The field-scale modeling activity, a partnership activity with the Interactive Design of Extreme-scale Application Software (IDEAS)–Watersheds project, is developing, evaluating, and refining multiscale modeling approaches and software frameworks that allow increasingly detailed understanding of fine-scale biogeochemical processes to be used in basin-scale river network models. Central to our strategy is our new multiscale ADELS model (Painter 2018; 2021) that extends highly successful residence-time models to accommodate nonlinear multicomponent reactions. The approach moves biogeochemical process representation to subgrid models that each represent an ensemble of hyporheic-zone flowpaths. Shifting to subgrid models allows biogeochemical processes to be represented in great detail at their native spatial scales without averaging over important, fine-scale variability in redox states that occurs within sediments and periphyton biofilms.
The broad objective of integrated activity on model development and parameterization is to facilitate state-of-theart advancements in process-based modeling of reactive transport in stream systems using Hg in EFPC as a representative use case. Specific objectives are to implement and test the new multiscale model in the ATS integrated hydrology software, develop and evaluate methods for estimating model parameters from stream tracer tests, and demonstrate the capability in watershedscale reactive transport simulations.
The implementation of ADELS in ATS was extended from conservative tracers to multicomponent reactive transport. The reactive transport capability uses the Alquimia interface to access the general geochemical reaction capability within the PFLOTRAN software. Multiple subgrid models with different reaction models may be specified for each channel grid cell to represent, for example, metabolically active and inactive transient storage. The implementation was verified against independent solutions. Mesh convergence studies using denitrification, as an example, revealed that semi-distributed conceptualizations, like that used for reactive transport in the Soil & Water Assessment Tool (SWAT) model, can produce significant spatial discretization error (see figure, top panel). Additional simulations using a subbasin of the Portage River Basin reveal complex spatial patterns in denitrification, including spatially localized hotspots (see figure, bottom panel) that are difficult to represent in existing simulation tools.
Multiscale Advanced Terrestrial Simulator (ATS) Simulations. Top: Difference in calculated denitrification between coarse- and fine-resolution representations of the stream network in EFPC. The coarse representation, which used one channel grid cell per channel segment as in semi-distributed models like SWAT, resulted in significant spatial discretization error. Bottom: Calculated nitrate concentration in a nitrate loading experiment in a subbasin of the Portage River Basins.
Methods for estimating model parameters from stream tracer test results were developed, implemented in Python workflows, and evaluated (Rathore et al. 2021) using a public dataset from tracer tests in the Hammer Stream in West Sussex, UK. For that work, we used Bayesian inference and the Markov Chain Monte Carlo (MCMC) method, which provides estimates of uncertainty in the various model parameters. The hyporheic travel time distributions were estimated without assumptions about the distribution shape. For the first time, the shape-free distributions were estimated simultaneously with channel properties. Moreover, we demonstrated that simultaneous analysis of observations from multiple locations constrains parameters better than analyzing those observations individually. Results from that work are shown in the figure. The observational correlations provide additional confidence in the ADELS model. In addition, the methods and workflow tools developed provide the basis for analyzing future tracer tests in EFPC to estimate model parameters.
Stream Tracer Test Breakthrough Curves. Predictive uncertainty plot for breakthrough curves at four locations simulated together using an ensemble of jointly estimated parameters. Observed concentrations are represented by solid lines and ensemble of simulated BTCs is shown in the corresponding semi-transparent lighter shade. [Reprinted by permission from Rathore, S. S., et al. 2021. “On the Reliability of Parameter Inferences in a Multiscale Model for Transport in Stream Corridors.” Water Resources Research. 57(5). DOI: 10.1029/2020WR028908. Copyright 2021. John Wiley & Sons, Inc.]
A novel Bayesian joint-fitting strategy was developed (see figure) and used to reinterpret results from the sorption and methylation experiments on EFPC sediments (see Theme 1 Accomplishments). Specifically, we applied Bayesian inference and MCMC simultaneously to multiple sorption and methylation experiments instead of the traditional approach of analyzing subprocess models individually. Joint-fitting more rigorously propagates uncertainties between different subprocess models and also allows information that is shared between datasets to be more fully used. Bayesian inference and MCMC also produce the full joint distribution of parameters, thus providing global uncertainty estimates and facilitating the detection of overparameterization manifesting as null spaces in the parameter space. The identification of null spaces guided the simplification of certain subprocess models: fast kinetic sorption was replaced by equilibrium sorption, and Monod demethylation was replaced by first-order demethylation based on the analysis. The proposed scheme will benefit parameter estimation for other complex biogeochemical systems characterized through multiple experiments.
The ADELS model was compared to an alternative multirate model that conceptualizes the hyporheic zone as a large number of well-mixed transient storage zones that exchange solute directly with the channel, but not with each other. That analysis showed that ADELS and the multirate model are mathematically equivalent in the special case of nonreacting tracers. However, they produce very different predictions of reactive transport. In particular, direct exchange of oxygen between each transient storage zone and the stream channel suppresses redox zonation in the multirate model, thus limiting its potential as a general-purpose model for reactive transport in stream corridors. The comparison also emphasizes that nonreacting tracers alone are unable to constrain models for hyporheic exchange and need to be augmented by additional observations, like reactive tracer tests.
Illustration of Bayesian Scheme. Schematics of Bayesian joint-fitting scheme for calibrating complex biogeochemical models offer multiple advantages with efficient utilization of datasets, provide full uncertainty propagation and robust estimation, and detect parameterization deficiencies.
References Cited
Painter, S. L. 2018. “Multiscale framework for modeling multicomponent reactive transport in stream corridors.” Water Resources Research. 54: 7216–30. DOI: 10.1029/2018WR022831.
Painter, S. L. 2021. “On the representation of hyporheic exchange in models for reactive transport in stream and river corridors.” Frontiers in Water. 2:69. DOI:10.3389/frwa.2020.595538.
Rathore, S. S., et al. 2021. “On the Reliability of Parameter Inferences in a Multiscale Model for Transport in Stream Corridors.” Water Resources Research. 57(5). DOI: 10.1029/2020WR028908.
Security Notice | Contact Eric Pierce, ORNL | Website Questions | Site Map
File last modified: Wednesday, September 01, 2021
The ORNL Mercury SFA is sponsored by the Subsurface Biogeochemical Research (SBR) program within the U.S. Department of Energy's Office of Biological and Environmental Research.