Strategies for Simplifying Reactive Transport Models: A Bayesian Model Comparison
DOI: https://doi.org/10.1029/2020WR028100
Persistent URL: http://resolver.sub.uni-goettingen.de/purl?gldocs-11858/8442
Persistent URL: http://resolver.sub.uni-goettingen.de/purl?gldocs-11858/8442
Schäfer Rodrigues Silva, Aline; Guthke, Anneli; Höge, Marvin; Cirpka, Olaf A.; Nowak, Wolfgang, 2020: Strategies for Simplifying Reactive Transport Models: A Bayesian Model Comparison. In: Water Resources Research, Band 56, 11, DOI: 10.1029/2020WR028100.
|
View/
|
For simulating reactive transport on aquifer scale, various modeling approaches have been proposed. They vary considerably in their computational demands and in the amount of data needed for their calibration. Typically, the more complex a model is, the more data are required to sufficiently constrain its parameters. In this study, we assess a set of five models that simulate aerobic respiration and denitrification in a heterogeneous aquifer at quasi steady state. In a probabilistic framework, we test whether simplified approaches can be used as alternatives to the most detailed model. The simplifications are achieved by neglecting processes such as dispersion or biomass dynamics, or by replacing spatial discretization with travel‐time‐based coordinates. We use the model justifiability analysis proposed by Schöniger, Illman, et al. (2015, https://doi.org/10.1016/j.jhydrol.2015.07.047) to determine how similar the simplified models are to the reference model. This analysis rests on the principles of Bayesian model selection and performs a tradeoff between goodness‐of‐fit to reference data and model complexity, which is important for the reliability of predictions. Results show that, in principle, the simplified models are able to reproduce the predictions of the reference model in the considered scenario. Yet, it became evident that it can be challenging to define appropriate ranges for effective parameters of simplified models. This issue can lead to overly wide predictive distributions, which counteract the apparent simplicity of the models. We found that performing the justifiability analysis on the case of model simplification is an objective and comprehensive approach to assess the suitability of candidate models with different levels of detail. Plain Language Summary:
In groundwater, chemical substances like nitrate are transported and undergo chemical reactions. Understanding such reactive transport processes plays a key role in securing our water resources and drinking water. We use computer models for understanding such reactive transport processes and for simulating their future behavior. In such models, we make many scientific decisions on which processes should be included and in what degree of detail. Here, we face a trade‐off: Usually, a complex model with many mathematical terms resolves many details of the process. Yet, such complex models require lots of data for calibration and lots of time for the computer simulation. In contrast, a simple model with fewer details comes with less effort in both respects. However, it might neglect important parts of the process. For the example of nitrate decay, we use a probabilistic approach to find the best simplification for a comparatively detailed reference model. Our results show that, in certain cases, it is justified to employ a simpler model instead of a complex alternative without deteriorating modeling results. Alongside, we explain how difficult it can be to define realistic parameter ranges for simplified models. Key Points:
We compare a set of four simplified models against a reference model for reactive transport at quasi steady state on aquifer scale.
A Bayesian model justifiability analysis helps identifying the most suitable model simplification strategy.
The proposed analysis reveals the difficulty of reasonably constraining parameter priors for simplified models.
Statistik:
View StatisticsCollection
- Geologie [933]
This is an open access article under the terms of the Creative Commons Attribution‐NonCommercial License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.