Abstract
Today, there is an increased use of artificial intelligence systems to solve the latest mixed science and technology problems. And Earth remote sensing (ERS) is one of these problems. As it applies to ERS problems, neural network approaches can be conventionally classified as algo- rithms indirectly using physical and non-physical quantities, i.e. measure values from satellite tools. In other terms, a function linking input quantity and output (such as the probability of a particular class in a classification problem) is approximated by the algorithms spec ified, unlike other approaches where this function is usually explicitly asserted. The widespread use of classification algorithms is the case of an explicit physical approach based on spectral analysis and threshold methods when physical quantities (e.g., spectral brightness ratio, luminous temperature, etc.) or spectral indices are evaluated by means of thresholds specified for each class. This approach requires complex dependencies to get more exact results in an unknown analytical form. However, using neural networks such difficulties can be overcome [1]. The article is aimed to highlight the primary results to practice the neural network in bathymetric and hydrographic parameters of water bodies identification.