Ensemble-based Data Assimilation Research Articles

Formation compaction vs land subsidence to constrain rock compressibility of hydrocarbon reservoirs

Quantification of uncertainty is becoming increasingly important in any general modelling activity. In this study, the ensemble smoother, i.e., an ensemble-based data assimilation algorithm, is used to quantify and reduce the uncertainty associated with the geomechanical parameters of deep hydrocarbon reservoirs. The aim is at estimating the vertical uniaxial compressibility cM of the producing layers by assimilation of: (i) ground or seabed vertical and horizontal displacements measured with InSAR, multibeam surveys, and GPS; and (ii) reservoir deformation obtained from specific well logs (e.g., the radioactive marker technique) and extensometer stations. Usually subsidence measurements are characterized by large datasets (in both time and space) with a relatively low accuracy. Conversely, the compaction monitoring techniques provide more accurate measurements, although their availability is at limited points and over few time intervals. In this contribution, we test the capability of these two types of data to reduce the uncertainty associated to cM for a producing reservoir. Although dealing with a test case application, this investigation originates from the need of properly addressing and explaining the seafloor displacements observed over a real offshore gas field. The numerical tests are carried out with two different conceptual models for cM, based on the common structure of gas fields. The first model considers a compressibility distribution varying with depth and effective vertical stress, but uniformly distributed within the reservoir. In this case, compaction measurements at the reservoir depth result very effective. However, when the reservoir is composed of several compartments bounded by faults and thrusts, the possible heterogeneity of cM among different blocks reduces the effectiveness of compaction measurements in data assimilation algorithms compared to that of surface displacements.

State-of-the-art stochastic data assimilation methods for high-dimensional non-Gaussian problems

This paper compares several commonly used state-of-the-art ensemble-based data assimilation methods in a coherent mathematical notation. The study encompasses different methods that are applicable to high-dimensional geophysical systems, like ocean and atmosphere and provide an uncertainty estimate. Most variants of Ensemble Kalman Filters, Particle Filters and second-order exact methods are discussed, including Gaussian Mixture Filters, while methods that require an adjoint model or a tangent linear formulation of the model are excluded. The detailed description of all the methods in a mathematically coherent way provides both novices and experienced researchers with a unique overview and new insight in the workings and relative advantages of each method, theoretically and algorithmically, even leading to new filters. Furthermore, the practical implementation details of all ensemble and particle filter methods are discussed to show similarities and differences in the filters aiding the users in what to use when. Finally, pseudo-codes are provided for all of the methods presented in this paper.

Dynamically adaptive data-driven simulation of extreme hydrological flows

Hydrological hazards such as storm surges, tsunamis, and rainfall-induced flooding are physically complex events that are costly in loss of human life and economic productivity. Many such disasters could be mitigated through improved emergency evacuation in real-time and through the development of resilient infrastructure based on knowledge of how systems respond to extreme events. Data-driven computational modeling is a critical technology underpinning these efforts. This investigation focuses on the novel combination of methodologies in forward simulation and data assimilation. The forward geophysical model utilizes adaptive mesh refinement (AMR), a process by which a computational mesh can adapt in time and space based on the current state of a simulation. The forward solution is combined with ensemble based data assimilation methods, whereby observations from an event are assimilated into the forward simulation to improve the veracity of the solution, or used to invert for uncertain physical parameters. The novelty in our approach is the tight two-way coupling of AMR and ensemble filtering techniques. The technology is tested using actual data from the Chile tsunami event of February 27, 2010. These advances offer the promise of significantly transforming data-driven, real-time modeling of hydrological hazards, with potentially broader applications in other science domains.

Impact of Representing Model Error in a Hybrid Ensemble-Variational Data Assimilation System for Track Forecast of Tropical Cyclones over the Bay of Bengal

Uncertainties in the numerical weather prediction models are generally not well-represented in ensemble-based data assimilation (DA) systems. The performance of an ensemble-based DA system becomes suboptimal, if the sources of error are undersampled in the forecast system. The present study examines the effect of accounting for model error treatments in the hybrid ensemble transform Kalman filter—three-dimensional variational (3DVAR) DA system (hybrid) in the track forecast of two tropical cyclones viz. Hudhud and Thane, formed over the Bay of Bengal, using Advanced Research Weather Research and Forecasting (ARW-WRF) model. We investigated the effect of two types of model error treatment schemes and their combination on the hybrid DA system; (i) multiphysics approach, which uses different combination of cumulus, microphysics and planetary boundary layer schemes, (ii) stochastic kinetic energy backscatter (SKEB) scheme, which perturbs the horizontal wind and potential temperature tendencies, (iii) a combination of both multiphysics and SKEB scheme. Substantial improvements are noticed in the track positions of both the cyclones, when flow-dependent ensemble covariance is used in 3DVAR framework. Explicit model error representation is found to be beneficial in treating the underdispersive ensembles. Among the model error schemes used in this study, a combination of multiphysics and SKEB schemes has outperformed the other two schemes with improved track forecast for both the tropical cyclones.

