"Unfortunately, such misuse of the NN technique is common –. In addition to –, another paper published in the subject of remote sensing has the same problem of mathematical indeterminacy due to too many hidden neurons used compared to the limited amounts of data available . Mutanga and Skidmore  used feedforward ANNs to map grass quality in the Kruger National Park, South Africa. "
[Show abstract][Hide abstract] ABSTRACT: A paper by Zhang , using a feedforward artificial neural network (ANN) for water quality retrievals from combined Thematic Mapper data and synthetic aperture radar data in the Gulf of Finland, has been published in this journal. This correspondence attempts to discuss and comment on the paper by Zhang The amount of data used in the paper by Zhang is not enough to determine the number of fitting parameters in the networks. Therefore, the models are not mathematically sound or justified. The conclusion is that ANN modeling should be used with care and enough data
IEEE Transactions on Geoscience and Remote Sensing 07/2007; 45(6-45):1896 - 1897. DOI:10.1109/TGRS.2007.895432 · 3.51 Impact Factor
[Show abstract][Hide abstract] ABSTRACT: This thesis deals with surface water quality estimation using remote sensing in the Gulf of Finland and the Archipelago Sea. Satellite remote sensing of water and empirical algorithms for surface water quality variables in coastal waters in the Gulf of Finland and the Archipelago Sea are explained and results from the studies in the area are presented. Concurrent in situ surface water measurements, AISA data, Landsat TM data, ERS-2 SAR data, AVHRR and MODIS data were obtained for selected locations in the Gulf of Finland and the Archipelago Sea in August 1997 and from April to May 2000, respectively. The AISA, TM, SAR, AVHRR and MODIS data from locations of water samples were extracted and digital data were examined. Significant correlations were observed between digital data and surface water quality variables. Semi-empirical, simple and multivariate regression analyses, and neural network algorithms were developed and applied in the study area. Application of neural networks appears to yield a superior performance in modelling radiative transfer functions describing the relation between satellite observations and surface water characteristics. The results show that the estimated accuracy for major characteristics of surface waters using the neural network method is much better than retrieval by using regression analysis. Since radar observations of water are strongly affected by surface geometry but not by water quality, radar data should be useful to eliminate the effects of surface roughness from the results when combined with optical observations. However, our results suggest that microwave data improve estimation of water quality very little or not at all. The technique, however, should be examined with new data sets obtained under various weather and water quality conditions in order to estimate its feasibility for estimating surface water quality parameters in the Finnish coastal waters. Report / Helsinki University of Technology, Laboratory of Space Technology, ISSN 0786-8154; 55
[Show abstract][Hide abstract] ABSTRACT: This paper presents the applicability of combined Landsat Thematic Mapper and European Remote Sensing 2 synthetic aperture radar (SAR) data to turbidity, Secchi disk depth, and suspended sediment concentration retrievals in the Gulf of Finland. The results show that the estimated accuracy of these water quality variables using a neural network is much higher than the accuracy using simple and multivariate regression approaches. The results also demonstrate that SAR is only a marginally helpful to improve the estimation of these three variables for the practical use in the study area. However, the method still needs to be refined in the area under study.
IEEE Transactions on Geoscience and Remote Sensing 04/2003; 41(3-41):622 - 629. DOI:10.1109/TGRS.2003.808906 · 3.51 Impact Factor
Data provided are for informational purposes only. Although carefully collected, accuracy cannot be guaranteed. The impact factor represents a rough estimation of the journal's impact factor and does not reflect the actual current impact factor. Publisher conditions are provided by RoMEO. Differing provisions from the publisher's actual policy or licence agreement may be applicable.