Quels apports hydrologiques pour les modèles hydrauliques? Vers ...

algorithm using the Nash and Sutcliffe (Nash and Sutcliffe, 1970) efficiencycriterion (NSE) calculated at the Tahlequah station as a single objectivefunction. The models were tested following a split sample test (Klemes, 1986)using the two sub-periods defined in the DMIP protocol (Smith et al., 2009)360including 1 year of warm-up (sub-period P1 started on 1995-10-01 and endedon 2002-09-30, sub-period P2 started on 2001-10-01 and ended on 2006-09-30).The models were calibrated on the first period and validated on the secondone, and then the role of the two periods was reversed to evaluate the modelsin validation mode on all data at hand. The mean statistics obtained over the365two validation and calibration tests, were considered for performance analysis.tel-00392240, version 1 - 5 Jun 2009370For each period, NSE as well as statistics on 15 flood events (see Section 3)were calculated on three gauging stations: at the downstream end of the reach(Tahlequah), on an interior point located at the middle of the reach (SiloamSprings) and on one tributary (Kansas).Event-based statistics were calculated according to Smith et al. (2004):Absolute peak error (%): E P =Absolute peak time error (h): E T =∑ Nek=1 |Q P k − Q P Sk |× 100 (3)N e × Q P AV G∑ Nek=1 |T P k − T P Sk |(4)N e375where N e is the number of events (15 here), Q P k is the observed peak dischargeof the k th event (m 3 /s), Q P Sk is the simulated peak, Q P AV G is the averagedobserved peak discharge over the N e events, T P k is the observed time to thek th peak, T P Sk is the simulated time to the k th peak.Note that the models were calibrated only on downstream flows. The two otherstations are considered ungauged for calibration, with blind simulations, so18