BIO-OPTICS AND BIOGEOCHEMISTRY III

DEVELOPMENT OF MERIS LAKE WATER ALGORITHMS: VALIDATION RESULTS

Koponen, Sampsa¹; Ruiz-Verdu, Antonio Ruiz-Verdú²; Heege, Thomas Heege³; Doerffer, Roland⁴; Brockmann, Carsten Brockmann⁵; Kallio, Kari ⁶; Pyhälahti, Timo ⁶; Pena, Ramon²; Polvorinos, Angel ⁷; Heblinski, Jörg Heblinski³; Ylöstalo, Pasi ⁶; Conde, Laura ²; Odermatt, Daniel ⁸; Estelles, Victor ⁹; Pulliainen, Jouni ¹⁰; Moreno, Jose⁹; Sorensen, Kai <sup>11

1</sup>Helsinki University of Technology (TKK) Otakaari 5a, Espoo, --, 02015, Finland; ²Centre for Hydrographic Studies - CEDEX, Pº Bajo de la Virgen del Puerto, 3, Madrid, 28005 , Spain; ³GKSS Forschungszentrum Geesthacht, Airport Oberpfaffenhofen, Gilching, D-82205 , Germany; ⁴GKSS Forschungszentrum Geesthacht, Max-Planck-Str., Geesthacht, 21502 , Germany; ⁵Brockmann Consult, Max-Planck-Str., Geesthacht, 21502 , Germany; ⁶Finnish Environment Institute (SYKE), Mechelininkatu 34a, Helsinki , 00251, Finland; ⁷University of Sevilla, Profesor García González, s/n, Sevilla, 41012 , Spain; ⁸University of Zurich, Winterthurerstr. 190, Zurich, CH-8057, Switzerland; ⁹University of Valencia, Dr. Moliner 50, Valencia, 46100 , Spain; ¹⁰Finnish Meteorological Institute (FMI), Erik Palménin aukio, Helsinki, 00560 , Finland; ¹¹Norwegian Institute for water research (NIVA), Gaustadalléen 21 , Oslo, NO-0349, Norway

During the ESA project “Development of MERIS Lake Water Algorithms” (Jan. 2007 – June 2008) two plug-in processors for deriving water quality information from lakes with MERIS data were developed and validated. The processors were developed for the BEAM toolbox and are based on the architecture of the MERIS Case 2 Regional –processor. They include an enhanced algorithm for atmospheric correction and new bio-optical algorithms for deriving inherent optical properties (IOP) and concentration (chl a and total suspended matter) data from atmospherically corrected reflectances. The neural networks for interpreting water bio-optical properties in boreal and eutrophic lakes were developed with IOP data from Finnish and Spanish lakes. Correction for the adjacency effect is tested with the ICOL-processor.

The processors were validated with in situ data collected from eleven lakes in Finland, Spain and Germany during April – August 2007. The lakes cover a wide range of water types from oligotrophic to hypereutrophic and to humic. The validation data include water leaving radiance reflectances, IOPs and concentrations of chl a and total suspended matter. Validation was also performed in Africa (Lake Victoria and Lake Manzalah).

The main results of the validation activities are presented and the applicability of the processor in different types of lakes is discussed. The results indicate the necessity of proper adjacency effect correction as well as necessity of algorithms trained to cover the observed water optical variability for different lake types.

BLENDING DISCRETE BIO-OPTICAL ALGORITHMS: A UNIFIED APPROACH TOWARDS REDUCING ERROR IN CHLOROPHYLL RETRIEVALS

Moore, Timothy¹; Campbell, Janet<sup>1

1</sup>University of New Hampshire OPAL, 142 Morse Hall, UNH, Durham, NH, 03824-3525, United States

Phytoplankton are the primary source of optical variability in the ocean, and thus their concentration in surface waters can be observed by ocean color sensors. Currently, there are two families of algorithms that derive chlorophyll a concentration (Chl) – a proxy for phytoplankton biomass. These are the empirical and semi-analytic algorithms. Such algorithms, parameterized from in-situ data, are currently used operationally to produce global maps of Chl. However, it is generally accepted that a single universal algorithm is not accurate everywhere, regardless of which type of algorithm is used.

