Michael Schaale

Atmospheric correction algorithm for MERIS above case-2 waters

The development and validation of an atmospheric correction algorithm designed for the Medium Resolution Imaging Spectrometer (MERIS) with special emphasis on case‐2 waters is described. The algorithm is based on inverse modelling of radiative transfer (RT) calculations using artificial neural network (ANN) techniques. The presented correction scheme is implemented as a direct inversion of spectral top‐of‐atmosphere (TOA) radiances into spectral remote sensing reflectances at the bottom‐of‐atmosphere (BOA), with additional output of the aerosol optical thickness (AOT) at four wavelengths for validation purposes. The inversion algorithm was applied to 13 MERIS Level1b data tracks of 2002–2003, covering the optically complex waters of the North and Baltic Sea region. A validation of the retrieved AOTs was performed with coincident in situ automatic sun–sky scanning radiometer measurements of the Aerosol Robotic Network (AERONET) from Helgoland Island located in the German Bight. The accuracy of the derived reflectances was validated with concurrent ship‐borne reflectance measurements of the SIMBADA hand‐held field radiometer. Compared to the MERIS Level2 standard reflectance product generated by the processor versions 3.55, 4.06 and 6.3, the results of the proposed algorithm show a significant improvement in accuracy, especially in the blue part of the spectrum, where the MERIS Level2 reflectances result in errors up to 122% compared to only 19% with the proposed algorithm. The overall mean errors within the spectral range of 412.5–708.75 nm are calculated to be 46.2% and 18.9% for the MERIS Level2 product and the presented algorithm, respectively.

Retrieval of atmospheric and oceanic properties from MERIS measurements: A new Case-2 water processor for BEAM

A freely available data processor for the B asic E RS & ENVISAT ( A )ATSR and M ERIS Toolbox (BEAM) was developed to retrieve atmospheric and oceanic properties above and of Case‐2 waters from Medium Resolution Imaging Spectrometer (MERIS) Level1b data. The processor was especially designed for European coastal waters and uses MERIS Level1b Top‐Of‐Atmosphere (TOA) radiances to retrieve atmospherically corrected remote sensing reflectances at the Bottom‐Of‐Atmosphere (BOA), spectral aerosol optical thicknesses (AOT) and the concentration of three water constituents, namely chlorophyll‐a (CHL), total suspended matter (TSM) and the absorption of yellow substance at 443 nm (YEL). The retrieval is based on four separate artificial neural networks which were trained on the basis of the results of extensive radiative transfer (RT) simulations by taking various atmospheric and oceanic conditions into account. The accuracy of the retrievals was acquired by comparisons with concurrent in situ ground measurements and was published in full detail elsewhere. For the remote sensing reflectance product a mean absolute percentage error (MAPE) of 18% was derived within the spectral range 412.5–708.75 nm while the accuracy of the AOT at 550 nm in terms of MAPE was calculated to be 40%. The accuracies for CHL, TSM and YEL were derived from match‐up analysis with MAPEs of 50%, 60% and 71%, respectively.

Application of the active learning method to the retrieval of pigment from spectral remote sensing reflectance data

Due to the noise that is present in remote sensing data, a robust method to retrieve information is needed. In this study, the active learning method (ALM) is applied to spectral remote sensing reflectance data to retrieve in‐water pigment. The heart of the ALM is a fuzzy interpolation method that is called the ink drop spread (IDS). Three datasets (SeaBAM, synthetic and NOMAD) are used for the evaluation of the selected ALM approach. Comparison of the ALM with the ocean colour 4 (OC4) algorithm and the artificial neural network (ANN) algorithm demonstrated the robustness of the ALM approach in retrieval of in‐water constituents from remote sensing reflectance data. In addition, the ALM identified and ranked the most relevant wavelengths for chlorophyll and pigment retrieval.

