Application of an empirical neural network to surface water quality estimation in the Gulf of Finland using combined optical data and microwave data

Since neural networks have been widely applied to the nonlinear transfer function approximation, we present an empirical neural network algorithm to estimate major parameters in surface waters from combined optical data and microwave data in the Gulf of Finland. Concurrent in situ surface water quality measurements, optical (Landsat TM) data and microwave (ERS-2 SAR) data were obtained in selected locations in August 1997. The TM and ERS-2 SAR data from locations of water samples were extracted and digital data were examined in numerous transformations. Although significant correlations were observed between digital data and chlorophyll-a (Chl-a), suspended sediment concentration (SSC), turbidity (Turb), and Secchi disk depth (SDD), application of neural networks appears to yield a superior performance in modeling transfer functions in this study area. Here, an empirical neural network algorithm is applied to estimate the transfer functions between the major characteristics of surface waters and the satellite optical and microwave data. The results show that the estimation accuracy for major characteristics of surface waters using the neural network is much better than those from regression analysis. The results also indicate that microwave data can assist to improve the estimation of these characteristics. Therefore, it may be possible to develop surface water quality algorithms in which microwave data are used as supplementary data to optical observations. However, this improvement of optical data retrieval algorithm is limited in this case study. The technique still needs to be refined in detail in order to detect differences within the typical range of these water quality parameters found in the area under study.