Systematic Analysis of Ocean Colour Uncertainties
Lavender, Samantha
Pixalytics Ltd, UNITED KINGDOM

Earth Observation ocean colour products, such as maps of chlorophyll concentration, from future spaceborne missions will have continuity with existing / historical missions such as Envisat/MERIS and MODIS-Terra in terms of both the algorithms and products alongside the introduction of new approaches. Space agencies have also recognised the need for error and/or uncertainty estimates so that end users are provided with knowledge that allows them to have confidence in the data they're using. Previous research (Lavender et. al. 2012) has primarily focused on the impact of the input data and algorithm/modelling uncertainties, but it's known that there are other contributory factors to the uncertainties e.g. the auxiliary data, such as wind speed estimates, and the variability at sub-pixel scales for inhomogeneous surfaces and the scientific methodology.

Semi-analytical approaches can often be model inversions based on multiple non-linear regression techniques, such as Levenberg-Marquardt, and artificial neural network (NNets). In the development process, a critical step is choosing/simulating a representative test dataset. Research by Luo and Mountrakis (2012) has demonstrated that choosing randomly extracted test datasets from an image can be less effective, for land classification accuracy, than situations where there has been a pre-analysis of pixel variability.

This research is focused on investigating pixel based uncertainty metrics that will aid non-specialist users in determining the uncertainty and therefore potential limitations in space derived information. Initially, the research is being based on an analysis of MERIS and MODIS imagery, but future research would also include Sentinel-3 data. The ultimate aim is to develop methodologies and implement solutions that will allow error/uncertainty estimates to be calculated on a pixel-by-pixel basis whilst being efficiently processed for operational implementation.

Lavender, S., Fanton d'Andon, O., Kay, S., Bourg, L., Emsley, S., Gilles, N., Nightingale, T., Quast, R., Bates, M., Storm, T., Hedley, J., Knul, M., Sotis, G., Nasir-Habeeb, R., Goryl, P. and Sentinel-3 L2 Products and Algorithm Team. 2012. Applying Uncertainties to Ocean Colour Data. Metrologia, 49, S17, doi:10.1088/0026-1394/49/2/S17

Luo, L. and Mountrakis, G. 2012. A multiprocess model of adaptable complexity for impervious surface detection, International Journal of Remote Sensing, 33:2, 365-381