Solomon White

Improving ocean model generalisation by including water type classification as pre-processing to the satellite image

24-Apr-24

Understanding and monitoring sea surface physical properties is crucial for gauging ocean health. This is particularly important for the global coastal ocean, which impact coastal communities through provisioning, regulating, supporting, and cultural ecosystem services. Climate change is leading to an increase in intensity and frequency of marine heat waves, eutrophication and acidification events which not only affect biodiversity but also impact coastal communities through food, human health and economic activities. Ocean colour models predicting sea surface properties such as sea surface salinity (SSS) and temperature (SST) are usually trained using pointwise ground based in-situ data . The model learning relationships between spectral signatures and water column inherent properties. However, this routinely operates at a pixel level and does not include any neighbourhood information which can inform on spatial distribution of features. The result is a loss of spatial information of the final product. Applying a clustering or segmentation approach to the input image can capture this spatial information by relating pixels to each other by spectral similarity or proximity. This paper aims to determine how using unsupervised learning to create clusters for water classification improves the performance of ocean colour regression models through appropriate algorithm selection as well as retaining spatial information.