Application of GMDH to Water Quality Data

On the variation with time of water quality in Kiso River

Abstract

The group method of data handling (GMDH) based on the heuristic self-organization, which has been proposed by A. G. Ivakhnenko was applied to the water quality data (electrical conductance, turbidity, dissolved oxygen and chemical oxygen demand) observed at Inuyama in Kiso River located in the middle of Japan. The GMDH is an identification method, advantageous to the systems characterized by many variables and parameters, complex structure and limited data. It is a main object of this paper to analyse the variation with time of the water quality in Kiso River due to the change of flow rate and discuss the algorithm for the successful application of the GMDH to the water quality prediction or forecasting problems.
The est!mation models for each water quality index, which resemble linear multiple regression equation, as shown in eqs.(9-1)-(23-3), were found out with the polynomial algorithms of the GMDH. Comparison of the estimated values using the models with those observed are shown in Figs. 3-6 and Fig. 8.
The variational characteristics with time of the water quality are as follows electrical conductance (EC) decreases in accordance with increasing rate of flow. It's variational pattern can be described with the 1 day antecedent EC and daily mean discharge. Turbidity varies in almost the same pattern as hydrograph, with a time lag of nearly 3 hours. The variation of concentration of DO seems to be related to water temperature. As time interval for analysis or unit time of this analysis was adopted 1 day, no more useful information could be obtained. However, more results will be obtained with the analysis in a shorter unit time. The variation of concentration of COD is different according to the runoff level. When the level is low, the variation of concentration of COD is temporal. On the other hand, when the level is high (above 2, 000 m³ /s/day), the concentration of COD increases rapidly in a short time and varies gradually there after.
To make more effective use of the GMDH, it is necessary to select a unit time appropriate for the purpose and to give consideration to the physical meaning of the detected models. Finally the GMDH is recomended for the identification or the construction of prediction model of complex systems such as environmental or ecological systems.