SECOND MOPEX WORKSHOP, 8-10 April 2002, Tucson, Arizona

Supported by the Office of Hydrology, National Weather Service, NOAA and SAHRA, Department of Hydrology and Water Resources, University of Arizona

Model Parameter Estimation Experiment (MOPEX) is funded by NOAA Office of Global Programs to investigate techniques for the a priori parameter estimation of hydrological models and land surface parameterization schemes used in atmospheric models. A major step is to assemble the necessary hydrometeorological data and the associated land surface characteristics data. A comprehensive database has been developed by the MOPEX project that contains historical hydrometeorological time series data and land surface characteristics data for many basins in the United States and in other countries. A number of international MOPEX workshops have been convened or are planned for MOPEX. The Second International MOPEX Workshop was held in Tucson, Arizona, April 8-10, 2002. This workshop was designed to bring together interested US and international hydrologists and land surface modellers to exchange experience in developing techniques for a priori estimation and calibration of hydrological model parameters.

WORKSHOP OBJECTIVES

The specific objectives of the Workshop were to seek answers to the following questions:
(1) How do we define the relationships between model parameters and basin characteristics?
(2) How can model calibration be used to refine the a priori parameters?
(3) How do we evaluate the uncertainty due to model structure, calibration data and model parameters?

LIST OF PRESENTATIONS

Qingyun Duan, National Weather Service, Hydrology Laboratory - Outline of MOPEX and data used in Workshop
Florence Habet, Meteo France - Description of model and data used
Jeff McDonnell, Oregon State Univeristy - Isotope response to observe catchment behaviour
George Leavesley, USGS - MMS application to Workshop data
Gopi Goteti, Princeton University - VIC model application to Workshop data
Tian Yew Gan, University of Alberta, Canada - Sacramento model application to Workshop data
Qingyun Duan, National Weather Service, Hydrology Laboratory - Sacramento model application to Workshop data
Stewart Franks, University of Newcastle, Australia - Parameter estimation methods
Emeil van Loon, Wageningen University, The Netherlands - "Black box" regression approach to parameter estimation
Jasper Urugt, University of Arizona - Correlation approach to a priori parameter estimation
Alain Pietroniro, National Hydrology Research Centre, Canada - MAGS approach to parameter estimation using Watflood and Watclass
Balazs Fekete, University of New Hampshire - Simple water balance global calibration
Andy Young, Center for Ecology and Hydrology, UK - UK flood study approach to regionalization
Terri Hogue, University of Arizona - Multi-step automatic calibration schemes (MACS)
Feyzan Misirli, University of Arizona - Baysian recursive estimation (BaRE) of parameters for conceptual models
Newsha Khodatalab, University of Arizona - Distributed modelling intercomparison project
Florence Habets, Meteo France - Rhone-AGGreg, a GLASS/GSWP experiment
Luis Bastidas University of Arizona - Calibration of coupled land surface models
John Schaake, National Weather Service, Hydrology Laboratory - Results of Workshop a priori parameter estimation procedures; Summary and future activities
Hoshin Gupta, University of Arizona - Summary of Day 1 discussions
Alan Hall, IAHS/WMO Working Group on GEWEX - Summary of scientific issues for a priori parameter estimation put forward by Workshop participants

RESULTS OF ANALYSIS OF SUBMITTED MODEL RESULTS

Results from eight models were submitted, four from outside the US (France, Canada, Japan and Russia).

Two of the international participants provided data sets for future MOPEX studies. This included data sets for 12 basins in the Rhone (France) and 15 basins in the UK. These data sets will likely become some of the most important in the MOPEX database.

Analysis of all of the model results is only partially complete because some of the results were not submitted until during the Workshop and some of the results are still being prepared and will be submitted later. The main purpose of the present model runs is to provide a benchmark on how well several different models can function in an a priori mode (ie. Prediction of Ungauged Basins - PUBs) and to show how much improvement can be achieved through model calibration. Another objective is to test the potential usefulness of the MOPEX database.

