Figure 2 - available via license: Creative Commons Attribution 3.0 Unported
Content may be subject to copyright.
Density scatter plots of in situ vs. satellite-retrieved Chl a for all algorithms providing meaningful values. The line of best fit (blue) and that of equal value (black) are superimposed, with relevant statistics.
Source publication
A 15-year (1997–2012) time series of chlorophyll a (Chl a) in the Baltic Sea, based on merged multi-sensor satellite data was analysed. Several available Chl a algorithms were sea-truthed against the largest in situ publicly available Chl a data set ever used for calibration and validation over the Baltic region. To account for the known biogeochem...
Similar publications
Various drought indices have been developed to monitor drought, which is a result of climate change, and mitigate its adverse effects on water resources, especially in agriculture. Vegetation indices determined by remote sensing were examined by many recent studies and shed light on drought risk management. In the current study, one of the 25 drain...
Citations
... Satellites enable quantitative assessments of algal biomass (Harvey et al., 2015) that can be used as an ecological indicator (Platt and Sathyendranath, 2008) to understand the eutrophication effects. Recent studies indicate the potential of remote sensing to supplement and optimize marine monitoring programs (Harvey et al., 2015;Hossain et al., 2015;Kratzer et al., 2014;Markager et al., 2019), However, while there are some clear advantage of using satellites for obtaining high temporal and spatial resolution data, studies have also shed light towards strong needs to improve the accuracy and precision of the applied algorithms to obtain more reliable estimate of parameters such as chlorophyll (O'Reilly et al., 2000;Pitarch et al., 2016). ...
... A major factor adding to the complexity of the use of a robust chlorophyll algorithm, especially in the optically complex region such as the Baltic region, relates to the presence of high concentrations of CDOM caused by humic rich freshwater inflow (Kowalczuk et al., 2006;Mélin and Vantrepotte, 2015;Pitarch et al., 2016). For optically complex water bodies with high CDOM concentrations, whose properties change significantly throughout the seasons, it is challenging to obtain accurate values of chlorophyll concentrations from satellites (Sathyendranath, 2000), leading often to an overestimation of chlorophyll concentrations through the use of standard algorithm (Attila et al., 2013;Darecki and Stramski, 2004). ...
... The chlorophyll concentrations from satellite sensors were derived from merging of SeaWiFS, MODIS/AQUA-MERIS-VIIRS remote-sensing reflectance similar to Gohin et al. (2019), processed by the combination of Copernicus Marine Environment Monitoring Services (CMEMS) Baltic Sea-Specific algorithm (BalAlg) (Pitarch et al., 2016), and the Free University Berlin (FUB) neural network (v4.01, (Schroeder et al., 2007)) algorithm. BalAlg is an adaptation of the OC4v6 algorithm for the Baltic Sea, modified by applying the coefficients of the linear regression between in situ and OC4v6-derived chlorophyll data to reduce the biasness between satellite observations and in situ measurements. ...
We investigated the use of multisensory satellite data to determine long-term changes in surface chlorophyll concentrations using a 19-year (1998–2016) time series of chlorophyll data in the Danish Kattegat region of the Baltic Sea. Merged satellite estimates (SeaWiFS-MODIS/Aqua-MERIS-VIIRS) were compared with in situ ship based time series from four monitoring stations situated with increasing distance from land and nutrient sources. In situ and satellite derived estimates showed similar trend in chlorophyll with several fold higher values closer to land. Satellites aligned very well with in situ estimates in the open water stations but showed significant differences in magnitude and inter-annual variability, in particular in shallow coastal waters. Some systematic deviation was observed with satellite underestimating the growing season average for the earlier periods (1998–2002) and overestimating for the later period (2012–2016) compared to in situ estimates. Comparing growing season chlorophyll means over the 19 year period showed increasing magnitude and variability in nearshore and shallower areas, most pronounced for the satellite derived chlorophyll. Satellites overestimated chlorophyll in nearshore areas 2–4 fold, despite excluding shallow nearshore areas with possible benthic interferences from the analyses. This bias needs further validation and requires correction to improve the overall applicability of satellites for long-term monitoring of chlorophyll in the Kattegat region. From analysis of normalized data, we developed a simple correction model, which reduced deviations considerably between methods, underlying the importance of in situ data for application of satellite observations. While significant deviations were observed from in situ data, satellites are clearly advantageous in the much higher temporal and high spatial coverage they provide. Multisensory satellites can, however, not be used currently as a standalone technique for long-term assessment of chlorophyll. They require validation with in situ measurements, which provide essential data for calibration, validation and correction of satellite based estimates. A complementary use of multisensory satellite and in situ measurements therefore remains essential to assess trends in the ecological status of optically complex waters such as the Kattegat region of the Baltic Sea.