A Monte Carlo Background Covariance Localization Method for an Ensemble–Variational Assimilation System

Spurious long-distance correlations in estimates of the background error covariance can deteriorate the performance of ensemble-based data assimilation methods. In this study, a localization method, called Monte Carlo (MC) localization, is presented to remove these correlations. It is particularly useful for use in high-dimensional ensemble–variational data assimilation systems. In this method, raw ensemble members are truncated by multiplying them with functions having compact support. This creates a larger ensemble, in which points spaced farther apart than the size of the compact support have zero correlation. The localized background error covariance is then estimated as the sample covariance of this larger ensemble. It is hypothesized that this localized background error covariance can be approximated by the MC approximation method using a limited set of the truncated ensemble members. This hypothesis is tested here on a grid with 1001 grid points and assuming a Gaussian true background error covariance. It is found that the mean relative error has an upper bound that scales with the inverse square root of the number of truncated ensemble members. In the case studied the size of the support for which the localized background covariance best approximates the true background covariance increases with increasing number of raw ensemble members and is close to 4 times the standard deviation of the Gaussian when 20 raw ensemble members are used. In the Fourier space the localization manifests itself as a convolution resulting in smoothing of the power spectral density of the ensemble members.

Polynomial Surrogates for Open-Channel Flows in Random Steady State

Assessing epistemic uncertainties is considered as a milestone for improving numerical predictions of a dynamical system. In hydrodynamics, uncertainties in input parameters translate into uncertainties in simulated water levels through the shallow water equations. We investigate the ability of generalized polynomial chaos (gPC) surrogate to evaluate the probabilistic features of water level simulated by a 1-D hydraulic model (MASCARET) with the same accuracy as a classical Monte Carlo method but at a reduced computational cost. This study highlights that the water level probability density function and covariance matrix are better estimated with the polynomial surrogate model than with a Monte Carlo approach on the forward model given a limited budget of MASCARET evaluations. The gPC-surrogate performance is first assessed on an idealized channel with uniform geometry and then applied on the more realistic case of the Garonne River (France) for which a global sensitivity analysis using sparse least-angle regression was performed to reduce the size of the stochastic problem. For both cases, Galerkin projection approximation coupled to Gaussian quadrature that involves a limited number of forward model evaluations is compared with least-square regression for computing the coefficients when the surrogate is parameterized with respect to the local friction coefficient and the upstream discharge. The results showed that a gPC-surrogate with total polynomial degree equal to 6 requiring 49 forward model evaluations is sufficient to represent the water level distribution (in the sense of the $\ell _2$ norm), the probability density function and the water level covariance matrix for further use in the framework of data assimilation. In locations where the flow dynamics is more complex due to bathymetry, a higher polynomial degree is needed to retrieve the water level distribution. The use of a surrogate is thus a promising strategy for uncertainty quantification studies in open-channel flows and should be extended to unsteady flows. It also paves the way toward cost-effective ensemble-based data assimilation for flood forecasting and water resource management.

Comparison of polynomial chaos and Gaussian process surrogates for uncertainty quantification and correlation estimation of spatially distributed open-channel steady flows

Data assimilation is widely used to improve flood forecasting capability, especially through parameter inference requiring statistical information on the uncertain input parameters (upstream discharge, friction coefficient) as well as on the variability of the water level and its sensitivity with respect to the inputs. For particle filter or ensemble Kalman filter, stochastically estimating probability density function and covariance matrices from a Monte Carlo random sampling requires a large ensemble of model evaluations, limiting their use in real-time application. To tackle this issue, fast surrogate models based on Polynomial Chaos and Gaussian Process can be used to represent the spatially distributed water level in place of solving the shallow water equations. This study investigates the use of these surrogates to estimate probability density functions and covariance matrices at a reduced computational cost and without the loss of accuracy, in the perspective of ensemble-based data assimilation. This study focuses on 1-D steady state flow simulated with MASCARET over the Garonne River (South-West France). Results show that both surrogates feature similar performance to the Monte-Carlo random sampling, but for a much smaller computational budget; a few MASCARET simulations (on the order of 10-100) are sufficient to accurately retrieve covariance matrices and probability density functions all along the river, even where the flow dynamic is more complex due to heterogeneous bathymetry. This paves the way for the design of surrogate strategies suitable for representing unsteady open-channel flows in data assimilation.

