Diagnostic Evaluation of Hydrological Models
Convener: Bettina Schaefli  | Co-Conveners: Jim Freer , Martyn Clark , Thorsten Wagener 
Oral Programme
 / Mon, 20 Apr, 15:30–17:00  / Room 35
Poster Programme
 / Attendance Mon, 20 Apr, 17:30–19:00  / Hall A

Reliable predictions of the hydrologic response of gauged and ungauged watersheds is a key objective of hydrological science and of crucial importance for many operational and scientific applications. Our current ability to achieve reliable predictions is restricted due to limitations in hydrological theory, limitations in mathematical models of watershed systems, and due to limitations in our capacity to observe important physical watershed characteristics and hydrologic
state variables. In gauged watersheds, we can compensate for these shortcomings through model calibration against observed system responses, e.g. discharge; for many applications- for prediction of ungauged catchments as well as for predictions of climate change impacts, model calibration to observed hydrologic behaviour is, however, only of limited use.

In these situations, reliable predictions can only be achieved if we can ensure that our model is consistent with the underlying system behaviour. Measuring this consistency is a key objective of what we call "model diagnostic evaluation" does the model give the right result for the right reasons? And if not, how can we improve the model?

The present session thus invites contributions that advance our ability to measure a model's consistency with the predicted natural behaviour. Of particular interests are studies that combine experimental and modeling techniques to address the following questions:

- How should we measure model performance? Which model performance measures have diagnostic power and how do they contribute to model improvement?
- What is the information content of hydrologic data? How much information do our models need? What is the definition of information content and how can we measure it?
- How can we separate and identify observation error versus model error? Assuming that that observational errors create mismatches between hydrologic model predictions and observations, how can we ensure that we are not trying to ‘fix’ the model to accommodate for observational errors?
- Models can often not simulate all observed system responses with constant model parameters and dynamic variations in parameter values are often an indication of model structural problems. How can we analyze temporal and process-related parameter variability and what strategies exist to modify our models to overcome these structural shortcomings?

This session is organized as part of the scientific activities of the PUB (Prediction in Ungauged Basins, www.pub.iwmi.org) working group "Uncertainty estimation in hydrological modeling" currently lead by the conveners of this session. The main task of this working group is to stimulate progress and to provide guidance in the theory and application of techniques for uncertainty estimation and diagnostic model evaluation.