Keywords

Model structure uncertainty, Modular Modelling Framework, Model genealogy, Dendrogram, Sampling model space

Start Date

16-9-2020 4:00 PM

End Date

16-9-2020 4:20 PM

Abstract

This study explores the compatibility of two approaches for hydrological model structure uncertainty assessment: modular modelling frame-works, and model ensemble runs. We created dendrograms, a method that stems from biology to determine family trees, based on both model structure and model output. We analysed how well model structure families could be inferred from signatures in model output over different climates, and as such, if model structure uncertainty can correctly be quantified based on model output. Results from a synthetic experiment over 671 climate in-stances showed that the performance of the inference depends on the type of signature evaluated, and the climate. However, the performance of the inference is overall low, implying that model structure families are different from model output families. The novel combination of approaches employed in this study demonstrates the need to clarify how we want to sample model space when investigating model structure uncertainty.

Stream and Session

false

Share

COinS
 
Sep 16th, 4:00 PM Sep 16th, 4:20 PM

Can model structure families be inferred from model output?

This study explores the compatibility of two approaches for hydrological model structure uncertainty assessment: modular modelling frame-works, and model ensemble runs. We created dendrograms, a method that stems from biology to determine family trees, based on both model structure and model output. We analysed how well model structure families could be inferred from signatures in model output over different climates, and as such, if model structure uncertainty can correctly be quantified based on model output. Results from a synthetic experiment over 671 climate in-stances showed that the performance of the inference depends on the type of signature evaluated, and the climate. However, the performance of the inference is overall low, implying that model structure families are different from model output families. The novel combination of approaches employed in this study demonstrates the need to clarify how we want to sample model space when investigating model structure uncertainty.