ESTIMATING COD LOADS IN COMBINED SEWER OVERFLOWS WITH MULTIVARIATE AND NEURAL NETWORK MODELS UNDER SEMI-ARID RAINFALL CONDITIONS

摘要

Estimation of pollution loads from combined sewer overflows (CSO) is a major issue for mitigation of impacts on the water environment. According to EU directives, pollutant loads must be estimated in a frequency-magnitude analysis to better assess their impact on the water bodies. This study focuses on estimating the chemical oxygen demand (COD) load in CSO from one of the main sewer trunks in Valencia (Spain) to assess impacts on the waterfront. 42 events were recorded, modelled and analysed during the period 2008-2012 (quantity and quality data). For each event, antecedent dry period (T), rainfall duration (D), peak rainfall intensity (I), rainfall volume (R), runoff volume (V) and COD load (M) spilled into the receiving water body were obtained. T is related to pollutant accumulation in the catchment (build-up), R and V to the event magnitude and I to erosive processes (wash-off). In this paper, two different models are analysed to estimate M. First, an analytical multivariate regressive model is adjusted considering relevant explanatory variables. On the same basis, an artificial pruned neural network (NN) was trained to estimate M, depending on input variables with a hidden layer. Both models highlight the same counterintuitive result in the studied case: M does not depend on T. The multivariate model best fit shows a quite linear relationship between R (or V) and the COD loads. This strong dependence between R and M is also deduced from the NN model, which eliminates the T, D and I inputs, and only considers R to estimate the COD load (M) with a 10% relative mean squared error on test data. Semi-arid conditions of the Valencia rainfall regime lead to very large antecedent dry periods. Accumulated pollutants in the catchment have reached their maximum rates and are not already influenced by T. Consequently, the higher rainfall or runoff volumes are, the higher pollutant loads because of the huge amount of pollutants accumulated in the system and mobilised during each event.