Validation of Envisat MERIS algorithms for chlorophyll retrieval in a large, turbid and optically-complex shallow lake

• Published in: Remote sensing of environment Vol. 157; pp. 158 - 169
• Main Authors: Palmer, Stephanie C.J., Hunter, Peter D., Lankester, Thomas, Hubbard, Steven, Spyrakos, Evangelos, N. Tyler, Andrew, Présing, Mátyás, Horváth, Hajnalka, Lamb, Alistair, Balzter, Heiko, Tóth, Viktor R.
• Format: Journal Article
• Language: English
• Published: Elsevier Inc 01.02.2015
• Subjects: Algorithm; Algorithms; Archives; Chlorophyll-a; Chlorophylls; Freshwater; Imaging spectrometers; Inland waters; Lake Balaton; Lakes; MERIS; Neural networks; Processors; Retrieval; Validation; Hungary, Balaton L; Validation; MERIS; Chlorophyll-a; Lake Balaton; Algorithm; Inland waters
• ISSN: 0034-4257; 1879-0704
• DOI: 10.1016/j.rse.2014.07.024

The 10-year archive of MEdium Resolution Imaging Spectrometer (MERIS) data is an invaluable resource for studies on lake system dynamics at regional and global scales. MERIS data are no longer actively acquired but their capacity for global scale monitoring of lakes from satellites will soon be re-established through the forthcoming Sentinel-3 Ocean and Land Colour Instrument (OLCI). The development and validation of in-water algorithms for the accurate retrieval of biogeochemical parameters is thus of key importance if the potential of MERIS and OLCI data is to be fully exploited for lake monitoring. This study presents the first extensive validation of algorithms for chlorophyll-a (chl-a) retrieval by MERIS in the highly turbid and productive waters of Lake Balaton, Hungary. Six algorithms for chl-a retrieval from MERIS over optically complex Case 2 waters, including band-difference and neural network architectures, were compared using the MERIS archive for 2007–2012. The algorithms were locally-tuned and validated using in situ chl-a data (n=289) spanning the five year processed image time series and from all four lake basins. In general, both band-difference algorithms tested (Fluorescence Line Height (FLH) and Maximum Chlorophyll Index (MCI)) performed well, whereas the neural network processors were generally found to much less accurately retrieve in situ chl-a concentrations. The Level 1b FLH algorithm performed best overall in terms of chl-a retrieval (R2=0.87; RMSE=4.19mgm−3; relative RMSE=30.75%) and particularly at chl-a concentrations of ≥10mgm−3 (R2=0.85; RMSE=4.81mgm−3; relative RMSE=20.77%). However, under mesotrophic conditions (i.e., chl-a<10mgm−3) FLH was outperformed by the locally-tuned FUB/WeW processor (relative FLH RMSE<10mgm−3=57.57% versus relative FUB/WeW RMSE<10mgm−3=46.96%). An ensemble selection of in-water algorithms is demonstrated to improve chl-a retrievals. •6 chlorophyll-a (chl-a) retrieval algorithms are validated for a large turbid lake.•Extensive in situ matchup dataset allowed insight into temporal and spatial nuances.•Fluorescence Line Height (FLH) most accurately retrieved chl-a overall.•Neural network-based processors were generally inaccurate, even when locally-tuned.•FUB/WeW processor outperformed FLH under mesotrophic conditions.