Regional differences in the global empirical algorithm (OC4v4) have been shown to exhibit biases specific to the geographic ocean basin (e.g., Southern Ocean, North Atlantic). Similarly, semi-analytic algorithms require empirical parameterizations derived from in-situ data. Most models parameterize the relationship between the inherent optical properties (IOPs) of the water (absorption, scattering) and its constituents. On the global scale, IOPs vary over two orders of magnitude (Barnard et al. 1998) due to variations in particle size, pigment composition and packaging of algal cells, and overall particle composition. Since the constituents can vary from place to place and seasonally, it is believed that model parameterizations have to be locally derived for a particular water type, thus requiring the algorithm to decide when and where to use appropriate parameters.

Little attention has been given to the challenge of working with a suite of algorithms that switch at natural oceanic boundaries. How does one decide when to choose one algorithm over another? How can we avoid artificial discontinuities at boundaries where two algorithms meet? The work reported here develops a concept of blending algorithms for different water types using a fuzzy logic approach, and is shown to work with both families of algorithms.

REMOTE SENSING OF SUSPENDED PARTICULATE MATTER IN TURBID WATERS: STATE OF THE ART AND FUTURE PERSPECTIVES

Ruddick, Kevin¹; Nechad, Bouchra¹; Neukermans, Griet¹; Park, Youngje¹; Sirjacobs, Damien²; Beckers, Jean-Marie<sup>2

1</sup>MUMM 100 Gulledelle, Brussels, --, B-1200, Belgium; ²GHER University of Liège, Liège, Liège, B-4000, Belgium

The state of the art and future perspectives for remote sensing of suspended particulate matter (SPM) in turbid waters will be assessed and illustrated with results from recent and ongoing research.

Interest in remote sensing of SPM in turbid waters is motivated by the environmental and economic importance of sediment transport in coastal waters. Applications include the optimisation of dredging/dumping operations, environmental impact of offshore construction activities, understanding of geomorphological change, etc. and are generally addressed by model studies supported by remote sensing data. While Total Suspended Matter (TSM) is a relatively easy parameter to detect optically even with the crudest of instruments, there is a need to continually improve the accuracy, the quality control and the spatio-temporal coverage of satellite-derived TSM maps. There is also a desire for more detailed information on particle size and composition.

Algorithms for TSM retrieval will be presented, including a hyperspectrally-calibrated algorithm which can be adapted to almost any remote sensor from AVHRR or SPOT to MODIS or MERIS. Spectral variation of the algorithms coefficients obtained by calibration from in situ measurements of water-leaving reflectance and TSM is shown to correspond, via optical closure, to spectral variation of the relevant inherent optical properties. Results for satellite data products from MODIS and MERIS are compared with other algorithms including the MERIS neural network algorithm. Methods for automatic quality control of satellite data products are described. Finally, some future perspectives are outlined including the use of spatio-temporal correlations to improve quality control and fill data fields for cloudy periods/regions and the use of geostationary sensors to give a dramatic improvement in temporal resolution.

THE IDENTIFICATION OF OPTICALLY DISTINCT WATER TYPES AT GLOBAL AND REGIONAL SCALES: IMPLICATIONS FOR ALGORITHM DEVELOPMENT AND VALIDATION.

Dowell, Mark¹; Berthon, Jean-François¹; Zibordi, Giuseppe<sup>1

1</sup>Joint Research Centre - European Commission Via E. Fermi, Ispra, --, 21027, Italy

There is a requirement to extensively characterise the optical variability of marine environments, so that developed algorithms are applicable at the global scale but remain quantitatively accurate for both the open ocean and regional/shelf seas. It is unclear at present that a single parameterisation of reflectance model sub-components is capable of achieving this satisfactorily. The present study addresses this concern by analysing the results of a statistical method (based on fuzzy logic) to discriminate distinct optical water types. The method itself is particularly conducive to the dealing with the intrinsic uncertainty present in in-situ and satellite datasets, their spatial/temporal variability and the transition from one type to another. Eight distinct optical water types have been identified at the global scale, and these are mapped on a time-series of global and regional products from SeaWiFS, MODIS and MERIS. Alternative bandsets for the same sensors are considered to analyse the sensitivity of the classification to bands which contain greater uncertainty, and to account for band shifting from one sensor to the other. Two specific applications are illustrated as examples of the methods potential. The first makes use of the in-situ datasets, NOMAD representative of global variability and the JRC bio-optical database more focused on European coastal waters, in the identification of optical classes. When the statistics from these analysis are applied to satellite data, they allow to identify “black holes” in the optical-representativity of existing datasets at global and regional scales. The second application considers the parameterisation of bio-optical algorithms for both empirical (i.e. an OC4-Prov), where the parametric fit coefficients are obtained for the distinct optical water-types, and semi-analytical algorithms, where inherent optical property sub-component models (for aph*, Scdom and Sb_bp) are paramterised for each optical water type.

