A.G. Romanova - Engineer, Project department of construction company «Evrodorstroy». E-mail: romalenka@mail.ru
D.A. Tarkhov - Dr.Sc. (Eng.), Professor, Department of Mathematics, St.-Petersburg State Polytechnic University. E-mail: dtarkhov@gmail.com
T.A. Shemyakina - Associate Professor, Department of Mathematics, St.-Petersburg State Polytechnic University. E-mail: st_tat@mail.ru

The article substantiates the need for hierarchical systems of environmental monitoring and forecasting, and also proposed neural network approach to their construction. The main advantage of this approach is the ability to efficiently utilize the processing power, working with neural network models instead of large amounts of data. This article presents two practical examples of the application of neural network approach. The first example is connected with the restoration of the initial distribution of contamination on the final distribution, the second - to identify the source of contamination at the border for measurements inside the area. These three problems are incorrect and complex, it is difficult to solve them using traditional methods. The third task shows the advantage of sending neural network model changing of wind speed over the traditional interval information in the spread of clouds of pollution.