Validation of empirical and semi-analytical remote sensing algorithms for estimating absorption by Coloured Dissolved Organic Matter in the Baltic Sea from SeaWiFS and MODIS imagery

Abstract

An extensive bio-optical data set obtained from field measurements was used to evaluate the performance of an empirical (Kowalczuket al. 2005) and two semi-analytical algorithms: Carder et al. (1999) and GSM01 (Maritorena et al. 2002) for estimating CDOMabsorption in the Baltic Sea. The data set includes coincident measurements of radiometric quantities and absorption coefficientsof CDOM made during 43 cruises between 2000 and 2008. In the first stage of the analysis, the accuracy of the empirical algorithmby Kowalczuk et al. (2005) was assessed using in situ measurements of remote sensing reflectance. Validation results improved whenmatching points located in Gulf of Gda&#x0144;sk close to the Vistula River mouth were eliminated from the data set. The calculatederrors in the estimation of <i>a</i>_CDOM(400) in the first phase of the analysis were <i>Bias</i> = -0.02, RMSE = 0.46 and R² = 0.70.In the second stage, the empirical algorithm was tested on satellite data from SeaWiFS and MODIS imagery. The satellite data werecorrected atmospherically with the MUMM algorithm designed for turbid coastal and inland waters and implemented in the SeaDASsoftware. The results of the best case scenario for estimating the CDOM absorption coefficient <i>a</i>_CDOM(400), based on SeaWiFSdata, were <i>Bias</i> = -0.02, RMSE = 0.23 and R² = 0.40. The validation of the Kowalczuk et al. (2005) empirical algorithm applied toMODIS data led to a less accurate estimate of <i>a</i>_CDOM(400): <i>Bias</i> = -0.03, RMSE = 0.19 and R² = 0.29. This assessment of the accuracy of standard semi-analytical algorithms available in the SeaWiFS and MODIS imagery processing software revealed that both algorithms (GSM_01 and Carder) underestimate CDOM absorption in the Baltic Sea with mean systematic and random errors in excess of 70%. The paper presents examples of theapplication of the Kowalczuk et al. (2005) empirical algorithm for producing maps of the seasonal distribution of <i>a</i>_CDOM(400) in the Baltic Sea between 2004 and 2008.