Data assimilation-based reconstruction of urban pollutant release characteristics

The development of efficient methodologies for the reconstruction of pollutant source characteristics in urban areas is a challenging issue due to the complexity of the flow in such environments, requiring the use of expensive computational fluid dynamics (CFD) solvers for the numerical estimation of the flow. In the present study, it is proposed to apply data assimilation (DA) techniques to the identification of both pollutant source characteristics and wind parameters from observations of the gas concentration at discrete locations in an urban environment. A non-intrusive ensemble-based variational DA scheme enhanced by proper orthogonal decomposition (POD) is deployed in conjunction with a Large Eddy Simulation (LES) solver, which provides high-fidelity estimation of the urban flow. The proposed methodology is proved to be able to accurately identify both pollutant source and wind characteristics with a computational cost of O(10) LES simulations. In addition, a sensor placement procedure based on global sensitivity analysis techniques is proposed in order to improve the performances of the assimilation process, which appear to be very sensitive to the choice of the measurement locations, in terms of computational cost and quality of the reconstructed parameters. Despite of several limitations in the use of the present approach for real time applications due to computational costs, this study demonstrates the potentiality of deploying sensitivity analysis and DA methodologies to solve inverse problems for urban flows in conjunction with high-fidelity CFD solvers.

A hybrid approach to generating search subspaces in dynamically constrained 4-dimensional data assimilation

Development and maintenance of the linearized and adjoint code for advanced circulation models is a challenging issue, requiring a significant proportion of total effort in operational data assimilation (DA). The ensemble-based DA techniques provide a derivative-free alternative, which appears to be competitive with variational methods in many practical applications. This article proposes a hybrid scheme for generating the search subspaces in the adjoint-free 4-dimensional DA method (a4dVar) that does not use a predefined ensemble. The method resembles 4dVar in that the optimal solution is strongly constrained by model dynamics and search directions are supplied iteratively using information from the current and previous model trajectories generated in the process of optimization. In contrast to 4dVar, which produces a single search direction from exact gradient information, a4dVar employs an ensemble of directions to form a subspace in order to proceed. In the earlier versions of a4dVar, search subspaces were built using the leading EOFs of either the model trajectory or the projections of the model-data misfits onto the range of the background error covariance (BEC) matrix at the current iteration. In the present study, we blend both approaches and explore a hybrid scheme of ensemble generation in order to improve the performance and flexibility of the algorithm. In addition, we introduce balance constraints into the BEC structure and periodically augment the search ensemble with BEC eigenvectors to avoid repeating minimization over already explored subspaces. Performance of the proposed hybrid a4dVar (ha4dVar) method is compared with that of standard 4dVar in a realistic regional configuration assimilating real data into the Navy Coastal Ocean Model (NCOM). It is shown that the ha4dVar converges faster than a4dVar and can be potentially competitive with 4dvar both in terms of the required computational time and the forecast skill.

Ensemble-based data assimilation for operational flood forecasting – On the merits of state estimation for 1D hydrodynamic forecasting through the example of the “Adour Maritime” river

This study presents the implementation and the merits of an Ensemble Kalman Filter (EnKF) algorithm with an inflation procedure on the 1D shallow water model MASCARET in the framework of operational flood forecasting on the “Adour Maritime” river (South West France). In situ water level observations are sequentially assimilated to correct both water level and discharge. The stochastic estimation of the background error statistics is achieved over an ensemble of MASCARET integrations with perturbed hydrological boundary conditions. It is shown that the geometric characteristics of the network as well as the hydrological forcings and their temporal variability have a significant impact on the shape of the univariate (water level) and multivariate (water level and discharge) background error covariance functions and thus on the EnKF analysis. The performance of the EnKF algorithm is examined for observing system simulation experiments as well as for a set of eight real flood events (2009–2014). The quality of the ensemble is deemed satisfactory as long as the forecast lead time remains under the transfer time of the network, when perfect hydrological forcings are considered. Results demonstrate that the simulated hydraulic state variables can be improved over the entire network, even where no data are available, with a limited ensemble size and thus a computational cost compatible with operational constraints. The improvement in the water level Root-Mean-Square Error obtained with the EnKF reaches up to 88% at the analysis time and 40% at a 4-h forecast lead time compared to the standalone model.

