Data Assimilation Component Research Articles

A new global daily sea‐surface temperature analysis system at Environment and Climate Change Canada

AbstractA new global daily sea‐surface temperature (SST) analysis system has been developed at Environment and Climate Change Canada (ECCC). All components of the new SST analysis system are implemented within the Modular and Integrated Data Assimilation System (MIDAS) software. MIDAS is already used for the data assimilation component of the main operational numerical weather prediction (NWP) systems at ECCC. The new SST analysis system, integrated together with the global sea‐ice analysis, will be part of the combined ocean surface analysis used for all operational prediction systems at ECCC. The data assimilation method used to compute the new SST analyses is two‐dimensional variational method with a diffusion operator for representing the horizontal background‐error correlations. A new algorithm for satellite data bias estimation has also been developed employing gridded bias estimates computed from a spatial averaging of the differences between collocated satellite and in‐situ data. New algorithms for quality control and thinning of satellite data have also been implemented, making each type of observational dataset more evenly distributed over the globe. The performance of the new SST system is examined relative to the current operational SST system by using independent data. The impact of using the new SST analysis within NWP and ocean prediction systems is also evaluated. When compared with the operational system currently in use, the experiments employing the new SST analysis system produce a nearly neutral impact on the NWP and ocean prediction systems. This validation of the new system is an important first step towards the ability to use MIDAS to perform ensemble‐based three‐dimensional ocean and coupled ocean‐ice–atmosphere data assimilation.

Improving ensemble data assimilation through Probit-space Ensemble Size Expansion for Gaussian Copulas (PESE-GC)

Abstract. Small forecast ensemble sizes (< 100) are common in the ensemble data assimilation (EnsDA) component of geophysical forecast systems, thus limiting the error-constraining power of EnsDA. This study proposes an efficient and embarrassingly parallel method to generate additional ensemble members: the Probit-space Ensemble Size Expansion for Gaussian Copulas (PESE-GC; “peace gee see”). Such members are called “virtual members”. PESE-GC utilizes the users' knowledge of the marginal distributions of forecast model variables. Virtual members can be generated from any (potentially non-Gaussian) multivariate forecast distribution that has a Gaussian copula. PESE-GC's impact on EnsDA is evaluated using the 40-variable Lorenz 1996 model, several EnsDA algorithms, several observation operators, a range of EnsDA cycling intervals, and a range of forecast ensemble sizes. Significant improvements to EnsDA (p<0.01) are observed when either (1) the forecast ensemble size is small (≤20 members), (2) the user selects marginal distributions that improve the forecast model variable statistics, and/or (3) the rank histogram filter is used with non-parametric priors in high-forecast-spread situations. These results motivate development and testing of PESE-GC for EnsDA with high-order geophysical models.

The importance of data assimilation components for initial conditions and subsequent error growth

The importance of data assimilation components for initial conditions and subsequent error growth

Maximizing the potential of numerical weather prediction models: lessons learned from combining high-performance computing and cloud computing

Abstract. To promote cloud and HPC computing, GRAPEVINE* project objectives include using these tools along with open data sources to provide a reusable IT service. In this service a predictive model based on Machine learning (ML) techniques is created with the aim of preventing and controlling grape vine diseases in the wine cultivation sector. Aside from the predictive ML, meteorological forecasts are crucial input to train the ML models and on a second step to be used as input for the operational prediction of grapevine diseases. To this end, the Weather and Research Forecasting model (WRF) has been deployed in CESGA's HPC infrastructure to produce medium-range and sub-seasonal forecasts for the targeted pilot areas (Greece and Spain). The data assimilation component of WRF – WRFDA – has been also introduced for improving the initial conditions of the WRF model by assimilating observations from weather stations and satellite precipitation products (Integrated Multi-satellitE Retrieval for GPM – IMERG). This methodology for assimilation was developed during STARGATE* project, allowing the testing of the methodology in the operational service of GRAPEVINE. The operational production of the forecasts is achieved by the cloudify orchestrator on a Kubernetes cluster. The connections between the Kubernetes cluster and the HPC infrastructure, where the model resides, is achieved with the croupier plugin of cloudify. Blueprints that encapsule the workflows of the meteorological model and its dependencies were created. The instances of the blueprints (deployments) were created automatically to produce operationally weather forecasts and they were made available to the ML models via a THREDDS server. Valuable lessons were learned with regards the automation of the process and the coupling with the HPC in terms of reservations and operational production.