Artificial-neural-network-based atmospheric correction algorithm: application to MERIS data

After the successful launch of the Medium Resolution Imaging Spectrometer (MERIS) on board of the European Space Agency (ESA) Environmental Satellite (ENVISAT) on March 1st 2002, first MERIS data are available for validation purposes. The primary goal of the MERIS mission is to measure the color of the sea with respect to oceanic biology and marine water quality. We present an atmospheric correction algorithm for case-I waters based on the inverse modeling of radiative transfer calculations by artificial neural networks. The proposed correction scheme accounts for multiple scattering and high concentrations of absorbing aerosols (e.g. desert dust). Above case-I waters, the measured near infrared path radiance at Top-Of-Atmosphere (TOA) is assumed to originate from atmospheric processes only and is used to determine the aerosol properties with the help of an additional classification test in the visible spectral region. A synthetic data set is generated from radiative transfer simulations and is subsequently used to train different Multi-Layer-Perceptrons (MLP). The atmospheric correction scheme consists of two steps. First a set of MLPs is used to derive the aerosol optical thickness (AOT) and the aerosol type for each pixel. Second these quantities are fed into a further MLP trained with simulated data for various chlorophyll concentrations to perform the radiative transfer inversion and to obtain the water-leaving radiance. In this work we apply the inversion algorithm to a MERIS Level 1b data track covering the Indian Ocean along the west coast of Madagascar.

Application of the Active Learning Method for the estimation of geophysical variables in the Caspian Sea from satellite ocean colour observations

Remotely sensed data inherently contain noise. The development of inverse modelling methods with a low sensitivity to noise is in demand for the estimation of geophysical variables from remotely sensed data. The Active Learning Method (ALM) is well known to have a low sensitivity to noise. For the first time, ALM was utilized for the inversion of radiative transfer calculations with the aim of estimating chlorophyll a (Chl a), coloured dissolved organic matter (CDOM), and suspended particulate matter (SPM) in the Caspian Sea using MERIS (MEdium Resolution Imaging Spectrometer) data. ALM training is straightforward and fast. The ALM inversion models revealed the most relevant variables and showed a short processing time in operational applications for the estimation of geophysical variables. The mean absolute percentage errors of Chl a, SPM, and CDOM estimation using ALM inversion models were 44, 70, and 73%, respectively. According to the ALM results, it can be introduced as a new method for inverse modelling of ocean colour observations.

Optical closure of parameterized bio-optical relationships*

An optical closure study on bio-optical relationships was carried out using radiative transfer model matrix operator method developed by Freie Universität Berlin. As a case study, the optical closure of bio-optical relationships empirically parameterized with in situ data for the East China Sea was examined. Remote-sensing reflectance (Rrs) was computed from the inherent optical properties predicted by these biooptical relationships and compared with published in situ data. It was found that the simulated Rrs was overestimated for turbid water. To achieve optical closure, bio-optical relationships for absorption and scattering coefficients for suspended particulate matter were adjusted. Furthermore, the results show that the Fournier and Forand phase functions obtained from the adjusted relationships perform better than the Petzold phase function. Therefore, before bio-optical relationships are used for a local sea area, the optical closure should be examined.

An extended validation test for data input into parameterized retrieval algorithms

The retrieval of environmental data from multi-spectral remotely sensed data is very often based on the (partial) inversion of extensive radiative transfer simulations (RTS). The inversion can be utilized in different ways, e.g. through the usage of polynomials or artificial neural networks. The inversion algorithms (IA) usually contain numerous parameters, which have to be adapted by regression schemes in a training phase with the help of the RTS data. The subsequent processing of real remotely sensed data by an adapted IA requires a validity test (VT) of the input data (usually a vector consisting of TOA radiances, environmental and geometric data) before inputting them into the IA. This test ensures that these or similar data were included in the training phase of the IA and thus helps to avoid unpredictable extrapolation effects. In standard procedures these “out-of-scope” data are identified by a simple convexity test (CT). CT means that each element of the input vector is tested to lie between the m...

Optimization of Water Sampling Locations Using Remote Sensing Data Analyzed by Neural Networks

Abstract In water sampling it is very common to use human experience to determine sampling locations. We present results from a neural network analysis of multispectral imaging data from the Compact Airborne Spectrographic Imager (casi) to determine significant water sampling locations. In this study Lake Tegel in Berlin, Germany, was overflown on different days. The analysis of the remote sensing data results in a clustering of the overflown water body for each pass. The neural network clusters found for each pass have been related to each other. This procedure enables us to optimize the number and location of water sampling stations.