Laserfiche WebLink
<br />.. ~ , '" <br />OG13J~ <br /> <br />33 <br /> <br />These findings indicate that. on the average, the conceptual models <br />will better predict the actual runoff response. The conceptual models' <br />mean overprediction of the actual peak discharge is 4% with 90% of the <br />prediction being within ~30% of this value. The physically-based models' <br />mean overprediction of actual peak discharge is 10% with 90% of the <br />predictions being within 35% of this value. <br />3) ConsistenCy test - This test evaluated the ability of <br />different modelers to get similar predictions with the same rainfall- <br />runoff model. The writers use the published results of several modelers <br />who were using the same rainfall-runoff model to predict the runoff <br />response of the same storm event in the same basin. Table 1I-6 lists the <br /> <br />results. <br /> <br />Any conclusions drawn from this test are questionable due to the <br /> <br /> <br />relatively few published results available for the test. Notwithstanding <br /> <br /> <br />these limitations, Table 1I-6 indicates an understandable trend. The <br /> <br /> <br />results from the conceptual models are more consistent (have a smaller <br /> <br /> <br />spread) than the results from the physically-based models. This trend <br /> <br />is not surprising when one considers the calibration process. During <br /> <br /> <br />calibration of a model, various calibration parameters are adjusted to <br /> <br />shift the model-generated runoff response into closer agreement with the <br />actual runoff response. Thus, prediction of future runoff responses <br /> <br />with this "calibrated" model becomes a function of the parameter ad- <br /> <br /> <br />justments. If the runoff response is very sensitive to a certain <br /> <br />calibration parameter, a slight difference in calibration will result in <br />a significant change in predicted runoff response. With this in mind, <br />the consistency trend found in Table 1I-6 is expected. It is more <br />likely that several modelers will assign a similar value to the few <br />