Integration of neural networks in a geographical information system for the monitoring of a catchment area

This paper presents a study done on the catchment area of the Têt, main river of the PyrénéesOrientales (P-O, south of France). As much of other Mediterranean coastal rivers, the hydrological system of the Têt river presents lowest water level long periods, intercepted by violent and devastating short rising period, essentially due to rain. This contrasted system, is precisely one of the main reason for which the researchers are interested in the coastal rivers like the Têt. Most of the pollution contributions to the sea is carried out during these flash-risings which are difficult to follow with both tradionnal sampling strategies and reliable mathematical models. So, this article presents in a first part, the implementation and the results of a rain flow model development, allowing to know the flow state at a precise point of the river. In a second part, a qualitative prediction is implemented in the WasteWater Treatment Plant of Perpignan (main city of the P-O) allowing to know the water biological state. Parameters used for the prediction are easily online measurable and allow the COD prediction at the WWTP exit of Perpignan, which is a parameter not easily online measurable. Both of these models are implemented with Recurrent Neural Network tools. The last part of this article shows the integration of these models in a GIS, allowing a real time visualization and supervision of the bassin.