HR: 09:00h
AN: H11B-05    [PDF]
TI: Comparison of Optimization Algorithms for the Automatic Calibration of a Watershed Model to Measured Flow, Sediment and Phosphorus Data
AU: * Tolson, B A
EM: bat9@cornell.edu
AF: Cornell University, School of Civil and Environmental Engineering, Hollister Hall, Ithaca, NY 14853 United States
AU: Shoemaker, C A
EM: cas12@cornell.edu
AF: Cornell University, School of Civil and Environmental Engineering, Hollister Hall, Ithaca, NY 14853 United States
AU: M‚ndez, F
EM: mendezf@unican.es
AF: Universidad de Cantabria, Dpto. de Ciencias y T‚cnicas del Agua y del Medio Ambiente, Santander, s/n 39005 Spain
AU: Regis, R
EM: rregis@orie.cornell.edu
AF: Cornell University, School of Operations Research, Ithaca, NY 14853 United States
AB: This study compares multiple heuristic optimization algorithms for automatic calibration of a watershed model to flow and water quality data. The automatic calibration of the watershed model in this study involves simultaneously calibrating the model to flow, suspended sediment, particulate and total dissolved phosphorus data by modifying up to 20 model parameters. A single-objective function is defined by a weighted sum of the measures of model and data agreement for each of the measured constituents. The heuristic optimization algorithms compared in this study are the Shuffled Complex Evolution (SCE-UA) algorithm developed by Duan et al., a Genetic Algorithm (GA) and a newly developed function approximation algorithm by Regis and Shoemaker described in another paper appearing in this session. Although the SCE and GA algorithms have been previously used to calibrate hydrologic models, their application to calibration of watershed models to measured sediment or nutrient data is much less common. This is the first application of a function approximation algorithm to the calibration of a watershed model to flow, sediment and phosphorus data. The watershed model applied in this study is a modified version of the Soil and Water Assessment Tool (SWAT2000). The case study area is a small (37 km$^{2}$), mainly rural watershed in Upstate New York called Town Brook that drains to New York City's (NYC) drinking water supply system. Town Brook is a sub-watershed in NYC's Cannonsville drinking water reservoir watershed (1178 km$^{2}$). Mitigating phosphorus loading to this and other NYC drinking water reservoirs is critical to avoid construction of an estimated 8 billion dollar water filtration plant. A SWAT2000 model was previously developed by the authors to model phosphorus loading from the entire watershed to the Cannonsville Reservoir. Initial SWAT2000 model parameters for the Town Brook scale model were derived from the larger scale Cannonsville Basin model. This study is focused on automatically calibrating the Town Brook model to approximately two years of Town Brook water quality data (suspended sediment loading and phosphorus) with the various heuristic optimization algorithms. The best automatic calibration results are compared with the initial trial-and-error calibration results to gauge the improvement in model predictive performance achieved by automatic calibration.
DE: 1803 Anthropogenic effects
DE: 1860 Runoff and streamflow
DE: 1871 Surface water quality
DE: 1899 General or miscellaneous
SC: Hydrology [H]
MN: 2003 Fall Meeting