A Statistical Modeling Framework for Characterising Uncertainty in Large Datasets: Application to Ocean Colour

Abstract

Uncertainty estimation is crucial to establishing confidence in any data analysis, and this is especially true for Essential Climate Variables, including ocean colour. Methods for deriving uncertainty vary greatly across data types, so a generic statistics-based approach applicable to multiple data types is an advantage to simplify the use and understanding of uncertainty data. Progress towards rigorous uncertainty analysis of ocean colour has been slow, in part because of the complexity of ocean colour processing. Here, we present a general approach to uncertainty characterisation, using a database of satellite-in situ matchups to generate a statistical model of satellite uncertainty as a function of its contributing variables. With an example NASA MODIS-Aqua chlorophyll-a matchups database mostly covering the north Atlantic, we demonstrate a model that explains 67% of the squared error in log(chlorophyll-a) as a potentially correctable bias, with the remaining uncertainty being characterised as standard deviation and standard error at each pixel. The method is quite general, depending only on the existence of a suitable database of matchups or reference values, and can be applied to other sensors and data types such as other satellite observed Essential Climate Variables, empirical algorithms derived from in situ data, or even model data.

A Global Ocean Observing System (GOOS), Delivered Through Enhanced Collaboration Across Regions, Communities, and New Technologies

“…For biogeochemical variables, Amin et al (2015) assessed GOES satellite-based ocean color products using in situ networks (Amin et al, 2015). Land et al (2018) used a database of satellite in situ matchups to generate a statistical model of satellite uncertainty as a function of its contributing variables for ocean color chlorophyll-a and showed that most errors are correctable biases (Land et al, 2018). Martínez-Vicente et al (2017) examined the differences among phytoplankton carbon (Cphy) estimations from six satellite ocean color algorithms by comparison with in situ estimates, and large (>100%) biases have been found (Martínez-Vicente et al, 2017).…”

Applications of DINEOF to Satellite-Derived Chlorophyll-a from a Productive Coastal Region

“…Improved results may be achieved in the future by using a longer time series to better constrain the EOFs near long gaps, or by using a longer temporal covariance matrix filter [47]. Additionally, the quality of input data is an important consideration for using DINEOF and is affected by the inherited uncertainties of the chla sat [57]. In this study, quality control prior to DINEOF processing included standard ocean color flags, a reduced straylight filter (3 × 3), removal of chla sat pixels exceeding 40.00 mg m −3 (Section 2.2.1), and removal of scenes and pixels with less than 2% ocean coverage (Section 2.3).…”