Retrieval of water quality parameters using different channel configurations

KurzfassungRecently launched satellite instruments, such as SeaWiFS and MODIS, and instruments to be launched in the near future, such as MERIS, provide a significant increase of remote sensing data suitable for water quality monitoring. Numerous channels in visible and near infrared wavelengths provide a variety of possibilities for water quality parameter estimation, starting from simple band ratio algorithms and extending towards more complex neural network solutions and inverse algorithms. As all these instruments have individual measurement channel characteristics - number of channels, central wavelengths and widths - exactly the same detection methods cannot be used in different satellites. The different detection methods and channel configuration differences will most likely in practice lead to different estimates by different instruments even if the water and atmospheric state is constant. From the end users point of view it is important to know these differences, if different data sources are to be used. In order to understand the error characteristics and to assess the capabilities of different instruments, a comparison with ground truth data must be made. Hyperspectral data from various sources (airborne imaging spectrometer measurements, shipborne spectrometer measurements, water optical models) is converted with suitable channel response functions to different satellite instrument multispectral configurations. This data is compared with the water sample laboratory analysis using an inverse water optical model, which converts spectral measurements to water quality parameters (chlorophyll, gelbstoff and suspended sediment content, algae species etc.). The optical model parameters are adjusted by using the full hyperspectral and laboratory analysis data to minimise the inversion errors caused by the model itself. Thus the results reflect the ability of different instruments to reproduce the key features of the hyperspectral signature of water in order to adequately detect the different water quality parameters. Another source of error in the water quality measurements is the atmospheric correction of the measured channels. In lakes and coastal areas also the adjacency effect plays an important role: Some light is originally reflected from the neighbouring land area and further scattered from the atmosphere above water to the satellite sensor. The influence is large specifically in the infrared region, where vegetation is very bright and water dark. As some atmospheric correction algorithms assume water dark in this region, the phenomenon may cause cumulating errors by leading to a distracted correction of channel values. The influence of atmospherically induced errors is modelled by using 6S model twice: Reflectance on water surface is converted to a measurement by the satellite using one atmospheric state model and back to a water surface reflectance with another. This distracted measurement is then used as an input to the inverse water optical model. The results are based on measurements in Lake Constance. The cases studied can be considered as case 2 water, in which there is also other optically active components than plankton. In such waters the water optical models are relatively complex. Specially in lakes the model parameters may require region-specific optimisation, thus the presented methods can be used to create local estimates of the different instrument accuracy