Event Title

Over Parameterized Models and the Illusion of Reality

Presenter Information

Karthik Kumarasamy
Patrick Belmont

Location

Eccles Conference Center Auditorium

Event Website

http://water.usu.edu

Start Date

3-31-2015 12:30 PM

End Date

3-31-2015 12:40 PM

Description

Hydrology and land use models are commonly used to predict effects of changes in climate, land or water management on water quality, nutrients and sediment loading. Such models often form the primary basis for large scale policy decisions and management decisions that have far reaching implications. Yet, such models are typically constrained by a multitude of parameters, leading to problems of equifinality that remain unresolved and are rarely addressed. Inclusion of the validation step in the model development process has provided a sense of scientific credibility especially when measured quantities are satisfactorily predicted for the period of record where calibration was not attempted. This effort describes how multiple parameter combinations can result in similar outcomes and which have very different policy implications. We use the Le Sueur River Basin as a prototype study area, located in south central Minnesota where high turbidity in rivers are a major concern. First, we describe challenges involved in properly calibrating the hydrologic model in an area where extensive use of sub-surface tile drains has significantly altered rainfall-runoff patterns, and yet modelers are provided little information regarding the location, density, or connectivity of the drainage system. Next, we show that one could carefully calibrate a model to obtain acceptable performance indicator values (e.g., Nash-Sutcliff), while implicating completely different sources for the sediment (upland versus near-channel sediment sources). Such equifinality problems can be introduced as a result of model structure, input data, and/or certain simplifications or approximations used in model development and such problems can be overcome when the models are constrained by key independent, empirical observations.

This document is currently not available here.

Share

COinS
 
Mar 31st, 12:30 PM Mar 31st, 12:40 PM

Over Parameterized Models and the Illusion of Reality

Eccles Conference Center Auditorium

Hydrology and land use models are commonly used to predict effects of changes in climate, land or water management on water quality, nutrients and sediment loading. Such models often form the primary basis for large scale policy decisions and management decisions that have far reaching implications. Yet, such models are typically constrained by a multitude of parameters, leading to problems of equifinality that remain unresolved and are rarely addressed. Inclusion of the validation step in the model development process has provided a sense of scientific credibility especially when measured quantities are satisfactorily predicted for the period of record where calibration was not attempted. This effort describes how multiple parameter combinations can result in similar outcomes and which have very different policy implications. We use the Le Sueur River Basin as a prototype study area, located in south central Minnesota where high turbidity in rivers are a major concern. First, we describe challenges involved in properly calibrating the hydrologic model in an area where extensive use of sub-surface tile drains has significantly altered rainfall-runoff patterns, and yet modelers are provided little information regarding the location, density, or connectivity of the drainage system. Next, we show that one could carefully calibrate a model to obtain acceptable performance indicator values (e.g., Nash-Sutcliff), while implicating completely different sources for the sediment (upland versus near-channel sediment sources). Such equifinality problems can be introduced as a result of model structure, input data, and/or certain simplifications or approximations used in model development and such problems can be overcome when the models are constrained by key independent, empirical observations.

https://digitalcommons.usu.edu/runoff/2015/2015Posters/12