Reliable satellite estimates of chlorophyll-a concentration (Chl-a) are needed in coastal waters for applications such as eutrophication monitoring. However, because of the optical complexity of coastal waters, retrieving accurate Chl-a is still challenging. Many algorithms exist and give quite different performance for different optical conditions but there is no clear definition of the limits of applicability of each algorithm and no clear basis for deciding which algorithm to apply to any given image pixel (reflectance spectrum). Poor quality satellite Chl-a data can easily reach end-users. To remedy this and provide a clear decision on when a specific Chl-a algorithm can be used, we propose simple quality control tests, based on MERIS water leaving reflectance (ρw) bands, to determine on a pixel-by-pixel basis if any of three popular and complementary algorithms can be used. The algorithms being tested are: 1. the OC4 blue-green band ratio algorithm which was designed for open ocean waters; 2. the OC5 algorithm which is based on look-up tables and corrects OC4 overestimation in moderately turbid waters and 3. a near infrared-red (NIR-red) band ratio algorithm designed for eutrophic waters.

Using a dataset of 348 in situ Chl-a / MERIS matchups, the conditions for reliable performance of each of the selected algorithms are determined. The approach proposed here looks for the best compromise between the minimization of the relative difference between In situ measurements and satellite estimations and the number of pixels processed. Conditions for a reliable application of OC4 and OC5 depend on ρw412/ρw443 and ρw560, used as proxies of coloured dissolved organic matter and suspended particulate matter (SPM), as compared to ρw560/ρw490, used as a proxy for Chl-a. Conditions for reliable application of the NIR-red band ratio algorithm depend on Chl-a and SPM. These conditions are translated into pixel-based quality control (QC) tests with appropriately chosen thresholds. Results show that by removing data which do not pass QC, the performance of the three selected algorithms is significantly improved. After combining these algorithms, 70% of the dataset could be processed with a median absolute percent difference of 30.5%. The QC tests and algorithm merging methodology were then tested on four MERIS images of European waters. The OC5 algorithm was found to be suitable for most pixels, except in very turbid and eutrophic waters along the coasts where the NIR-red band ratio algorithm helps to fill the gap. Finally, a test was performed on an OLCI-S3A image. Although some validations of water reflectance are still needed for the OLCI sensors, results show similar behavior to the MERIS applications which suggests that when applied to OLCI data the present methodology will help to accurately estimate Chl-a in coastal waters for the next decade.