Multisite- Multivariate Calibration of Watershed Models · 2009-08-17 · Watershed Models • Computer-based hydrologic models have become popular for – performing hydrologic forecasts

Multisite- Multivariate Calibration of Watershed Models

Mahdi AhmadiMazdak Arabi

August 06, 20095th International SWAT Conference

Objective• Evaluate the

“Efficiency” and “Applicability”of

“Optimization Techniques”for

“Parameter Estimation”in

“Multiple Locations” for ”Multiple Variables“

of a Comprehensive Watershed model (i.e., Soil and Water Assessment Tool; SWAT) in a watersheds in Indiana

Watershed Models• Computer-based hydrologic models have become

popular for– performing hydrologic forecasts– managing watersheds

• Models should be calibrated before being applied in decision-making process

Model Calibration• To specify values for model "parameters" in such a

way that the model's behavior (Simulation Results) closely matches that of the real system it represents (Historical Data)

• Most of the parameters are conceptual representations of abstract watershed characteristicsand should be determined through a trial-and-errorprocess not by direct measurements

Model Calibration• Manual vs. Automatic Calibration

– Manual• Tedious and time consuming• Subject to experience of the user

– Automatic(use of an optimization algorithm to determine best-fit parameters)

• Faster and less experience• Less subjective

Auto-calibration• Optimization

– Single objectiveor/ to lump all different objectives into one• Shuffled Complex Evolution (SCE-UA)• Single-objective Genetic Algorithm (GA)

– Multi-objective• Multi-objective GA (MOGA)

to choose the best trade-offs among all the defined and conflicting objectives

• Water resources systems analysis is often confronted with multiple conflicting decision objectives that should be optimized

… outline• Auto-calibration Methods• Auto-calibration Tool• Case Study• Results/ Discussion• Conclusion

Auto-calibration• Shuffled Complex Evolution (SCE, Duan 1992)

– Combination of random and deterministic approaches– Includes concept of clustering

• Sampling using Random or/ Prior Distribution• Partitions results in several “Complexes”• Evolving sequences

– Used in the auto-calibration of ArcSWAT

Auto-calibration• Multi-Objective GA (MOGA: NSGA-II, Deb 2001)

– Ranking of individuals– New individuals

• Selection• Mating• Crossover• Mutation

Auto-calibration Tool

Auto-calibration Tool

108 parameters included

Auto-calibration Tool

(MO)GA Settings

SCE settings

Case StudyWildcat Creek, Kokomo Region, Indiana (588 km2)

Reservoir

Case Study• Objective functions

• Group 1– outlet # 1: RMSE of Total pesticide (Tpest)– outlet # 4: RMSE of Streamflow (SF) in

• Group 2– outlet # 1: (1-NS) of Tpest– outlet # 4: (1-NS) of SF

(18 parameters of Streamflow, Reservoir, and pesticide included)

• Group 3– outlet # 4: (1-NS) of SF– outlet #11: (1-NS) of SF– outlet # 1: (1-NS) of Tpest– outlet # 1: (1-NS) of Total Nitrogen (TN)

(29 parameters of Streamflow, Reservoir, Pesticide, and Nitrogen included)

Case Study

Local Sensitivity Analysis Global Sensitivity Analysis

0

2

4

6

8

10

0 20 40 60 80 100 120 140 160 180 200

Screening

Prioritize Parameters

5 Streamflow

2 Pesticide

11 reservoir

11 Nitrogen

Group 1 $ 2

Group 3

• Parameters Selection

Results• Group 1 (2000 Runs- 7 years)

Corresponding NS= 0.67

Corresponding NS= 0.39

Results• Group 2 (3500 Runs- 7 years)

Results• Group 3 (4500 Runs- 7 years)

Conclusion• Flow processes and pesticide can be reasonably

modeled, however representation of pesticide processes is more challenging

• Examined auto-calibration algorithms:– provide a systematic approach for parameter estimation– reduce the subjectivity and time requirements of the

manual calibration exercises

• Based on the same number of runs, Multi-Objective Optimization method gave better results

Conclusion• A wider range of alternatives is usually identified

when a multi-objective methodology is employed

• More realistic results are available when many objectives are considered

• Multiple objectives promotes more appropriate roles for the participants in the planning (analyst) and decision-making processes

Thank you!

Mahdi Ahmadi Mazdak Arabi

[email protected] [email protected]