Representing Uncertainty with Diverse Model Ensembles: A Test Case in an Alpine Karst System

Streaming Media

Session Type

Technical Sessions: Conservation Science

Start Date

18-8-2020 1:00 PM

Description

Karst aquifers are difficult to model because the heterogeneous nature of groundwater flow through conduits, rather than through distributed pore spaces, leads to high structural uncertainty. Most existing models rely either on detailed conduit maps, or on effective flow parameters approximating a porous medium. Both approaches have significant failings, because karst systems are rarely fully mapped, and their flow behavior is fundamentally different from porous aquifers. Our approach links three components: geologic modeling, conduit network generation, and pipe flow modeling. We use pre-existing data from a long-term research site, the Gottesacker karst system in the German/Austrian Alps. First, we build several geologic models in GemPy, a Python package. Each geologic model is fed to the Stochastic Karst Simulator, a pseudo-genetic conduit evolution model, generating many conduit network maps. For each network, we estimate hydraulic parameters, and model the flow behavior using the E.P.A. Storm Water Management Model. This yields an ensemble of competing models, organized into a model tree recording the geologic structure, conduit network map, and hydraulic parameters for each model. The models in the ensemble will be ranked based on the fit of model-predicted spring discharge to observed data, and a sub-set of high-performing models used to project future discharge under different scenarios. We expect that the benefit of this structurally diverse model ensemble is that it will more fully represent the range of possible system behaviors. Finally, the ensemble will be com-pared to the known conduit network, to assess the effectiveness of our approach.

Comments

This presentation was part of the Technical Sessions on Conservation Science. Presentation topics ranged from cave conservation techniques, environmental education, community engagement, resource protection assessment, and scientific and cultural research from across the globe. Formats vary from traditional PowerPoints to films to story maps and informal interviews.

Share

COinS
 
Aug 18th, 1:00 PM

Representing Uncertainty with Diverse Model Ensembles: A Test Case in an Alpine Karst System

Karst aquifers are difficult to model because the heterogeneous nature of groundwater flow through conduits, rather than through distributed pore spaces, leads to high structural uncertainty. Most existing models rely either on detailed conduit maps, or on effective flow parameters approximating a porous medium. Both approaches have significant failings, because karst systems are rarely fully mapped, and their flow behavior is fundamentally different from porous aquifers. Our approach links three components: geologic modeling, conduit network generation, and pipe flow modeling. We use pre-existing data from a long-term research site, the Gottesacker karst system in the German/Austrian Alps. First, we build several geologic models in GemPy, a Python package. Each geologic model is fed to the Stochastic Karst Simulator, a pseudo-genetic conduit evolution model, generating many conduit network maps. For each network, we estimate hydraulic parameters, and model the flow behavior using the E.P.A. Storm Water Management Model. This yields an ensemble of competing models, organized into a model tree recording the geologic structure, conduit network map, and hydraulic parameters for each model. The models in the ensemble will be ranked based on the fit of model-predicted spring discharge to observed data, and a sub-set of high-performing models used to project future discharge under different scenarios. We expect that the benefit of this structurally diverse model ensemble is that it will more fully represent the range of possible system behaviors. Finally, the ensemble will be com-pared to the known conduit network, to assess the effectiveness of our approach.