... Suppose that multiple sensors simultaneously record and monitor the cross-country skier's movement, so that R 1 and R n represent the data obtained by the first sensor and the nth sensor. If R 1 and R n both satisfy the characteristics of Gaussian distribution [24], to express the degree of difference between the two, the formula is expressed as ...
The status and role of science and technology in the field of modern competitive sports have become increasingly prominent. The construction of a scientific training command system is of great significance for improving the scientific level of the training process and deepening the digital cognition of ski training. This paper is based on the multisensor combination to conduct a digital research on cross-country skiing training, aiming to conduct in-depth research on the realization of human motion capture and the theory of motion inertial sensing. To build a scientific, formal, and malleable ski training program, the requirements for data acquisition, recording, and analysis are quite strict. For this, it is necessary to use scientific and reasonable tools combined with multiple algorithms to process information and data. During the experiment, accelerometers, gyroscopes, and magnetometers are selected as sensors to receive motion information, and recognition algorithms for identifying weightlessness, hybrid filtering algorithm, displacement estimation algorithm, and kinematic principles are adapted to process multisensor data using information integration technology. A human body motion model was established based on kinematic principles, and a cross-country skiing motion measurement program was designed. The experimental results show that, according to the combination of multisensing and video platform, the athlete’s posture prediction is adjusted, and the action on the track is more consistent, which can accelerate the athlete’s skiing speed and the size of the inclination angle to a large extent. It can affect the direction of the athlete’s borrowing force and the adjustment of gravity during the exercise. The tilt angle is expanded from 135° to 170°, and it can maintain good continuity during the exercise.
... Bootstrapping methods (Efron, 1979) provide an adequate methodology to strengthen the uncertainty estimation in algorithm selection exercises, as such are often used in the OCR community for the selection of CHL algorithms Pitarch et al., 2016). Alas they are used seldom for radiometric comparisons as they are not recommended in the current protocols. ...
Multiple approaches have been used by the Ocean Colour community for validating satellite-derived products using in situ data, with most of them derived from mainly two approaches, one suggested by Bailey and Werdell (2006) (BW06) and one suggested by Zibordi et al. (2009a) (Z09), each with a different set of quality checking and spatiotemporal collocation criteria. The question remains what sort of information is added or missed when choosing one over the other. In this work, the differences among validation approaches were determined by using a common dataset of in situ and satellite data. The match-up exercise was separated into two groups of datasets based on the spatial resolution of the sensors to be validated. Sentinel-3A/OLCI data were selected as a representation of medium spatial resolution sensors, and two validation approaches were selected to this match-up dataset. The high spatial resolution sensors were represented by Sentinel-2A/MSI data, and three validation approaches were tested. Data from the AERONET-OC network were chosen as the common in situ dataset. For Sentinel-3A/OLCI, the number of match-ups varies depending on the validation approach used. Bailey and Werdell (2006) produces 20% more match-ups for Sentinel-3A/OLCI due to its more relaxed filtering criteria compared to the criteria applied by Zibordi et al. (2009a). The validation metrics vary between different validation methods giving a different impression of accuracy of the satellite products. Also, the satellite data selected by BW06 have a statistical distribution with a higher median and standard deviation when compared to Z09. Similar findings are also confirmed for the match-up analysis conducted for Sentinel-2A/MSI. Therefore, although a common reference dataset was used, the validation statistical results were influenced by the validation approach selected. This does not suggest that one validation protocol is better than the other, but it implies that validation statistics reported in different studies may not always be directly comparable. Additionally, it was determined that BW06 could be a better fit when trying to obtain a sufficient number of match-ups for calibration purposes in the shortest time.