Design and Implementation of a GSI-Based Convection-Allowing Ensemble-Based Data Assimilation and Forecast System for the PECAN Field Experiment. Part II: Overview and Evaluation of a Real-Time System

Abstract Multiscale ensemble-based data assimilation and forecasts were performed in real time during the Plains Elevated Convection At Night (PECAN) field experiment. A 20-member ensemble of forecasts at 4-km grid spacing was initialized daily at both 1300 and 1900 UTC, together with a deterministic forecast at 1-km grid spacing initialized at 1300 UTC. The configuration of the GSI-based data assimilation and forecast system was guided by results presented in Part I of this two-part study. The present paper describes the implementation of the real-time system and the extensive forecast products that were generated to support the unique interests of PECAN researchers. Subjective and objective verification of the real-time forecasts from 1 June through 15 July 2015 is conducted, with an emphasis on nocturnal mesoscale convective systems (MCSs), nocturnal convective initiation (CI), nocturnal low-level jets (LLJs), and bores on the nocturnal stable layer. Verification of nocturnal precipitation during overnight hours, a proxy for MCSs, shows both greater skill and spread for the 1300 UTC forecasts than the 1900 UTC forecasts. Verification against observed soundings reveals that the forecast LLJs systematically peak, veer, and dissipate several hours before the observations. Comparisons with bores that passed over an Atmospheric Emitted Radiance Interferometer reveal an ability to predict borelike features that is greatly improved at 1-km, compared with 4-km, grid spacing. Objective verification of forecast CI timing reveals strong sensitivity to the PBL scheme but an overall unbiased ensemble.

Accelerating volcanic ash data assimilation using a mask-state algorithm based on an ensemble Kalman filter: a case study with the LOTOS-EUROS model (version 1.10)

Abstract. In this study, we investigate a strategy to accelerate the data assimilation (DA) algorithm. Based on evaluations of the computational time, the analysis step of the assimilation turns out to be the most expensive part. After a study of the characteristics of the ensemble ash state, we propose a mask-state algorithm which records the sparsity information of the full ensemble state matrix and transforms the full matrix into a relatively small one. This will reduce the computational cost in the analysis step. Experimental results show the mask-state algorithm significantly speeds up the analysis step. Subsequently, the total amount of computing time for volcanic ash DA is reduced to an acceptable level. The mask-state algorithm is generic and thus can be embedded in any ensemble-based DA framework. Moreover, ensemble-based DA with the mask-state algorithm is promising and flexible, because it implements exactly the standard DA without any approximation and it realizes the satisfying performance without any change in the full model.

Evaluation of a data-driven wildland fire spread forecast model with spatially-distributed parameter estimation in simulations of the FireFlux I field-scale experiment

The general objective of this research is to develop a prototype data-driven wildland fire spread simulator, called FIREFLY, using an ensemble-based data assimilation approach with the objective to forecast the location and speed of the fire. The specific focus of the present study is on evaluating the new features of FIREFLY at field scale in a controlled grassland fire experiment known as FireFlux I. FIREFLY features the following components: an Eulerian front-tracking solver that treats the fire as a propagating front and uses Rothermel's model for the rate of spread (ROS); a series of observations of the fire front position (based here on high-resolution fireline data previously generated by validated numerical simulations); and a data assimilation algorithm based on an ensemble Kalman filter configured in a parameter estimation mode to address model bias and uncertainties in the input data to the ROS model. In this work, FIREFLY is modified to allow for an estimation of spatially-distributed surface wind speed and direction. To generate a reliable ensemble and ensure an accurate correction, the ensemble Kalman filter requires sampling truncated probability density functions as well as localizing, i.e., dynamically selecting the areas where the wind parameters are corrected. Results show that the spatialized parameter estimation approach allows for a successful reconstruction of observed fireline position and shape as well as a substantial improvement in the forecast performance compared to the standalone fire spread model. Results also show that the inferred wind parameters may not be accurate and should be viewed as effective values that incorporate multiple sources of uncertainties. Developing a better representation of fire-wind interactions is thus viewed as a key aspect to improve the FIREFLY forecast capability.

