Bathymetric Mapping for Shallow Water Using Landsat 8 via Artificial Neural Network Technique

Bathymetric measurement using remote sensing can be replaced by the conventional technique which reduces the cost and labor required for bathymetric measurement. It can also overcome the complications in provision required during spatial and temporal depth estimates. But to obtain bathymetric information from multispectral satellite imagery using remote sensing techniques requires many corrections such as atmospheric, bottom albedo water and bottom reflectance, attenuation coefficient, and concentration of suspended solid constituents such as organic and inorganic, etc. Sometimes it is practically impossible to apply such correction to imagery because of non availability of field data. Therefore, there is a need to have faster and practical approach to hand the complex relationship between satellite reflectance and water depth for finding bathymetric measurement. The methodology based on Artificial Neural Network (ANN) is very simple to derive bathymetric maps in shallow water via reflectance’s value of imagery and sample depth measurement. ANN techniques used simple approach for deriving of the depth estimation possibility, without refining image reflectance value in the depth causing scattering from environmental parameters such as type of vegetation and bottom material available. In this paper, the best linear or non-linear mathematical models to be fitted for bathymetric application are applied on Upper Lake Bhopal using Landsat 8. For this, the best fitting curve, linear with single and multi band, polynomial fit for first, second and higher degree, ratio and exponential-based algorithm, and ANN models were tried. The main conclusions are ANN model have produced lowest pass, chi-square test, and RMSE value as compared to other models.