... Bailey and Werdell (2006) proposed a 5 × 5 or 3 × 3 box for averaging satellite data with a pixel size of 1 km × 1 km. For satellite data with bigger pixel size (i.e. 4 km or 9 km), a single-pixel without spatial windowing in the Baltic Sea (a semi-enclosed marginal sea) (Pitarch et al., 2016) and coastal areas of the eastern China Sea (Hao et al., 2019), a 3 × 3 and 5 × 5 box along coasts of Southern Ocean (Johnson et al., 2013), and spatial averaging in Mediterranean Sea (Volpe et al., 2019) have been used. In this study, at the first step the spatial homogeneity of Chl-a in a 3 × 3 box centered on the location of an in situ sample was tested using the coefficient of variation (CV) and a two-tailed critical value of Student's t-test (Zar, 1996;Bailey and Werdell, 2006). ...
... In addition, to select the best match-up pairs, only in situ match-up data pairs were selected which had difference of <15% with the associated Chl-a pixel (Harding et al., 2005). Also, the matchup data pairs which were outside the range of 1/20 and 20 times of the in situ Chl-a averages, were considered as outliers and discarded (Pitarch et al., 2016). As a result, 32%, 33%, 41%, 29%, 20%, and 18% of available in situ samples were considered for evaluation of OC-CCI, MGCL, MODIS, MERIS, SeaWiFS, and VIIRS datasets, respectively (Table 1). ...
... The merged multi-sensor Chl-a datasets have been used in several studies to evaluate the Chl-a retrieval algorithms in marginal seas. Most of these studies have shown that the OC-CCI and GlobColour Chl-a datasets can be used properly for monitoring the chlorophyll concentrations in the coastal turbid waters (Johnson et al., 2013;Brotas et al., 2014;Sá et al., 2015;Johnson et al., 2013;Pitarch et al., 2016;Mélin et al., 2017;Hao et al., 2019;Volpe et al., 2019). ...
Persian Gulf is a highly turbid and optically complex marginal sea. Satellite remotely sensed chlorophyll-a (Chl-a) products have been widely used in this marine area, despite uncertainties due to complex oceanic and atmospheric optical properties. In this study, accuracy of daily merged multi-sensor Ocean Color Climate Change Initiative (OC–CCI), Copernicus Marine Environmental Monitoring Service (CMEMS) GlobColour, and OC5 single-sensor products of SeaWiFS, MERIS, MODIS, and VIIRS datasets were evaluated using in situ chlorophyll concentrations collected from 2008 to 2018 in the Iranian territorial waters. The results showed that the MAPE, RMSE, bias(δ) and R² values between in situ and satellite-4 derived Chl-a vary in the range of 131–273%, 0.38–0.69, 0.27–0.43, and 0.27–0.44, respectively. Satellite-derived Chl-a concentrations overestimated the field observations by 131–232% in the northern parts of the middle deeper areas and up to 173–273% in Iranian coastal areas. The OC-CCI and GlobColour merged datasets overestimate the Chl-a concentrations by 19% more than the average of OC5 single-sensor products. Systematic errors were observed in the log-normal distributions of difference between in situ and satellite-derived Chl-a. A bootstrapping-like assessment was performed to eliminate the systematic errors, and to reduce the bias from satellite datasets. The results of statistical adjustment were applied on daily matchup data-pairs and time-series datasets. Furthermore, an inter-comparison was made between merged multi-sensor (OC–CCI, GlobColour CHL2) and OC5 single-sensor (SeaWiFS, MERIS, MODIS, and VIIRS) Chl-a products using 8-day time-series datasets during the years 2000–2020. The results showed that the OC5 single-sensor Chl-a datasets were more consistent with OC-CCI than GlobColour merged CHL2 in the deep and shallow regions of the study area. In contrast, the merged multi-sensor products were more similar to each other than the OC5 single-sensor datasets in the river plume zone. After performing statistical adjustment of time-series datasets, the bias values between OC-CCI, GlobColour CHL2, and OC5 single-sensor datasets decreased by 14–22%, and the single-sensor datasets showed more similarity to OC-CCI datasets.