Design and Implementation of a GSI-Based Convection-Allowing Ensemble Data Assimilation and Forecast System for the PECAN Field Experiment. Part I: Optimal Configurations for Nocturnal Convection Prediction Using Retrospective Cases

Abstract A real-time GSI-based and ensemble-based data assimilation (DA) and forecast system was implemented at the University of Oklahoma during the 2015 Plains Elevated Convection at Night (PECAN) experiment. Extensive experiments on the configuration of the cycled DA and on both the DA and forecast physics ensembles were conducted using retrospective cases to optimize the system design for nocturnal convection. The impacts of radar DA between 1200 and 1300 UTC, as well as the frequency and number of DA cycles and the DA physics configuration, extend through the following night. Ten-minute cycling of radar DA leads to more skillful forecasts than both more and less frequent cycling. The Thompson microphysics scheme for DA better analyzes the effects of morning convection on environmental moisture than WSM6, which improves the convection forecast the following night. A multi-PBL configuration during DA leads to less skillful short-term forecasts than even a relatively poorly performing single-PBL scheme. Deterministic and ensemble forecast physics configurations are also evaluated. Thompson microphysics and the Mellor–Yamada–Nakanishi–Niino (MYNN) PBL provide the most skillful nocturnal precipitation forecasts. A well thought out multiphysics configuration is shown to provide advantages over evenly distributing three of the best-performing microphysics and PBL schemes or a fixed MYNN/Thompson ensemble. This is shown using objective and subjective verification of precipitation and nonprecipitation variables, including convective initiation. Predictions of the low-level jet are sensitive to the PBL scheme, with the best scheme being variable and time dependent. These results guided the implementation and verification of a real-time ensemble DA and forecast system for PECAN.

Ring current pressure estimation with RAM‐SCB using data assimilation and Van Allen Probe flux data

AbstractCapturing and subsequently modeling the influence of tail plasma injections on the inner magnetosphere is important for understanding the formation and evolution of the ring current. In this study, the ring current distribution is estimated with the Ring Current‐Atmosphere Interactions Model with Self‐Consistent Magnetic field (RAM‐SCB) using, for the first time, data assimilation techniques and particle flux data from the Van Allen Probes. The state of the ring current within the RAM‐SCB model is corrected via an ensemble based data assimilation technique by using proton flux from one of the Van Allen Probes, to capture the enhancement of the ring current following an isolated substorm event on 18 July 2013. The results show significant improvement in the estimation of the ring current particle distributions in the RAM‐SCB model, leading to better agreement with observations. This newly implemented data assimilation technique in the global modeling of the ring current thus provides a promising tool to improve the characterization of particle distribution in the near‐Earth regions.

Ensemble-based approximation of observation impact using an observation-based verification metric

Knowledge on the contribution of observations to forecast accuracy is crucial for the refinement of observing and data assimilation systems. Several recent publications highlighted the benefits of efficiently approximating this observation impact using adjoint methods or ensembles. This study proposes a modification of an existing method for computing observation impact in an ensemble-based data assimilation and forecasting system and applies the method to a pre-operational, convective-scale regional modelling environment. Instead of the analysis, the modified approach uses observation-based verification metrics to mitigate the effect of correlation between the forecast and its verification norm. Furthermore, a peculiar property in the distribution of individual observation impact values is used to define a reliability indicator for the accuracy of the impact approximation. Applying this method to a 3-day test period shows that a well-defined observation impact value can be approximated for most observation types and the reliability indicator successfully depicts where results are not significant.

To what extent is your data assimilation scheme designed to find the posterior mean, the posterior mode or something else?

Recently there has been a surge in interest in coupling ensemble-based data assimilation methods with variational methods (commonly referred to as 4DVar). Here we discuss a number of important differences between ensemble-based and variational methods that ought to be considered when attempting to fuse these methods. We note that the Best Linear Unbiased Estimate (BLUE) of the posterior mean over a data assimilation window can only be delivered by data assimilation schemes that utilise the 4-dimensional (4D) forecast covariance of a prior distribution of non-linear forecasts across the data assimilation window. An ensemble Kalman smoother (EnKS) may be viewed as a BLUE approximating data assimilation scheme. In contrast, we use the dual form of 4DVar to show that the most likely non-linear trajectory corresponding to the posterior mode across a data assimilation window can only be delivered by data assimilation schemes that create counterparts of the 4D prior forecast covariance using a tangent linear model. Since 4DVar schemes have the required structural framework to identify posterior modes, in contrast to the EnKS, they may be viewed as mode approximating data assimilation schemes. Hence, when aspects of the EnKS and 4DVar data assimilation schemes are blended together in a hybrid, one would like to be able to understand how such changes would affect the mode- or mean-finding abilities of the data assimilation schemes. This article helps build such understanding using a series of simple examples. We argue that this understanding has important implications to both the interpretation of the hybrid state estimates and to their design.

