Estimating the Uncertainty of Hydrologic Signatures through Model-free Discharge Resampling and its use for Model Diagnostics
Abstract
The diagnostic approach to model evaluation quantifies model performance using major behavioral functions, so-called signatures, of the system of interest. This method requires (i) a portfolio of signatures that measure complementary information about system behavior, (ii) signature tolerances to partition the model space into behavioral and nonbehavioral solutions, and (iii) a search method that explores the behavioral model space. This presentation is concerned with the second requirement, specifically, with a statistically rigorous quantification of signature uncertainty. To this end, we will present a model-free discharge resampling method that generates many probable replicates of the measured discharge records. Using discharge time series of 500+ watersheds of the CAMELS data set, we show that the replicates portray accurately the estimated discharge measurement uncertainty and preserve statistical and hydrologic properties of the measured discharge record. Next, we use the replicates to quantify signature uncertainty. If time permits, we will present preliminary results about the use of the estimated signature uncertainty to analyze model structural errors.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFM.H52C..04D