... Optical remote sensing of water quality is dramatically improving by developing new approaches for integration of EO data with in-situ measurements in support of critical activities such as the effective monitoring of complex coastal regions and aquatic ecosystems (Bonansea et al., 2019a(Bonansea et al., , 2019b. Indeed, the combination of WCC data obtained from in-situ and satellite observations is an optimal remote sensing approach to provide high temporal resolution information on water surface properties in coastal regions (Lawford et al., 2013;Pitarch et al., 2016). ...
A R T I C L E I N F O Keywords: Long-term water quality monitoring Remote sensing Integration Multi-sensor satellites In-situ measurements Hyperspectral observations MERIS MSI OLCI ENVISAT Sentinel-2 Sentinel-3 2SeaColor MODTRAN Radiative Transfer modeling Coastal waters The Wadden Sea A B S T R A C T Recently, there have been significant efforts in the integration of in-situ and satellite observations for effective monitoring of coastal areas (e.g., the Copernicus program of the European Space Agency). In this study, a 15-year diurnal variation of Water Constituent Concentrations (WCCs) was retrieved from multi-sensor satellite images and in-situ hyperspectral measurements using Radiative Transfer (RT) modeling in the Dutch Wadden Sea. The existing RT model 2SeaColor was inverted against time series of in-situ hyperspectral measurements of water leaving reflectances (R rs [sr −1 ]) for the simultaneous retrieval of WCCs (i.e., Chlorophyll-a (Chla), Suspended Particulate Matter (SPM), Dissolved Organic Matter (CDOM)) on a daily basis between 2003 and 2018 at the NIOZ jetty station (the NJS) located in the Dutch part of the Wadden Sea. At the same time, the existing coupled atmosphere-hydro-optical RT model MOD2SEA was used for the simultaneous retrieval of WCCs from time series of multi-sensor satellite images of the MEdium Resolution Imaging Spectrometer (MERIS) onboard ENVISAT, Multispectral Instrument (MSI) onboard Sentinel-2 and Ocean and Land Colour Instrument (OLCI) onboard Sentinel-3 between 2003 and 2018 over the Dutch Wadden Sea. At the NJS, a direct comparison (Taylor diagram and statistical analysis) showed strong agreement between in-situ and satellite-derived WCC values (Chla: R 2 ≥ 0.70, RMSE ≤7.5 [mg m −3 ]; SPM: R 2 ≥ 0.72, RMSE ≤5.5 [g m −3 ]; CDOM absorption at 440 nm: R 2 ≥ 0.67, RMSE ≤1.7 [m −1 ]). Next, the plausibility of the spatial variation of retrieved WCCs over the study area was evaluated by generating maps of Chla [mg m −3 ], SPM [g m −3 ], and CDOM absorption at 440 nm [m −1 ] from MERIS and OLCI images using the MOD2SEA model. The integration of the spatio-temporal WCC data obtained from in-situ measurements and satellite images in this study finds applications for the detection of anomaly events and serves as a warning for management actions in the complex coastal waters of the Wadden Sea.
... Shelf seas are more sensitive to terrestrial runoff and bottom resuspension, and upwelling coastal areas are known for the high phytoplankton biomass. Enclosed and semi-enclosed seas follow their own dynamics Kopelevich et al., 2004;Pitarch et al., 2016). ...
... SBAL displays seasonal dynamics influenced by an intense summer bloom (Pitarch et al., 2016) that leads to a green-brown color (FU = 8-9). The rest of the year, color is dominated by high amounts of CDOM. ...
