Highly parameterized numerical models are frequently used in environmental sciences to interpret physicochemical processes. Despite their popularity, unjustified model complexity often undermines model generalizability, resulting in poor model predictive capabilities. A trade-off between model performance and complexity can be achieved using model selection techniques, which are largely unexplored in biogeochemical modeling. Our study aims to address this scientific gap by investigating the main advantages and findings of model selection in a typical biogeochemical modeling application. In particular, the study focuses on the numerical description, at five different levels of complexity, of the reactive transport processes of nitrogen species in a controlled aquifer model laboratory setup. To this end, the mechanistic model HYDRUS is coupled with the multimodal Nested Sampling algorithm for the estimation of the Bayesian model evidence and posterior parameter distributions. After successfully calibrating the HYDRUS-2D model against measurements from a tracer test, a lower-fidelity surrogate-based analysis is carried out to tackle the problem of computational cost, as well as to compare five different models which combine first-order and Monod-based parameterization of the microbial nitrogen degradation. Results demonstrate that measured data contain enough information content to support an increase in the complexity of the Monod model, which can reproduce the transient accumulation of nitrite at the beginning of the experiment. The Bayesian analysis reveals that such behavior is related to the particular structure of measured data, which plays a fundamental role in the model selection. The analysis concludes with a validation of the surrogate-based analysis, which confirms the reliability of the proposed numerical approach, opening up new perspectives on the use of multi-fidelity surrogates for the Bayesian model selection in environmental modeling.

Handling model complexity with parsimony: Numerical analysis of the nitrogen turnover in a controlled aquifer model setup

Brunetti, G
;
2020-01-01

Abstract

Highly parameterized numerical models are frequently used in environmental sciences to interpret physicochemical processes. Despite their popularity, unjustified model complexity often undermines model generalizability, resulting in poor model predictive capabilities. A trade-off between model performance and complexity can be achieved using model selection techniques, which are largely unexplored in biogeochemical modeling. Our study aims to address this scientific gap by investigating the main advantages and findings of model selection in a typical biogeochemical modeling application. In particular, the study focuses on the numerical description, at five different levels of complexity, of the reactive transport processes of nitrogen species in a controlled aquifer model laboratory setup. To this end, the mechanistic model HYDRUS is coupled with the multimodal Nested Sampling algorithm for the estimation of the Bayesian model evidence and posterior parameter distributions. After successfully calibrating the HYDRUS-2D model against measurements from a tracer test, a lower-fidelity surrogate-based analysis is carried out to tackle the problem of computational cost, as well as to compare five different models which combine first-order and Monod-based parameterization of the microbial nitrogen degradation. Results demonstrate that measured data contain enough information content to support an increase in the complexity of the Monod model, which can reproduce the transient accumulation of nitrite at the beginning of the experiment. The Bayesian analysis reveals that such behavior is related to the particular structure of measured data, which plays a fundamental role in the model selection. The analysis concludes with a validation of the surrogate-based analysis, which confirms the reliability of the proposed numerical approach, opening up new perspectives on the use of multi-fidelity surrogates for the Bayesian model selection in environmental modeling.
2020
Modeling
Model selection
Lower-fidelity surrogate model
HYDRUS
Nitrification
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.11770/345904
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 16
  • ???jsp.display-item.citation.isi??? 17
social impact