaluation of the historic ocean colour time series, with Marine Scotland’s existing database of in situ Chl data used to determine Chl algorithm performance. Specific objectives are:
1. Use of machine learning approaches (region growing and support vector regression) to automatically segment historical remote sensing imagery (freely available from NASA, ESA) into optical water types which can be used to partition existing MS data sets and establish performance of existing and proposed Chl algorithms.
2. Identify subsets of imagery where algorithm performance can be quality assured and other areas where data quality is either uncertain or known to be degraded. This will inform future in situ sampling strategies, including use of near real time (NRT) satellite imagery to position research vessel assets.
3. Integrate ocean colour imagery with recently established fine resolution hydrodynamic models to establish potential drivers of ecosystem and algorithm performance variability, along with coupling future runs of the model with an ecosystem modelling component (ERSEM).
4. Apply image classification to multi-scale imagery in specific areas of interest or that are particularly data-rich (e.g. Loch Ewe, Stonehaven, Clyde Sea, Western Shelf) to resolve biogeochemical-physical interactions relevant to multiple end applications including: offshore installation licensing (e.g. aquaculture and marine renewable energy), fisheries, marine conservation and spatial planning.
Candidates for this project must have strong numeracy skills and experience with Matlab and / or other statistcal software would be beneficial.
This project is jointly funded by the Data Lab and MASTS Industrial Doctorate program and by the University of Strathclyde. The successful candidate will be based at the University of Strathclyde in the Physics Department but will work with a range of experts in machine learning (Dr Jinchang Ren, EEE, Strathclyde), remote sensing (Dr Jacqueline Tweddle, University of Aberdeen) and with Scottish Government scientists (Drs Alejandro Gallego, Matthew Gubbins and Eileen Bresnan, Marine Scotland, Aberdeen). The PhD is open to EU nationals and is fully funded for a total of 3.5 years, with preferred start date of 1st Oct 2018.
How good is research at University of Strathclyde in Physics?
FTE Category A staff submitted: 27.00
Research output data provided by the Research Excellence Framework (REF)
Click here to see the results for all UK universities