The Forel-Ule (FU) color comparator scale is the oldest set of optical water types (OWTs). This scale was originally developed for visual comparison and generated an immense amount of data, with hundreds of thousands of observations being gathered from the last 130 years. Since recently, the FU scale is also applicable to remote sensing data. This has been possible thanks to an optical characterization of the 21 FU colors in terms of the (x,y) CIE standards and new algorithms that convert remote-sensing reflectances (R rs) from satellite-borne ocean color sensors to FU. R rs-derived hue angle and FU have been recently applied with success in the assessment of color variability of lakes and specific shelf areas, but an evaluation over global oceanic waters is still missing. By clustering global climatological ESA-OC-CCI v2.0 R rs with the derived FU, we obtain a set of R rs to be used as optical water types (OWTs). Diffuse attenuation coefficient, Secchi disk depth and chlorophyll concentration are also associated to the FU classes. The angular distances of a given R rs to the two nearest FU classes are proposed as simple and robust membership indexes, adding up to one. We also evaluate the advantages and limitations of FU and the hue angle as monitoring tools over the full marine range, from the most oligotrophic areas to the turbid and productive coastal zones. The first 7 FU indexes cover 99% of global surface waters. Unlike the hue angle, that resolves all spatio-temporal color variations, the FU scale is coarse as a monitoring tool for oligotrophic waters as all the subtropical gyres saturate to FU = 1, while the color of other seas varies across 2, 3 or even 4 FU classes. We illustrate the introduction of a new "zero" FU class that increases monitoring resolution at the blue end of the color range. Finally, we show how optical diversity varies across the color range and compare several sets of OWTs from a color perspective. Overall, we provide a valuable and self-consistent dataset that enhances the usefulness of the FU scale by converting it to useful information for the oceanographic community. This OWT scheme keeps the advantages of other datasets, like being useful to study ocean color product quality and characterize the uncertainties, but also allows to continue to monitor long-term change in optical diversity over the global ocean color. Integration into the optical modules of ecosystem models can help verify past simulations that predate the satellite age, through comparisons with in-situ FU data collected at the time.
... These studies indicated that the inversion model developed at the Free University of Berlin, the "FUB model" (Schroeder et al., 2007a(Schroeder et al., , 2007b, estimated chl-a concentrations better than the other examined processors, such as MERIS ground segment processor (MEGS) and the Case-2 regional processor (C2R) (Doerffer and Schiller, 2007). The assessment of coastal water bodies is not feasible with the Baltic Sea chl-a products generated by the Copernicus Marine Environment Monitoring Service (CMEMS) (Pitarch et al., 2016;CMEMS Quality Information Document, 2016) due to its 1-4 km spatial resolution and its limited accuracy in comparison to in situ data over the Baltic Sea (r 2 = 0.2 and r 2 = 0.46 for 1 km product and reprocessed time series (4 km, REP), respectively, Pitarch et al., 2016). ...
... These studies indicated that the inversion model developed at the Free University of Berlin, the "FUB model" (Schroeder et al., 2007a(Schroeder et al., , 2007b, estimated chl-a concentrations better than the other examined processors, such as MERIS ground segment processor (MEGS) and the Case-2 regional processor (C2R) (Doerffer and Schiller, 2007). The assessment of coastal water bodies is not feasible with the Baltic Sea chl-a products generated by the Copernicus Marine Environment Monitoring Service (CMEMS) (Pitarch et al., 2016;CMEMS Quality Information Document, 2016) due to its 1-4 km spatial resolution and its limited accuracy in comparison to in situ data over the Baltic Sea (r 2 = 0.2 and r 2 = 0.46 for 1 km product and reprocessed time series (4 km, REP), respectively, Pitarch et al., 2016). ...
