干旱区生态环境知识资源中心

The Generalized Likelihood Uncertainty Estimation (GLUE) methodology is used for model calibration, testing and predictive uncertainty estimation in the application of the MIKE SHE hydrologic model for estimating monthly streamflow in a semi-arid shrubland (chaparral) catchment in central California. Monte Carlo simulation is used to randomly generate one thousand parameter sets for a 20-year calibration period encompassing variable climatic and wildfire conditions, from which behavioural (acceptable) MIKE SHE parameter sets are identified and 5% and 95% uncertainty bounds for monthly streamflow are calculated. This group of behavioural parameter sets is subsequently used to predict streamflow and to construct uncertainty bounds for a 12-year test period with climatic and fire characteristics different from those of the calibration period. More than two-thirds of the observations ill each period fell within the corresponding uncertainty bounds, suggesting a similar level of model performance in the calibration and test periods. Prediction errors (i.e. observations falling Outside the uncertainty bounds) were generally associated with large rainfall and wildfire events and are indicative of deficiencies in model structure, uncertainty in input data, and/or errors in observed streamflow. The effect of uncertainty in remote sensing-based LAI model inputs on the uncertainty associated with MIKE SHE streamflow predictions receives special attention in this work due to the fire-prone nature of the study area and the increasing use of remotely sensed LAI estimates in distributed hydrological modelling applications. Results from MIKE SHE simulations for seven LAI input scenarios (including the baseline LAI sequence used in the model calibration and testing phase of this study) indicate that differences in predictive uncertainty between scenarios are usually less than +/- 10%. This is evidence that the baseline LAI trajectory is generally appropriate for distributed hydrological modelling of chaparral catchments. (c) 2005 Elsevier B.V. All rights reserved.