University of Edinburgh logo British Antarctic Survey logo National Oceanographic Centre logo University of Leeds logo


Improving the utility of ocean colour satellites in marine systems science

Ocean colour satellites have revolutionised our capacity to characterise surface ocean ecosystems and biogeochemistry, allowing continuous global observation that would be impossible from any other platform. More ocean ecosystem variables are derived from ocean colour satellites (e.g. chlorophyll a, particulate (in)organic carbon, net primary production, size distribution) than there are degrees of freedom (i.e. the number of independent parameters/variables/quantities) in the raw data that satellites collect, so not all of these variables can be estimated independently – even without considering measurement error. It is therefore essential to quantify the degrees of freedom contained in satellite ocean colour data to establish how detailed a picture of marine ecosystems they can provide, and how best to use these degrees of freedom when characterising marine ecosystems. For example, it may not be possible to independently estimate the abundance of many phytoplankton types simultaneously, but it should be possible to do better than describing ecosystem characteristics (e.g. size structure, community composition, biogeochemical concentrations and rates) as just a function of chlorophyll and temperature. For understanding and modelling the contemporary ocean state and for predicting ecosystem responses to natural variability and anthropogenic climate change, it is critical to know where in between these two limits capabilities of past, current, and future satellites lie. Statistical methods to do so rigorously and efficiently are widely used in the atmospheric sciences, but have not been sufficiently applied in the oceanographic community.

This PhD’s objective is to quantify the degrees of freedom in past, current and future ocean color satellites’ measurements, and hence to what extent they can be used to characterise crucial marine ecosystem metrics. The student will accomplish this via statistical analyses of available datasets:
D1. Reflectances and derived ecosystem variables from current and past multispectral ocean color satellites (MODIS, SeaWiFS, Sentinel-3), globally and by season and region (parsing the ocean into e.g. Longhurst provinces and repeating analyses on spatial subsets).
D2. Output from a marine ecosystem model that resolves a complex virtual ecosystem’s optical imprint, to compare a surface ecosystem’s degrees of freedom with that of its optical signature.
D3. Hyperspectral radiometry data from the Marine Optical BuoY (MOBY) deployed near Hawai’i, to compare the degrees of freedom in hyperspectral versus multispectral measurements.

The student will apply to these datasets the established large-dataset ‘Information Content Analysis’ statistical method commonly used in the atmospheric sciences alongside more recently developed machine learning/artificial intelligence generalisations of this classical approach (especially Self-Organised Maps and Kernel Principal Component Analysis). These methods quantify the degrees of freedom of large, complex datasets, allowing the student to answer questions critical for the improvement of satellite algorithms to map and monitor surface ocean ecology and biogeochemistry:
Q1. How independent are the wavelengths measured by ocean colour satellites, and how does that independence vary with time, space, and ecosystem type, captured by degrees of freedom metrics?
Q2. How many degrees of freedom are retained in converting reflectance to ecosystem variables? Are there under-utilised wavelengths that could be better leveraged to further constrain these variables?
Q3. How many degrees of freedom are gained when spectral resolution is increased?
Q4: What improvements in measurement uncertainty or algorithm structure are required to maximise our capability to capture key ecosystem characteristics (i.e. those listed above)?

In identifying a maximal subset of quantities that can be estimated independently and simultaneously via ocean colour satellites, the outcome of these efforts will be essential knowledge for improving remote sensing algorithms to best use the data provided by historical and future satellite missions to study the changing oceans of today and the future. The student will be hosted at the National Oceanography Centre (NOC), Southampton within the Ocean Biogeosciences group, and will work in close collaboration with the Oceans group at the University of Edinburgh. The student will receive training in the use of ‘big-data’ statistical, machine learning, and artificial intelligence techniques, satellite remote sensing data, and numerical models. The student will have the opportunity to visit NASA (for a 3+ month placement) and MIT to learn about NASA satellite processing and the Darwin model. A solid understanding of statistics and some coding experience is preferred.