Prediction‐Focused Approaches: An Opportunity for Hydrology

Hermans, Thomas

doi:10.1111/gwat.12548

Cited by 23 publications

(26 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The 5 day experiment results are more reliable but are obtained at the cost of additional data acquisition. This illustrates that PFA can be used within the context of experimental design (Hermans, 2017). The results are also coherent with the deterministic solution (Figure 3a, green curve).…”

Section: Heat Storage Stochastic Prediction and Uncertainty Quantificsupporting

confidence: 81%

Uncertainty Quantification of Medium‐Term Heat Storage From Short‐Term Geophysical Experiments Using Bayesian Evidential Learning

Hermans

Nguyen

Klepikova

et al. 2018

Water Resources Research

Self Cite

View full text Add to dashboard Cite

In theory, aquifer thermal energy storage (ATES) systems can recover in winter the heat stored in the aquifer during summer to increase the energy efficiency of the system. In practice, the energy efficiency is often lower than expected from simulations due to spatial heterogeneity of hydraulic properties or non‐favorable hydrogeological conditions. A proper design of ATES systems should therefore consider the uncertainty of the prediction related to those parameters. We use a novel framework called Bayesian Evidential Learning (BEL) to estimate the heat storage capacity of an alluvial aquifer using a heat tracing experiment. BEL is based on two main stages: pre‐ and postfield data acquisition. Before data acquisition, Monte Carlo simulations and global sensitivity analysis are used to assess the information content of the data to reduce the uncertainty of the prediction. After data acquisition, prior falsification and machine learning based on the same Monte Carlo are used to directly assess uncertainty on key prediction variables from observations. The result is a full quantification of the posterior distribution of the prediction conditioned to observed data, without any explicit full model inversion. We demonstrate the methodology in field conditions and validate the framework using independent measurements.

show abstract

Section: Heat Storage Stochastic Prediction and Uncertainty Quantificsupporting

confidence: 81%

Uncertainty Quantification of Medium‐Term Heat Storage From Short‐Term Geophysical Experiments Using Bayesian Evidential Learning

Hermans

Nguyen

Klepikova

et al. 2018

Water Resources Research

Self Cite

View full text Add to dashboard Cite

show abstract

“…In addition, taken as a whole, the model ensemble should represent what we know (e.g., physical laws), what we think we know (our prevailing assumptions regarding structures and relevant processes), and what we know that we do not know (less certain assumptions regarding structure, processes, and parameter values). These prescriptions for forming useful models and model ensembles have been discussed previously [2,4,[46][47][48][49][50][51][52]. Similarly, the element of surprise [53,54], which describes the discovery of unknown unknowns, represents the collection of elements that were not included in any model in the ensemble.…”

Section: Bayes From a Hydrologist's Perspective-the Basic Storymentioning

confidence: 99%

“…First, it is worth deciding if a marginally better fit to limited data truly constitutes a better model for decision support. Second, the computational demands of ensemble modeling may provide further support for preferring less complex models, as has been discussed elsewhere [10,14,29,46,48,52,[54][55][56][57][58][59][60].…”

Section: Bayes From a Hydrologist's Perspective-the Basic Storymentioning

confidence: 99%

Being Bayesian: Discussions from the Perspectives of Stakeholders and Hydrologists

Ferré

2020

Water

View full text Add to dashboard Cite

Bayes' Theorem is gaining acceptance in hydrology, but it is still far from standard practice to cast hydrologic analyses in a Bayesian context-especially in the realm of hydrologic practice. Three short discussions are presented to encourage more complete adoption of a Bayesian approach. The first, aimed at a stakeholder audience, seeks to explain that an informal Bayesian analysis is the default approach that we all take to any decision made under uncertainty. The second, aimed at a general hydrologist audience, seeks to establish multi-model approaches as the natural choice for Bayesian hydrologic analysis. The goal of this discussion is to provide a bridge from the stakeholder's natural approach to a more formal, quantitative Bayesian analysis. The third discussion is targeted to a more advanced hydrologist audience, suggesting that some elements of hydrologic practice do not yet reflect a Bayesian philosophy. In particular, an example is given that puts Bayes Theory to work to identify optimal observation sets before data are collected.

show abstract

“…Models also become more faithful to the point where synthetic catchments, synthetic aquifers, and virtual observatories no longer seem out of reach (Thomas et al ; Yu et al ). The limiting factor increasingly comes from the exploration and analysis of the resulting simulations (Hermans ). Deep networks might contribute to more systematic interpretation through interpolation and category classification.…”

Section: Deep Learning Prospects For Hydrological Inferencementioning

confidence: 99%

Prospective Interest of Deep Learning for Hydrological Inference

Marçais

Dreuzy

2017

Groundwater

View full text Add to dashboard Cite

International audienceFractured bedrock reservoirs are of socio-economical importance, as they may be used for storage or retrieval of fluids and energy. In particular, the hydromechanical behavior of fractures needs to be understood as it has implications on flow and governs stability issues (e.g. microseismicity). Laboratory, numerical or field experiments have brought considerable insights to this topic. Nevertheless, in situ hydromechanical experiments are relatively uncommon, mainly because of technical and instrumental limitations. Here, we present the early stage development and validation of a novel approach aiming at capturing the integrated hydromechanical behavior of natural fractures. It combines the use of surface tiltmeters to monitor the deformation associated with the periodic pressurization of fractures at depth in crystalline rocks. Periodic injection and withdrawal advantageously avoids mobilizing or extracting significant amounts of fluid and it hinders any risk of reservoir failure. The oscillatory perturbation is intended to: 1) facilitate the recognition of its signature in tilt measurements; 2) vary the hydraulic penetration depth in order to sample different volumes of the fractured bedrock around the inlet, and thereby assess scale effects typical of fractured systems. By stacking tilt signals, we managed to recover small tilt amplitudes associated with pressure-derived fracture deformation. Therewith, we distinguish differences in mechanical properties between the three tested fractures but we show that tilt amplitudes are weakly dependent on pressure penetration depth. Using an elastic model, we obtain fracture stiffness estimates that are consistent with published data. Our results should encourage further improvement of the method

show abstract

Prediction‐Focused Approaches: An Opportunity for Hydrology

Abstract: Article impact statement: Going directly from data to predictions through prediction‐focused approaches: A changing paradigm for solving prediction problems.

Cited by 23 publications

References 17 publications

Uncertainty Quantification of Medium‐Term Heat Storage From Short‐Term Geophysical Experiments Using Bayesian Evidential Learning

Uncertainty Quantification of Medium‐Term Heat Storage From Short‐Term Geophysical Experiments Using Bayesian Evidential Learning

Being Bayesian: Discussions from the Perspectives of Stakeholders and Hydrologists

Prospective Interest of Deep Learning for Hydrological Inference

Contact Info

Product

Resources

About