|Cloud filling of ocean colour and sea surface temperature remote sensing products over the Southern North Sea by the Data Interpolating Empirical Orthogonal Functions methodology|Sirjacobs, D.; Alvera-Azcárate, A.; Barth, A.; Lacroix, G.; Park, Y.; Nechad, B.; Ruddick, K.G.; Beckers, J.-M. (2011). Cloud filling of ocean colour and sea surface temperature remote sensing products over the Southern North Sea by the Data Interpolating Empirical Orthogonal Functions methodology. J. Sea Res. 65(1): 114-130. Dx.doi.org/10.1016/j.seares.2010.08.002
In: Journal of Sea Research. Elsevier/Netherlands Institute for Sea Research: Amsterdam. ISSN 1385-1101, more
Remote Sensing; Cloud Filling; Quality Control; Empirical Orthogonal Functions; Ocean Colour; SST; North Sea; English Channel
|Authors|| || Top |
Optical remote sensing data is now being used systematically for marine ecosystem applications, such as the forcing of biological models and the operational detection of harmful algae blooms. However, applications are hampered by the incompleteness of imagery and by some quality problems. The Data Interpolating Empirical Orthogonal Functions methodology (DINEOF) allows calculation of missing data in geophysical datasets without requiring a priori knowledge about statistics of the full dataset and has previously been applied to SST reconstructions. This study demonstrates the reconstruction of complete space–time information for 4 years of surface chlorophyll a (CHL), total suspended matter (TSM) and sea surface temperature (SST) over the Southern North Sea (SNS) and English Channel (EC). Optimal reconstructions were obtained when synthesising the original signal into 8 modes for MERIS CHL and into 18 modes for MERIS TSM. Despite the very high proportion of missing data (70%), the variability of original signals explained by the EOF synthesis reached 93.5% for CHL and 97.2% for TSM. For the MODIS TSM dataset, 97.5% of the original variability of the signal was synthesised into 14 modes. The MODIS SST dataset could be synthesised into 13 modes explaining 98% of the input signal variability. Validation of the method is achieved for 3 dates below 2 artificial clouds, by comparing reconstructed data with excluded input information. Complete weekly and monthly averaged climatologies, suitable for use with ecosystem models, were derived from regular daily reconstructions. Error maps associated with every reconstruction were produced according to Beckers et al. (2006). Embedded in this error calculation scheme, a methodology was implemented to produce maps of outliers, allowing identification of unusual or suspicious data points compared to the global dynamics of the dataset. Various algorithm artefacts were associated with high values in the outlier maps (undetected cloud edges, haze areas, contrails, and cloud shadows). With the production of outlier maps, the data reconstruction technique becomes also a very efficient tool for quality control of optical remote sensing data and for change detection within large databases.