Each modeller presented an assessment of their model results and their experience using the MOPEX data sets. There were 11 additional presentations, several by students from the University of Arizona, on related science issues. The main results so far show that, where there is adequate high quality data, calibration can significantly improve the a priori results. This means there is a substantial challenge to develop improved a priori parameter estimation techniques

FOCUS SCIENCE ISSUES

Based on the key science issues provided by each participant and the discussions during the Workshop the following topics and related science questions to address parameter estimation have been developed:

1. Model evaluation issue/diagnostic tools
A. How do you judge model successes and failures?
B. What measures should be used: fitting statistics, patterns, or others?
C. What diagnostic tools are available?
D. How do extract the maximum information from observations?

2. Tucson MOPEX Workshop results
A. What are the results from different models?
B. Why are they different?

3. Model calibration issues:
A. What calibration criteria should be used?
B. How much data are needed for meaningful calibration?
C. What are the latest calibration tools?
D. How do use a priori knowledge to constrain model calibration?
E. How can we "tune" distributed model parameters?

4. Data for model calibration and validation/use of soft data/effect of data errors:
A. What kind of data and how much data are needed?
B. What data can be used to evaluate models besides streamflow?
C. How can "soft" data be used to evaluate models and improve model structure?
D. What are effects of forcing data errors on model parameters and how do we account for forcing data errors?
E. How important is it to account for spatio-temporal variability in forcing data?

5. Quantification of uncertainty/model structure/model parameter estimation:
A. How do we assess uncertainty from data, model structure, and model parameters?
B. What are the implications of equifinality on model structure and parameter estimation?
C. Is there a scale at which the model uncertainty can be minimized?
D. How complex should models be, given the limits in data quantity and quality?

6. Model parameters and basin characteristics/regionalization issues:
A. Do model parameters have physical meaning? How do we distinguish parameters and fudge factors?
B. What information about watershed characteristics can be inferred and how can this be related to model structure/parameters?
C. How useful is it to represent spatial heterogeneity of land surface characteristics in models? How can we take advantage of the knowledge in heterogeneity?
D. How do we relate conceptual model parameters to land surface characteristics?
E. If the relationships between model parameters and land surface characteristics exist, do these relationships work different spatial scales?
F. How do climate related to model parameters?

7. Transferability/PUB:
A. How can knowledge about model parameters from one location be transferred to other locations?
B. How can model calibration help to improve relationships between model parameters and watershed characteristics?
C. Can knowledge about the parameters of one model be used by other models?
D. How can we take a given model and associated parameters at one scale and make them work at other scales?
E. What are the links to PUBs?

SUMMARY OF WORKSHOP AND FUTURE ACTIVITIES

1. The next comparisons of a priori parameter estimation techniques is to push forward on developing improved a priori parameter estimation techniques and on the regional estimation of parameters and to check their reality. A priori estimates should include estimates of their uncertainty. The goal is to narrow the uncertainty in the parameters and to improve the simulation of hydrological parameters.

2. The science issues have been discussed extensively. These will be used by the Workshop organisers to propose the focus science areas for MOPEX and foster collaborative activities in these areas.

3. Participants are requested to finish their comparison runs by the end of May.

4. The organisers are to prepare a paper/ report on MOPEX and the results of the Workshop. Several MOPEX publications were suggested. One would be a brief descriptive article for the EOS and IAHS newsletters. Another would be a report on the MOPEX database. And one would be a report on the model inter-comparison results of the Tucson workshop with discussion of related science issues.

5. The connection with GLASS is to be pursued by Luis Bastidas and with CEOP by John Schaake.

6. The next MOPEX workshop will be during the IUGG/IAHS General Assembly in Sapporo, July, 2003 - HW08 Parameter Estimation Techniques. This three-day workshop will: (i) consider progress being made to develop improved a priori parameter estimation techniques; (ii) review analyses of MOPEX data sets that were suggested at Tucson; and (iii) provide opportunity for discussion of results in the focus science areas identified by the Tucson workshop. The Workshop will be structured around the topics and related science questions described earlier. Each day of the workshop will end with a discussion of what the experience of that day suggests for future MOPEX activities. Contributions to HW08 are invited from researchers in this field and abstracts are due by January 2003.

Qingyun Duan, Hoshin Gupta, Alan Hall and John Schaake