All models are wrong, but some are useful: Assessing model limitations for use in decision making and future model development

Autor: Apostel, Anna Maria
Jazyk: angličtina
Rok vydání: 2021
Předmět:
Druh dokumentu: Text
Popis: The return of severe algal blooms to the Western Basin of Lake Erie has refocused efforts to manage nutrients in and around the Great Lakes. An important part of this effort has been extensive water quality modeling in the region, especially in watersheds responsible for excessive nutrient loading to at-risk lake basins. Models can expand the predictive impact of limited monitoring data, and therefore provide a powerful tool for water managers. However, models are limited by numerous shortcomings, including data availability, model structure, and equifinal model solutions. Bringing light to these potential issues in model development and implementation is key in the effective use of and public trust in modeling results.The work presented in this document is comprised of four main objectives aimed at examining model confidence. First (Chapter 2), a Soil and Water Assessment Tool (SWAT) model was developed for the Maumee River watershed at the near-farm-field scale, incorporating the best available data for the region. This new model was compared against previous model iterations using edge-of-field monitoring data. A key improvement in soil P initialization values revealed a potential structural limitation in the model to simulate phosphorus export in surface runoff. Second (Chapter 3), a retrospective analysis of land management changes simulated over the past several decades was completed to examine the influence of individual agricultural management practices on driving discharge and loading trends. While climate played a major role in driving discharge patterns, tillage had a significant impact on nutrient loading. Third (Chapter 4), the presence of equifinality—that many differing parameterizations can produce acceptable models—was examined, along with the potential to reduce equifinality using increased data in calibration. A Latin Hypercube Sampling approach was used to select values for 15 parameters, and then constraints were applied across data types and performance metrics to narrow parameter set selection. Nutrient-related constraints had the strongest tendency to restrict parameter space, leading to models with greater confidence. Finally (Chapter 5), the model was used to compare uncertainty regarding parameterization to that from agricultural management assumptions. While much equifinality study focuses on calibration techniques, there is less study on model inputs, and particularly those related to agricultural practices; yet earlier chapters suggest management assumptions may be critical for confidence in nutrient loading. Parameter scenarios were defined from a suite of parameter sets representing a range of plausible values for commonly calibrated parameters, while farm management scenarios were based on potential management assumptions or omissions. The model was run for each unique combination of parameter and management scenarios, and model output variance was attributed to variation in parameters and management representations. While parameterization of the model was the major driver of discharge uncertainty, nutrients were mainly driven by uncertainty in farm management assumptions.Models do not provide perfectly correct answers, they allow for the conceptualization of complex systems and the extrapolation of data over areas for which it may not yet exist. Transparency in investigating, reporting, and communicating the uncertainty that is intrinsic in a model is needed if models are to be effectively used.
Databáze: Networked Digital Library of Theses & Dissertations
Externí odkaz: