Article révisé par les pairs
Résumé : The implementation of biological filtration on granular activated carbon under various operating conditions has revealed that biodegradable organic matter (BOM) removal performance is not easily predicted. In this context, the statistical analysis of the differences between experimental data and data calculated by a deterministic model may provide insight into the source of prediction errors. The CHABROL model, developed by Billen et al., (1992), relates BOM consumption to biomass densities in order to predict BOM removal profiles in biological activated carbon filters (BAC) and rapid sand filters. This work presents the results of testing the CHABROL model using a large database from pilot and full-scale filters located in two Canadian cities: Laval and Montreal. This database includes data from two different water sources and three biological filtration configurations (direct filtration, first-stage dual-medium sand-BAC filters and second-stage mono-medium BAC filters). Since nearly half of all experimental data were obtained at very low temperatures (≤1°C), the impact of prolonged acclimation to very cold temperatures was investigated to ensure accurate prediction by the CHABROL model. Experimental results have shown that modifications in the CHABROL model have to be made to account for the acclimation to very cold temperature and to prevent under estimations of the BDOC removals. The results of the model testing show overall satisfactory results in the ability to predict BDOC removals in various types of filters, except for the sampling campaigns completed just prior to backwash. This comparison shows the interest of using a predictive model to predict performance in dynamically operated filters which may be used for design and operation.