THE COMPARISON OF NORMALIZED WATER LEAVING RADIANCES FROM SEAWIFS AND MODIS CONTRIBUTES TO DEFINING THE SPATIAL DISTRIBUTION OF THEIR UNCERTAINTIES.

MELIN, Frederic¹; ZIBORDI, Giuseppe<sup>1

1</sup>EC - JRC EC - JRC, via Fermi, 2749, TP272, ISPRA, --, 21027, Italy

The intercomparison of coincident satellite data from different sensors can contribute to characterizing the uncertainties in normalized water leaving radiances Lwn, particularly with respect to their spatial variations, and appears as a valuable support to the information provided by Lwn validation datasets obtained from field observations inherently limited in geographical coverage.

Considering two ensembles of measurements X(i) and Y(i) (i=1,N), the discrepancies between the two records can be documented by estimators like mean unbiased percent differences or the coefficient of determination. Moreover, taking X and Y as linear functions of a true value Z, X(i)=Z(i)+d(i), and Y(i)=alpha+beta Z(i)+e(i), we can define the additive and multiplicative biases alpha and beta between the two distributions, and zero-expectation measurement errors d and e associated with X and Y. Assuming the equality of the variance sigma of d and e, alpha, beta and sigma can be computed. In that context, sigma is a conservative estimate of the satellite product radiometric uncertainties.

The analysis relies on apparent and inherent optical properties (IOP) derived from SeaWiFS LAC and MODIS imagery covering the European seas for 2002-2007. The comparison is conducted on large databases with time (daily) and space (2-km) resolutions comparable to those of satellite overpasses. The SeaWiFS Lwn are expressed at the MODIS center-wavelengths with bio-optical relationships using satellite derived IOPs. Generally, sigma, in units of Lwn, decreases with wavelengths and shows a relative spatial homogeneity. Coherently, relative differences between SeaWiFS and MODIS show larger spatial variations, particularly in the blue, with higher differences in coastal regions, in the Baltic and Black Seas. Cases of seasonal dependence are also observed, like larger differences in Mediterranean regions in winter. The intercomparison is put in relation with validation results obtained with large datasets of field observations collected by autonomous systems in the North Adriatic and Baltic seas.

DETECTION OF ICE AND ICE-WATER MIXING PIXELS FOR THE MODIS OCEAN COLOR DATA PROCESSING

Wang, Menghua¹; Shi, Wei<sup>1

1</sup>NOAA/NESDIS/STAR E/RA3, Room 102, 5200 Auth Road, Camp Springs, MD, 20746, United States

Currently, the data processing for deriving ocean color products from the Moderate Resolution Imaging Spectroradiometer (MODIS) has no specific ice and ice-water detection procedure. The near-infrared (NIR) reflectance threshold at the MODIS 869 nm band, which has been used to discriminate clear sky from clouds (cloud masking) for the standard ocean color data processing, can eliminate most of the ice pixels. However, there are still many cases for which the ice and ice-water mixing pixels have been misidentified as ocean waters in the current ocean color data processing, leading to errors in the MODIS-derived ocean color product (e.g., chlorophyll-a concentration). This is particularly true for most of the ice-water mixing cases. For atmospheric correction using the shortwave infrared (SWIR) method, which also uses the SWIR reflectance for the cloud masking, the problem of ice misidentification is getting even worse. In this presentation, we describe a method for detection of ice and ice-water mixing pixels for the MODIS ocean color data processing. Using the MODIS-derived normalized water-leaving radiances at 412, 555, and 859 nm, a scheme for ice and ice-water detection has been developed and tested for producing MODIS global ocean color products. With the new ice detection scheme, pixels with ice and/or ice-water mixing can be discriminated, flagged and masked out. The detection results are compared with the MODIS ice mask product produced from the MODIS Land team, as well as ice product data obtained from the NOAA National Ice Center. We show improved results from the new masking algorithm for the purpose of the MODIS ocean color data processing, in particular, for the detection of ice-water mixing pixels.