Assessment of the data assimilation framework for the Rapid Refresh Forecast System v0.1 and impacts on forecasts of a convective storm case study

Abstract. The Rapid Refresh Forecast System (RRFS) is currently under development and aims to replace the National Centers for Environmental Prediction (NCEP) operational suite of regional- and convective-scale modeling systems in the next upgrade. In order to achieve skillful forecasts comparable to the current operational suite, each component of the RRFS needs to be configured through exhaustive testing and evaluation. The current data assimilation component uses the hybrid three-dimensional ensemble–variational data assimilation (3DEnVar) algorithm in the Gridpoint Statistical Interpolation (GSI) system. In this study, various data assimilation algorithms and configurations in GSI are assessed for their impacts on RRFS analyses and forecasts of a squall line over Oklahoma on 4 May 2020. A domain of 3 km horizontal grid spacing is configured, and hourly update cycles are performed using initial and lateral boundary conditions from the 3 km grid High-Resolution Rapid Refresh (HRRR). Results show that a baseline RRFS run is able to represent the observed convection, although with stronger cells and large location errors. With data assimilation, these errors are reduced, especially in the 4 and 6 h forecasts using 75 % of the ensemble background error covariance (BEC) and 25 % of the static BEC with the supersaturation removal function activated in GSI. Decreasing the vertical ensemble localization radius from 3 layers to 1 layer in the first 10 layers of the hybrid analysis results in overall less skillful forecasts. Convection is greatly improved when using planetary boundary layer pseudo-observations, especially at 4 h forecast, and the bias of the 2 h forecast of temperature is reduced below 800 hPa. Lighter hourly accumulated precipitation is predicted better when using 100 % ensemble BEC in the first 4 h forecast, but heavier hourly accumulated precipitation is better predicted with 75 % ensemble BEC. Our results provide insight into the current capabilities of the RRFS data assimilation system and identify configurations that should be considered as candidates for the first version of RRFS.

An online ensemble coupled data assimilation capability for the Community Earth System Model: system design and evaluation

Abstract. The Community Earth System Model (CESM) developed by the National Center for Atmospheric Research (NCAR) has been used worldwide for climate studies. This study extends the efforts of CESM development to include an online (i.e., in-core) ensemble coupled data assimilation system (CESM-ECDA) to enhance CESM's capability for climate predictability studies and prediction applications. The CESM-ECDA system consists of an online atmospheric data assimilation (ADA) component implemented in both the finite-volume and spectral-element dynamical cores and an online ocean data assimilation (ODA) component. In ADA, surface pressures (Ps) are assimilated, while in ODA, gridded sea surface temperature (SST) and ocean temperature and salinity profiles at real Argo locations are assimilated. The system has been evaluated within a perfect twin experiment framework, showing significantly reduced errors of the model atmosphere and ocean states through “observation” constraints by ADA and ODA. The weakly coupled data assimilation (CDA) in which both the online ADA and ODA are conducted during the coupled model integration shows smaller errors of air–sea fluxes than the single ADA and ODA, facilitating the future utilization of cross-covariance between the atmosphere and ocean at the air–sea interface. A 3-year CDA reanalysis experiment is also implemented by assimilating Ps, SST and ocean temperature and salinity profiles from the real world spanning the period 1978 to 1980 using 12 ensemble members. The success of the online CESM-ECDA system is the first step to implementing a high-resolution long-term climate reanalysis once the algorithm efficiency is much improved.

