Observation Quality Control with a Robust Ensemble Kalman Filter

Roh, S.; Genton, Marc G.; Jun, Mikyoung; Szunyogh, Istvan; Hoteit, Ibrahim

doi:10.1175/mwr-d-13-00091.1

Cited by 22 publications

(11 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For the rainfall data with unknown θ, we used state augmentation. For the heavy-tailed data, we also implemented the robust Huber-EnKF proposed in Roh et al (2013). The Huber-EnKF requires a cutoff or clipping value for each observation location, which we selected by sampling 1000 times from the true forecast distribution based on an update consisting of a single observation at the respective location.…”

Section: Simulation Study For Non-gaussian Observationsmentioning

confidence: 99%

“…2: Simulation results for non-Gaussian observations: heavy-tailed data (Example 1) and rainfall data (Example 2). MSPE = mean squared prediction error; CRPS = continuous ranked probability score; exact = posterior distribution obtained using MCMC; PF = SIR particle filter; Huber = robust Huber-EnKF fromRoh et al (2013) Mean squared prediction error (MSPE) versus time for GEnKF and particle filter (PF) updates in the simulated rain example for known κ = 3. Note that time on the x-axis is on a log scale.…”

mentioning

confidence: 99%

See 1 more Smart Citation

Ensemble Kalman Methods for High-Dimensional Hierarchical Dynamic Space-Time Models

Katzfuß

Stroud

Wikle

2019

Journal of the American Statistical Association

View full text Add to dashboard Cite

We propose a new class of filtering and smoothing methods for inference in highdimensional, nonlinear, non-Gaussian, spatio-temporal state-space models. The main idea is to combine the ensemble Kalman filter and smoother, developed in the geophysics literature, with state-space algorithms from the statistics literature. Our algorithms address a variety of estimation scenarios, including on-line and off-line state and parameter estimation. We take a Bayesian perspective, for which the goal is to generate samples from the joint posterior distribution of states and parameters. The key benefit of our approach is the use of ensemble Kalman methods for dimension reduction, which allows inference for high-dimensional state vectors. We compare our methods to existing ones, including ensemble Kalman filters, particle filters, and particle MCMC. Using a real data example of cloud motion and data simulated under a number of nonlinear and non-Gaussian scenarios, we show that our approaches outperform these existing methods.

show abstract

Section: Simulation Study For Non-gaussian Observationsmentioning

confidence: 99%

mentioning

confidence: 99%

Ensemble Kalman Methods for High-Dimensional Hierarchical Dynamic Space-Time Models

Katzfuß

Stroud

Wikle

2019

Journal of the American Statistical Association

View full text Add to dashboard Cite

show abstract

“…This reflects an increased uncertainty in the production rates resulting from less confidence in the accuracy of the geophysical data. Data quality is an important aspect for reservoir interpretation and reservoir history matching purposes, as large errors in the data may lead to overshooting or undershooting in the Kalman filtering update step (Roh et al 2013;Trani, Arts, and Leeuwenburgh 2012;Chen and Oliver 2010;Anderson 2009). This is exemplified in the fact that, for large levels of noise in the data, it becomes rather challenging to distinguish the individual phase components in the reservoir from the gravity and InSAR data.…”

Section: Sensitivity To Observation Errorsmentioning

confidence: 99%

Integrating gravimetric and interferometric synthetic aperture radar data for enhancing reservoir history matching of carbonate gas and volatile oil reservoirs

Katterbauer

Arango

Sun

et al. 2016

Geophysical Prospecting

Self Cite

View full text Add to dashboard Cite

Reservoir history matching is assuming a critical role in understanding reservoir characteristics, tracking water fronts, and forecasting production. While production data have been incorporated for matching reservoir production levels and estimating critical reservoir parameters, the sparse spatial nature of this dataset limits the efficiency of the history matching process. Recently, gravimetry techniques have significantly advanced to the point of providing measurement accuracy in the microgal range and consequently can be used for the tracking of gas displacement caused by water influx. While gravity measurements provide information on subsurface density changes, i.e., the composition of the reservoir, these data do only yield marginal information about temporal displacements of oil and inflowing water. We propose to complement gravimetric data with interferometric synthetic aperture radar surface deformation data to exploit the strong pressure deformation relationship for enhancing fluid flow direction forecasts. We have developed an ensemble Kalman‐filter‐based history matching framework for gas, gas condensate, and volatile oil reservoirs, which synergizes time‐lapse gravity and interferometric synthetic aperture radar data for improved reservoir management and reservoir forecasts. Based on a dual state–parameter estimation algorithm separating the estimation of static reservoir parameters from the dynamic reservoir parameters, our numerical experiments demonstrate that history matching gravity measurements allow monitoring the density changes caused by oil–gas phase transition and water influx to determine the saturation levels, whereas the interferometric synthetic aperture radar measurements help to improve the forecasts of hydrocarbon production and water displacement directions. The reservoir estimates resulting from the dual filtering scheme are on average 20%–40% better than those from the joint estimation scheme, but require about a 30% increase in computational cost.

show abstract

“…These robust statistical methodologies are essential because the data being modelled in fisheries science and management may be subject to large measurement error and the state and observation processes are only approximations with the consequence that there may well be outlying observations or deviating substructures. Much of the recent research on robustness for SSMs (Roh et al, ; Calvet, Czellar, & Ronchetti, ) has focussed on building in robustness against errors in the observation process without much attention to robust estimation of the parameters. The parameters are important here because the interest is in estimates of the proportion mature for a specific year and cohort.…”

Section: Introductionmentioning

confidence: 99%

Robust state space models for estimating fish stock maturities

Cantoni

Flemming

et al. 2015

Can J Statistics

View full text Add to dashboard Cite

Here, we formulate robust state space models (SSMs) and develop inference tools in the context of fisheries science and management. Our prototype model concerns the maturity of fish by age over time, knowledge of which is fundamental to understanding the dynamics and productivity of fish stocks, a key component in fish stock assessment. Our SSM incorporates dynamics over time and yields robust estimates of the proportion of fish mature at various ages for a collection of cohorts of interest. The estimates are obtained using an iterative weighted likelihood where the high-dimensional unobserved dynamics of the SSM and the optimization over the fixed parameters are handled with the Automatic Differentiation Model Builder. The data used to both demonstrate and validate our new approach comes from the annual sampling program of Fisheries and Oceans Canada, the primary scientific and regulatory body responsible for stock assessments. In addition, we carried out a comprehensive simulation study to demonstrate the robustness of our approach to realistic contamination

show abstract

Observation Quality Control with a Robust Ensemble Kalman Filter

Cited by 22 publications

References 31 publications

Ensemble Kalman Methods for High-Dimensional Hierarchical Dynamic Space-Time Models

Ensemble Kalman Methods for High-Dimensional Hierarchical Dynamic Space-Time Models

Integrating gravimetric and interferometric synthetic aperture radar data for enhancing reservoir history matching of carbonate gas and volatile oil reservoirs

Robust state space models for estimating fish stock maturities

Contact Info

Product

Resources

About