Localization and regularization for iterative ensemble smoothers

Ensemble-based data assimilation methods have recently become popular for solving reservoir history matching problems, but because of the practical limitation on ensemble size, using localization is necessary to reduce the effect of sampling error and to increase the degrees of freedom for incorporating large amounts of data. Local analysis in the ensemble Kalman filter has been used extensively for very large models in numerical weather prediction. It scales well with the model size and the number of data and is easily parallelized. In the petroleum literature, however, iterative ensemble smoothers with localization of the Kalman gain matrix have become the state-of-the-art approach for ensemble-based history matching. By forming the Kalman gain matrix row-by-row, the analysis step can also be parallelized. Localization regularizes updates to model parameters and state variables using information on the distance between the these variables and the observations. The truncation of small singular values in truncated singular value decomposition (TSVD) at the analysis step provides another type of regularization by projecting updates to dominant directions spanned by the simulated data ensemble. Typically, the combined use of localization and TSVD is necessary for problems with large amounts of data. In this paper, we compare the performance of Kalman gain localization to two forms of local analysis for parameter estimation problems with nonlocal data. The effect of TSVD with different localization methods and with the use of iteration is also analyzed. With several examples, we show that good results can be obtained for all localization methods if the localization range is chosen appropriately, but the optimal localization range differs for the various methods. In general, for local analysis with observation taper, the optimal range is somewhat shorter than the optimal range for other localization methods. Although all methods gave equivalent results when used in an iterative ensemble smoother, the local analysis methods generally converged more quickly than Kalman gain localization when the amount of data is large compared to ensemble size.

Investigation on Principal Component Analysis Parameterizations for History Matching Channelized Facies Models with Ensemble-Based Data Assimilation

Preserving plausible geological features when updating facies models is still one of the main challenges with ensemble-based history matching. This is particularly evident for fields with complex geological description (e.g., fluvial channels). There is an impressive amount of research published in the last few years about this subject. However, it appears that there is no definitive solution and both, academia and industry, are looking for practical and robust methods. Among the parameterizations traditionally investigated for history matching, the principal component analysis (PCA) of the prior covariance matrix is an efficient alternative to represent models described by two-point statistics. However, there are some recent developments extending PCA-based parameterizations for models described by multiple-point statistics. The first part of this paper presents an investigation on PCA-based schemes for parameterizing channelized facies models for history matching with ensemble-based methods. The following parameterizations are tested: standard PCA, two alternative implementations of kernel PCA and optimization-based PCA. In the second part of the paper, the optimization-based PCA is modified to allow the use of covariance localization and adapted for simultaneously adjusting the facies type and the permeability values within each facies when history matching production data with an ensemble-based method.

Integrated Work Flow of Preserving Facies Realism in History Matching: Application to the Brugge Field

Summary Data assimilation with ensemble-based inversion methods was successfully applied for parameter estimation in reservoir models. However, in certain complex-reservoir models, it remains challenging to estimate the model parameters and to preserve the geological realism simultaneously. In particular, when handling special-reservoir model parameters such as facies types concerning fluvial channels, one must realize that geological realism becomes one of the key concerns. The main objective of this work is to address these issues for a complex field with a newly extended version of a recently proposed facies-parameterization approach coupled with an ensemble-based data assimilation method. The proposed workflow combines the new facies parameterization and the adaptive gaussian mixture (AGM) filter into the data assimilation framework for channelized reservoirs. To handle discrete-facies parameters, we combine probability maps and truncated Gaussian fields to obtain a continuous parameterization of the facies fields. For the data assimilation, we use the AGM filter, which is an efficient history matching approach that incorporates a resampling routine that allows us to regenerate facies fields with information from the updated probability maps. This work flow is evaluated, for the first time, on a complex field case—the Brugge field. This reservoir model consists of layers with complex channelized structures and layers characterized by reservoir properties generated with variograms. With limited prior knowledge on the facies model, this work flow is shown to be able to preserve the channel continuity while reducing the reservoir model uncertainty with AGM. When applied to a complex reservoir, the proposed work flow provides a geologically consistent and realistic reservoir model that leads to improved capability of predicting subsurface flow behaviors.