Using Bayesian Networks for Sensitivity Analysis of Complex Biogeochemical Models

Dai, Heng; Ye, Ming; Song, Xianfang; Hammond, Glenn Edward; Hu, Bill X.; Zachara, John M.

doi:10.1029/2018wr023589

Cited by 27 publications

(14 citation statements)

References 51 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The aquifer‐river interface is composed of a low‐permeability sandy layer of recent fluvial deposition. The thin alluvial layer (0.5 to ~2 m) has an important influence on HEFs as it dampens river fluctuation propagation into the aquifer (Dai et al, 2019; Hammond & Lichtner, 2010; Song et al, 2018; Zachara et al, 2020; Zhou et al, 2018).…”

Section: Materials and Methodologymentioning

confidence: 99%

River Dynamics Control Transit Time Distributions and Biogeochemical Reactions in a Dam‐Regulated River Corridor

Song

Zachara

Shuai

et al. 2020

Water Resources Research

Self Cite

View full text Add to dashboard Cite

Transit time distributions (TTDs) exert important controls on biogeochemical processes in watershed systems. TTDs are often assumed to follow time-invariant exponential, lognormal, or heavy-tailed power law distributions in headwater or low-order streams. However, under dynamic hydrological forcing, transit time could exhibit more complex distribution patterns with strong spatial and temporal variability. In this study, we used a numerical particle tracking approach to characterize TTDs along the Hanford Reach of the Columbia River under the influences of river stage fluctuations and evaluate the associated effects on biogeochemical reaction potentials within the river corridor. Particle tracking was conducted using velocity fields simulated by high-resolution three-dimensional groundwater flow models that capture both the river stage fluctuations and physical heterogeneity. Our results revealed that multifrequency flow variations led to multimodal TTDs that varied in time and space. Such characteristics can only be captured by multiyear numerical simulations supported by multiyear field monitoring. Dam-induced high-frequency (subweekly) flow variations increased additional hydrologic exchange flows with short (subweekly) transit times, which accounted for up to 44% of reactant consumption in the river corridor along the Hanford Reach. The dam-induced river stage fluctuations have more significant impacts on faster biogeochemical reactions because they cause a larger fraction of shorter transit times. Numerical particle tracking provides an efficient alternative for characterizing TTDs for large complex systems where in situ field experiments are not feasible. Such a numerical approach is thus essential for improving large-scale biogeochemical modeling from watersheds to basins.

show abstract

Section: Materials and Methodologymentioning

confidence: 99%

River Dynamics Control Transit Time Distributions and Biogeochemical Reactions in a Dam‐Regulated River Corridor

Song

Zachara

Shuai

et al. 2020

Water Resources Research

Self Cite

View full text Add to dashboard Cite

show abstract

“…Nevertheless, our estimates of hydraulic properties can provide reliable inputs with reduced uncertainty toward improved numerical models in both accuracy and precision. This top of the permeability field is one of the most influential factors that control the magnitude (Lackey et al, 2015;Bao et al, 2018), spatial extent (Schilling et al, 2017), and associated biogeochemical hot spots (Song et al, 2018;Dai et al, 2019) of HEFs.…”

Section: Discussionmentioning

confidence: 99%

Spatial Mapping of Riverbed Grain-Size Distribution Using Machine Learning

et al. 2020

Self Cite

View full text Add to dashboard Cite

Recent alluvial sediments in riverbeds play a significant role in controlling hydrologic exchange flows (HEFs) in river systems. The alluvial layer is usually associated with strong heterogeneity in physical properties (e.g., permeability and hydraulic conductivity), which affects local HEFs and therefore biogeochemical processes. The spatial distribution of these physical properties needs to be determined to inform the numerical models used to reveal the realistic hydro-biogeochemical behaviors. Such information can be obtained based on the intrinsic link between sediment grain-size distribution and hydraulic properties where sediment texture information is available. However, grain-size measurements are usually spatially sparse and do not have adequate coverage and resolution, particularly for a relatively large domain such as the Hanford Reach of the Columbia River. In this paper, we adopted machine learning (ML) approaches for categorizing and mapping the spatial distributions of riverbed substrate grain size and filling in missing areas of substrate data using the ML models along the reach. Such ML models for substrate size mapping were trained at 13,372 locations using measured substrate sizes along with observed and simulated attributes, including bathymetric attributes (e.g., elevation, slope, and aspect ratio) from LIDAR and bathymetric surveys, and hydrodynamic properties (e.g., water depth, velocity, shear stress, and their statistical moments). An ensemble bagging-based ML technique, Random Forest, was adopted to identify the most influential factors as predictors to develop the predictive models with over-fitting issues addressed. The models were evaluated with respect to each individual substrate size class and the lumped group, and then used to generate the final substrate size maps covering all the grid cells in the numerical modeling domain.

show abstract

“…The model weights can be evaluated in more sophisticated ways, for example, using Bayesian network as shown in Dai et al. (2019).…”

Section: Methodsmentioning

confidence: 99%

“…The combination, B 1 C 1 , of process models B 1 and C 1 has the probability of P(B 1 C 1 ) = P(B 1 ) × P(C 1 ) = 0.7 × 0.6 = 0.42. The model weights can be evaluated in more sophisticated ways, for example, using Bayesian network as shown in Dai et al (2019).…”

Section: Evaluation Of Total-effect Process Sensitivity Indexmentioning

confidence: 99%

Process Interactions Can Change Process Ranking in a Coupled Complex System Under Process Model and Parametric Uncertainty

Jian

Dai

et al. 2022

Water Resources Research

Self Cite

View full text Add to dashboard Cite

For a complex hydrologic system with multiple processes and process interactions, global sensitivity analysis is often used to identify important or influential parameters for model development and improvement. The identification is complicated by process model uncertainty, when a system process can be represented by multiple process models. This study develops a new total‐effect process sensitivity index to identify influential processes under model uncertainty. This is done by extending Sobol's total‐effect parameter sensitivity index for one system model to total‐effect process sensitivity index for multiple system models to account for uncertainty in process models and model parameters. The total‐effect process sensitivity index includes not only the first‐order process sensitivity index for measuring the importance of individual processes but also higher‐order indices that account for process interactions. The total‐effect process sensitivity index can identify an influential process that itself and its interactions with other processes influence a model output. The total‐effect process sensitivity index is applied to two numerical examples: (a) Sobol's G*‐functions with analytical solutions of first‐order and total‐effect process sensitivity indices, and (b) groundwater flow models with interactions between recharge, geology, and snowmelt processes. The second evaluation shows that, due to second‐order and higher‐order process interactions, the first‐order and total‐effect process sensitivity indices give different process ranking. It is thus necessary to estimate both first‐order and total‐effect process sensitivity indices to appreciate the difference between the first‐order impact of a process alone and the overall total‐effect impact of the process itself and its interactions with other processes on a model output.

show abstract

Using Bayesian Networks for Sensitivity Analysis of Complex Biogeochemical Models

Cited by 27 publications

References 51 publications

River Dynamics Control Transit Time Distributions and Biogeochemical Reactions in a Dam‐Regulated River Corridor

River Dynamics Control Transit Time Distributions and Biogeochemical Reactions in a Dam‐Regulated River Corridor

Spatial Mapping of Riverbed Grain-Size Distribution Using Machine Learning

Process Interactions Can Change Process Ranking in a Coupled Complex System Under Process Model and Parametric Uncertainty

Contact Info

Product

Resources

About