Needs and influence of calibration and validation data sets in stormwater quality models of various levels of complexity

摘要

Stormwater quality modelling has become a fundamental issue in urban hydrology. Practitioners are led to work with Stormwater Quality Models (SQMs), the operational use of which remains a difficult task despite the progress achieved during the last two decades. A method to assess the variability in calibration and validation results of various SQMs with different levels of complexity due to the variability in calibration data sets is proposed. Preliminary results for site mean concentration (SMC) and Event Mean Concentration (EMC) models used to evaluate SS and COD concentrations in "Le Marais" catchment show that a minimum of 15 to 20 rainfall events is needed to calibrate them in order to get acceptable results compared to what is achievable by using more rainfall events. The size of validation data sets appears less critical for EMC models: if 20 to 40 events are used for validation, the validation quality remains rather constant for SS models and showed little variations for COD models. These conclusions strongly depend on the total available data sets and on the fraction used for calibration.