Combining Clustering Methods With MPS to Estimate Structural Uncertainty for Hydrological Models

Vilhelmsen, Troels Norvin; Auken, Esben; Christiansen, A.V.; Barfod, Adrian A. S.; Marker, Pernille Aabye; Bauer‐Gottwein, Peter

doi:10.3389/feart.2019.00181

Cited by 19 publications

(16 citation statements)

References 71 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Tahmasebi (2018) also showcase different applications of MPS in his state-of-the-art review of the method. Several studies have shown the applicability of MPS for modeling the geological architecture of major lithological units (Jørgensen et al 2015;Barfod et al 2018;Vilhelmsen et al 2019), although we agree that the application of MPS for real world scenarios is not overwhelming. Høyer et al (2017) are simulating the architecture of a thick succession of Miocene sand and clay units within a confined area of the subsurface.…”

Section: Interactive Commentmentioning

confidence: 55%

Respond to Donald Keefer

Madsen¹

2020

View full text Add to dashboard Cite

We would like to thank Donald Keefer for his referee comments. In the following, citations from the interactive comment are written with RC1 in front of them. From reading the comments, we have not been able to describe and state the work frame and purpose of the project thoroughly. We think that a lot of the comments are based on a misconception of the modeling purpose and detail. Before commenting, we therefore start by stating the purpose of the modeling. The work we present is part of a project (The Mapfield project) that aims at developing a concept for targeted N regulation.

show abstract

Section: Interactive Commentmentioning

confidence: 55%

Respond to Donald Keefer

Madsen¹

2020

View full text Add to dashboard Cite

show abstract

“…The issues of nonstationarity in MPS are discussed in detail in Mirowski et al (2009). Another point is that issues with non-stationarity can be remedied with large amounts of conditional data (see e.g., Barford et al (2018) or Vilhelmsen et al (2019)).…”

Section: Discussion Papermentioning

confidence: 99%

Response to RC2

Madsen¹

2020

View full text Add to dashboard Cite

We would like to thank the anonymous reviewer for his/her referee comments. In the following citations from the interactive comment is written with RC2 in front of them. RC2: "Overview: This paper presents an approach for joint stochastic simulation of subsurface geological and redox architectures in 3D using multiple point geostatistics (MPS). The method is demonstrated on a small catchment in Denmark, where simulations are conditioned on observed resistivities of the subsurface from towed transient electromagnetic measurements (tTEM) as well as on soil maps and borehole observations of lithology, sediment colours and water chemistry. The paper is interesting C1 HESSD

show abstract

“…Through experiments with geological interpretation of the uncertainty in boreholes, Randle et al (2019) argued that expert elicitations do not result in accurate predictions of interpretation error. Efforts have been made to make TI generators (Pyrcz et al, 2008) and data-driven TIs without the need for expert knowledge (Vilhelmsen et al, 2019). However, process-based TI generation 605 from expert elicitation is a common approach in MPS applications (Mariethoz and Caers, 2015).…”

Section: Geological Modeling Subjectivity and Data Reuse 600mentioning

confidence: 99%

“…Consequently, we might risk reproducing larger parts of the TI in the realizations. Such verbose copying is partially remedied by the addition of conditioning data and choosing 640 a smaller search radius as argued in Vilhelmsen et al (2019). However, a smaller search radius comes at a price of not reproducing the features in the TI and adding variability more related to algorithmic choices than geological variability.…”

Section: Training Images and Geological Elements 625mentioning

confidence: 99%

“…In recent years, some studies have adapted geostatistical simulation methods for geological mapping of the substratum in order to quantify and possibly account for some of these uncertainties. A few examples exist of multiple-point geostatistical 85 simulation (MPS) utilized for mapping 3D geology with AEM data (Barfod et al, 2018;He et al, 2014b;Høyer et al, 2017;Jørgensen et al, 2015;Vilhelmsen et al, 2019). However, AEM data provide structural information of the deeper subsurface (100-200 m deep) at a coarser resolution (Sørensen and Auken, 2004), and hence may not be adequate to provide structural information for simulations of N-transport at catchment level occurring mainly within the upper 30 m. A newly developed towed transient electromagnetic method (tTEM) (Auken et al, 2019) provides data at much higher resolution but with a lower 90 penetration depth than AEM.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

3D multiple point geostatistical simulation of joint subsurface redox and geological architectures

Madsen¹,

Kim²,

Kallesøe³

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

Abstract. Nitrate contamination of subsurface aquifers is an ongoing environmental challenge due to nitrogen (N) losses from intensive N fertilization and management on agricultural fields. The distribution and fate of nitrate in aquifers are primarily governed by geological, hydrological and geochemical conditions of the subsurface. Therefore, we propose a novel approach to model both geology and redox architectures simultaneously in high resolution 3D (25 m × 25 m × 2 m) using multiple point geostatistical simulation (MPS). Data consists of 1) mainly resistivities of the subsurface mapped with towed transient electromagnetic measurements (tTEM), and 2) information of lithology and redox conditions obtained from the borehole observations at point scale interpreted from the geological descriptions and colors of sediment samples, and chemistry analyses of water samples. The conceptual understandings of the geological and redox architectures of the study system were introduced to the simulation as training images. These data were combined with detailed soil maps and digital elevation models to identify main geological elements defined as volumes within the subsurface. From a geological perspective, these data are considered independent from each other in terms of formation. This approach became computationally attractive by simulating smaller geological elements individually instead of the entire catchment. The final realizations were stitched together using simulations of the individual geological elements, resulting in an ensemble of realizations representing a quantification of the uncertainty for the given setup. The joint simulation of geological and redox architectures, which is one of the strengths of the MPS simulations compared to other geostatistical methods, secures that the two architectures in general show coherent patterns. Despite the inherent subjectivity of interpretations of the training images and geological element boundaries, they enable an easy and intuitive incorporation of qualitative knowledge of geology and geochemistry in quantitative simulations of the subsurface architectures. Altogether, we conclude that our approach effectively simulates the coherent geological and redox architectures of the subsurface that can be used for hydrological modelling with N-transport, which may be fundamental to better understanding of the nitrogen (N) retention of the subsurface and to future more targeted regulation of agriculture.

show abstract

Combining Clustering Methods With MPS to Estimate Structural Uncertainty for Hydrological Models

Cited by 19 publications

References 71 publications

Respond to Donald Keefer

Respond to Donald Keefer

Response to RC2

3D multiple point geostatistical simulation of joint subsurface redox and geological architectures

Contact Info

Product

Resources

About