HR: 1340h
AN: H43A-0488    [Abstracts]
TI: Multi-Objective Calibration and Sensitivity Analysis of a Distributed Model using Similarity Measures
AU: * Bastidas, L A
EM: Luis.Bastidas@usu.edu
AF: Utah Water Research Laboratory, Utah State University, 8200 Old Main Hill, Logan, UT 84322-8200 United States
AU: Li, S
EM: shujun.li@usu.edu
AF: Utah Water Research Laboratory, Utah State University, 8200 Old Main Hill, Logan, UT 84322-8200 United States
AB: Better spatial pattern representation of hydrological variables is one of the important targets of distributed hydrologic and hydrometeorological models at basin scale. In the present work, we use a modified version of the set theory-based similarity measure - the Hausdorff Norm (HN), for a quantitative evaluation of distributed fields of hydro-meteorological variables from a land surface model against observations, for a global sensitivity analysis of the model parameters and for a multi-objective distributed model calibration. We show that the modified version of the HN overcomes its traditional limitations. The procedure is used for the evaluation of individual snapshots, for the evaluation of several snapshots fields simultaneously, and for the evaluation of a temporal sequence of multiple field snapshots. We applied the methodology to the NOAH-LSM over a semi-arid watershed, the San Pedro River basin in Arizona. The NOAH model was run using a 4 km resolution grid defined by the degree of land surface and sub-surface characteristics of this watershed (around 3000 grid cells). The model was driven in offline fashion with hourly forcing data from the NLDAS. Surface characteristics like vegetation, soil and topography were remapped from a fine resolution grid (1km). Using the Multi-Objective Generalized Sensitivity Analysis Algorithm (MOGSA) it was determined that around 20,000 distributed model runs are required to evaluate the parameter sensitivity (i.e., approximately 100,000 point function evaluations) and the number of model parameters was reduced. That information was used for the multi-objective parameter estimation of the distributed model. These computationally intensive and comprehensive tests have shown the extreme versatility and power of the HN and the multi-objective framework as a tool for evaluation, parameter identification, and comparison of distributed model performances at different spatial and temporal scales. The HN provides with a quantitative measure of the model performance and was successfully applied to the parameter identification of a large distributed model.
DE: 1833 Hydroclimatology
DE: 1843 Land/atmosphere interactions (1218, 1631, 3322)
DE: 1846 Model calibration (3333)
SC: Hydrology [H]
MN: Fall Meeting 2005