Earth Observation (EO) offers spatially and temporally unique data for generating information required under various environmental regulations for assessing the status of surface waters. These requirements, which are laid down in, for example, European Union directives and the Clean Water Act in the United States, share two core elements with respect to status assessments: 1) the status assessment is done using discrete classes, typically for water bodies, sub-areas or critical sites representative for certain area of interest, and 2) phytoplankton chlorophyll a (chl-a) is one of the main variables considered. We analysed the benefits of using chl-a concentrations derived from EO data for the status assessments specified in the EU Water Framework Directive (WFD). Our study focused especially on EO observations' ability to capture extreme and transient events (such as instances of cyanobacteria blooms) more frequently than the monitoring-station data conventionally employs. The accuracy of EO-based chl-a assessment was studied for, in all, 129 Finnish water bodies in the area of the Baltic Sea, in Northern Europe. Natural conditions in this coastal area – particularly its multitude of small bays, numerous estuaries, and mosaic of islands – impose exceptionally strict requirements for an EO instrument's spatial resolution. The analysis revealed that an instrument with a 300 m resolution, such as the MEdium Resolution Imaging Spectrometer (MERIS) or Ocean and Land Colour Instrument (OLCI), can be used to estimate the water quality in 62% of these water bodies. Processing of MERIS data into chl-a concentrations by means of a FUB inversion model demonstrated good accuracy relative to monitoring stations' measurements for the open-water season in 2003–2011. This extensive dataset showed a 23% difference in modal values between EO- and station-sampling-based chl-a concentrations. The bias in EO chl-a estimates was found to increase with low Secchi disk depth, elevated turbidity, and the presence of intensive phytoplankton blooms. The monitoring-station and EO data showed similar distributions of chl-a observations for a given day and location, a finding that supports the comprehensive use of EO-derived chl-a concentrations in assessment. For determination of a water body's status, the EO data required but also allowed for statistical analysis that differs from what has typically been utilised with sparse measurements from monitoring-station data. The geometric mean or the mode of the EO observations was found to represent the main bulk of the chl-a concentrations well statistically. In contrast, the arithmetic mean of EO observations yields chl-a concentrations that are roughly 1.1–1.6 μg/l higher and hence can lead to over-estimation in the associated status assessment. This paper also presents a new approach applicable for evaluating the validity of EO-based algorithms for any coastal water area requiring assessment. With this quality-grade (QG) method, the EO chl-a estimation accuracy is rated in terms of three grades, with water bodies taken as the evaluation units. For this, the method utilises statistical differences between EO and station-sampling chl-a concentrations and applies background information on optical properties obtained from measurements at routine-monitoring stations. The QG method showed the EO-based chl-a accuracy to suffice for assessing the status of 65% of the coastal water bodies examined. At concentrations representing the threshold for the target of “good status” under the WFD, the EO approach produced 0.6 μg/l higher chl-a values than the stations' sampling did. The MERIS results point to clear benefits of using OLCI-based status assessment throughout the Sentinel-3 era.
... Previous research in the Baltic Sea evaluated the performance of standard and Case 2-specific Chl a (Harvey et al., 2015;D'Alimonte et al., 2012;Kratzer et al., 2008;Melin et al., 2007;Reinart and Kutser, 2006) and k d (Stramska and Swirgon, 2014;Doron et al., 2011;Pierson et al., 2008) ocean colour products. Regionally calibrated blue-green ratio versions of OC4v6 (Pitarch et al., 2016;Darecki and Stramski, 2004) have allegedly improved the accuracy of Chl a retrieval in the Baltic Sea, but do not work for waters where CDOM dominates the absorption in the blue. Using longer wavelengths such as red-to-green (Woźniak et al., 2014) and red-to-near-infra red (Koponen et al., 2007;Krawczyk et al., 1997;Matthews, 2011) is therefore advisable in these optically complex, CDOM-rich waters. ...
