AbstractsThe subject of this article is runoff losses in an urban environment, specifically initial losses through impermeable surface depressions directly connected to the network.For this purpose the hydrological behavior of two urban watersheds (Batany and Trianon) of about 5 hectares each, in the Bordeaux region of France, have been studied to observe that the fluctuations around the "Rainfall Amount versus Runoff Volume" law essentially derive from initial runoff losses which differ from one rainfall event to another. The fluctuations around this law make it impossible to precisely estimate runoff volume based on rainfall. Improved knowledge of initial losses would result in better estimation of the runoff volume of more regular (monthly or bimonthly) rainfall events, which must increasingly be treated at water treatment plants in order to be able to better control the overflows. If the involvement of permeable surfaces is assumed to be negligible, we can postulate that the initial losses from a given rainfall event are directly linked to the water stored on the impervious surfaces connected to the network. The moisture content prior to the event is dependent upon the occurrence of a previous rainfall event and on the meteorological conditions prevailing between the previous rainfall event and the one under study. A model, called the EVA model, has been developed with the objective of predicting runoff losses corresponding to a rainfall event as a function of the previously prevailing watershed moisture conditions. The model evaluates the amount of evaporation from the water contained in the surface depressions between two successive rainfall events, called the initial rainfall event and the final rainfall event. The initial rainfall event represents the previous rainfall event, and the final rainfall event is the event for which the losses are to be estimated. The application of the model requires a very good record of the small rainfall events which have occurred during the modelling periods, and which are called the intermediate rainfall events. In practice, this constraint implies the need to monitor the dispersed rainfall events which, even if they cause only very light runoff, nonetheless contribute to a partial filling of the surface depressions present on the impervious terrain. The equations used in the model are those which correspond to the energy balance between air and water using the net radiation, the latent, sensible and storage heat (the soil heat flux is considered negligible). The EVA model uses meteorological data such as the air temperature, relative humidity, wind speed and solar radiation. The model evaluates two variables: water depth and water temperature. Since water depth after evaporation is known, the losses of a rainfall event can be estimated by subtracting the total volume of water which has evaporated during the dry period from a maximum value of the losses. The modeled losses are then compared with the measured losses. In order to simplify the resolution of the problem, the total water volume contained in the thousands of surface depressions present in the watershed is considered to be contained in a single depression called a representative depression. This representative depression can take different forms and have different initial heights, which have been tested while the work was in progress. The model is found to be coherent in terms of the variations in water depth in the surface depressions. The total water volume contained in the surface depressions takes from 1 to 3 days to evaporate depending on the season. The variation in water depth is caused by differences in evaporation rates occurring during the months close to the summer solstice and during the cooler months. The first version of the model was created in 1992 and tested on two watersheds of about 250 hectares each in the Paris area. The model was modified and the new results were compared with the measurements obtained in two watersheds in the city of Bordeaux. The performance of the model was evaluated for 17 rainfall events, of which 10 were in the Batany watershed and 7 in the Trianon basin. The model accurately predicts the losses corresponding to a rainfall event within 0.5 mm, in two cases out of three. The problem encountered in the remaining onethird of the cases was essentially that it is difficult to account for runoff during intermediate rainfall events because of very low flow rates and small rainfall depth measurements. Experimental equipment installed in two watersheds in Bordeaux has made it possible to obtain relatively precise pluviometric and discharge measurements. However, there is uncertainty concerning these measurements, which is inherent in this field, when it comes to validating a model like EVA because such a model is used for regular rainfall events for which the initial losses directly influence the runoff volume. However the a priori knowledge of initial runoff losses should enable better use of a model which, for example, assumes that the runoff coefficient increases progressively at the beginning of the rainfall event. The validation of such a model was attempted while the work was under way, but ran into the difficulty of selecting a set of rainfall events characterized by constant rainfall intensity. The development of a model like EVA requires rainfall and flow measurements which are free of uncertainties and accessible within experimental watersheds which are perfectly monitored and where the measurement uncertainties are the same for all observable measurement ranges (in particular for the discharge measurements). These requirements currently constitute a technical barrier in terms of measurement that will be difficult to surmount. However, the work carried out in this research hints at the possible improvement of the classical hydrological models used in urban hydrology, particularly those used for forecasting the runoff volumes of regular rainfall events. KeywordsInitial losses, depression storage, impervious surfaces, evaporation, energy balance, water depth. Corresponding authorD. Brûlé, École Polytechnique, Département de génie civil, CP 6079, Succursale Centreville, Montréal (Québec), H3C 3A7 CANADA  