... R rs is low in the blue region due to the high absorption by CDOM, and the performance of R rs (490)/R rs (560) ratios were better compared to R rs (443)/R rs (560) ratios since the R rs (490) signal was stronger than R rs (443), which may be relevant for Chl a algorithms such as OC3 and OC4 when they use the R rs (490)/R rs (560) ratios. Pitarch et al. (2016), however used the regional calibration of OC4v6 to map the Chl a concentration in the Baltic Sea, but they found that OC4v6 over-estimates Chl a resulting in a R 2 = 0.43 and bias of 0.44, suggesting that Chl a algorithms for the Baltic Sea, should use longer wavelengths than R rs (490). In their analysis, they also included data from the Kattegat and Skagerrak which proved to be more accurate with blue: green Chl a algorithms than for the Baltic Sea area. ...
The Baltic Sea is a semi-enclosed sea that is optically dominated by coloured dissolved organic material (CDOM) and has relatively low sun elevation which makes accurate ocean colour remote sensing challenging in these waters. The high absorption, low scattering properties of the Baltic Sea are representative of other optically similar water bodies including the Arctic Ocean, Black Sea, coastal regions adjacent to the CDOM-rich estuaries such as the Amazon, and highly absorbing lakes where radiometric validation is essential in order to develop accurate remote sensing algorithms. Previous studies in this region mainly focused on the validation and improvement of standard Chlorophyll-a (Chl a) and attenuation coefficient (kd) ocean colour products. The primary input to derive these is the water-leaving radiance (Lw) or remote sensing reflectance (Rrs) and it is therefore fundamental to obtain the most accurate Lw or Rrs before deriving higher level products. To this end, the retrieval accuracy of Rrs from Medium Resolution Imaging Spectrometer (MERIS) imagery using six atmospheric correction processors was assessed through above-water measurements at two sites of the Aerosol Robotic Network for Ocean Colour (AERONET-OC; 363 measurements) and a shipborne autonomous platform from which the highest number of measurements were obtained (4986 measurements). The six processors tested were the CoastColour processor (CC), the Case 2 Regional processor for lakes (C2R-Lakes), the Case 2 Regional CoastColour processor (C2R-CC), the FUB/WeW water processor (FUB), the MERIS ground segment processor (MEGS) and POLYMER. All processors except for CC had small average absolute percentage differences (ψ) in the wavelength range from 490 nm to 709 nm (ψ < 40%), while other bands had larger differences with ψ > 60%. Compared to in situ values, the Rrs(709)/Rrs(665) band ratio had ψ < 30% for all processors. The most accurate Rrs in the 490 to 709 nm domain was obtained from POLYMER with ψ < 30% and coefficients of determination (R²) > 0.6. Using a score system based on all statistical tests, POLYMER scored highest, while C2R-CC, C2R-Lakes and FUB had lower scores. This study represents the largest data base of in situ Rrs, the most comprehensive analysis of AC models for highly absorbing waters and for MERIS, conducted to date. The results have implications for the new generation of Copernicus Sentinel ocean colour satellites.
Remote sensing studies published up to now show that the performance of empirical (band-ratio type) algorithms in different parts of the Baltic Sea is highly variable. Best performing algorithms are different in the different regions of the Baltic Sea. Moreover, there is indication that the algorithms have to be seasonal as the optical properties of phytoplankton assemblages dominating in spring and summer are so different. We modelled 15,600 reflectance spectra using HydroLight radiative transfer model to test 58 previously published empirical algorithms. 7200 of the spectra were modelled using specific inherent optical properties (SIOPs) of the open parts of the Baltic Sea in summer and 8400 with SIOPs of spring season. Concentration range of chlorophyll-a, coloured dissolved organic matter (CDOM) and suspended matter used in the model simulations were based on the actually measured values available in literature. For each optically active constituent we added one concentration below actually measured minimum and one concentration above the actually measured maximum value in order to test the performance of the algorithms in wider range. 77 in situ reflectance spectra from rocky (Sweden) and sandy (Estonia, Latvia) coastal areas were used to evaluate the performance of the algorithms also in coastal waters. Seasonal differences in the algorithm performance were confirmed but we found also algorithms that can be used in both spring and summer conditions. The algorithms that use bands available on OLCI, launched in February 2016, are highlighted as this sensor will be available for Baltic Sea monitoring for coming decades.