Development of Coupled Data Assimilation With the BCC Climate System Model: Highlighting the Role of Sea‐Ice Assimilation for Global Analysis

AbstractThe coupled data assimilation (CDA) system consisting of ocean, sea‐ice, and atmosphere data assimilation components with the Beijing Climate Center (BCC) Climate System Model has been developed to provide reliable analyses of the atmosphere, ocean, and sea‐ice states. It incorporates ocean temperature/salinity profiles, sea surface temperature, sea level height, and sea‐ice concentration observations at a daily frequency, and atmosphere reanalysis at a 6‐hourly frequency. Results show that the system is capable of realistically reproducing the climatology and variability of ocean, sea‐ice, and atmosphere. The performances in analyzing ocean component are comparable to those of well‐known ocean reanalyses that were once used to initialize the BCC model for climate predictions. A series of experiments with and without sea‐ice observations in the CDA framework are designed to explore the role of sea‐ice data assimilation (DA). The addition of sea‐ice DA exerts very small influence to the analysis of upper ocean temperature over the Arctic area, but leads to an evident reduction of temperature error in the upper 1000 m of ocean south of 60°S. Particularly, only the inclusion of sea‐ice DA can make the ocean/ocean‐atmosphere DA effective in providing skillful analysis in the high‐latitude Southern Ocean. Furthermore, we find that only on the basis of ocean DA, can the addition of sea‐ice concentration assimilation improve the analysis of tropical tropospheric atmosphere, along with a better analysis of mid‐ and high‐latitude stratospheric atmosphere. These results address the importance of coordination of sea‐ice observations and ocean observations in CDA.

The impact of a new high-resolution ocean model on the Met Office North-West European Shelf forecasting system

Abstract. The North-West European Shelf ocean forecasting system has been providing oceanographic products for the European continental shelf seas for more than 15 years. In that time, several different configurations have been implemented, updating the model and the data assimilation components. The latest configuration to be put in operation, an eddy-resolving model at 1.5 km (AMM15), replaces the 7 km model (AMM7) that has been used for 8 years to deliver forecast products to the Copernicus Marine Environment Monitoring Service and its precursor projects. This has improved the ability to resolve the mesoscale variability in this area. An overview of this new system and its initial validation is provided in this paper, highlighting the differences with the previous version. Validation of the model with data assimilation is based on the results of 2 years (2016–2017) of trial experiments run with the low- and high-resolution systems in their operational configuration. The 1.5 km system has been validated against observations and the low-resolution system, trying to understand the impact of the high resolution on the quality of the products delivered to the users. Although the number of observations is a limiting factor, especially for the assessment of model variables like currents and salinity, the new system has been proven to be an improvement in resolving fine-scale structures and variability and provides more accurate information on the major physical variables, like temperature, salinity, and horizontal currents. AMM15 improvements are evident from the validation against high-resolution observations, available in some selected areas of the model domain. However, validation at the basin scale and using daily means penalized the high-resolution system and does not reflect its superior performance. This increment in resolution also improves the capabilities to provide marine information closer to the coast even if the coastal processes are not fully resolved by the model.

Adaptive Probability Thresholding in Automated Ice and Open Water Detection From RADARSAT-2 Images

In this letter, we introduce adaptive probability thresholding in addition to our previously developed technique for automated detection of ice and open water from RADARSAT-2 ScanSAR dual-polarization HH–HV images. Situations where the probability threshold needs to be modified were identified based on the analysis of misclassified ice and water samples when the static probability threshold of 0.95 is applied. We found that with the use of the proposed approach, the fraction of misclassified ice samples decreased from 0.98% to 0.24% and the fraction of misclassified water samples decreased from 0.35% to 0.09% in the most clean verification scenario against Canadian Ice Service Image Analysis pure ice and water data, while the fraction of correctly classified ice and water samples did not decrease appreciably, from 72.2% to 65.4%. The developed approach will be implemented as a part of the data assimilation component of the operational Environment and Climate Change Canada Regional Ice-Ocean Prediction System.

Exploratory studies into seasonal flow forecasting potential for large lakes

Abstract. In seasonal flow forecasting applications, one factor which can help predictability is a significant hydrological response time between rainfall and flows. On account of storage influences, large lakes therefore provide a useful test case although, due to the spatial scales involved, there are a number of modelling challenges related to data availability and understanding the individual components in the water balance. Here some possible model structures are investigated using a range of stochastic regression and transfer function techniques with additional insights gained from simple analytical approximations. The methods were evaluated using records for two of the largest lakes in the world – Lake Malawi and Lake Victoria – with forecast skill demonstrated several months ahead using water balance models formulated in terms of net inflows. In both cases slight improvements were obtained for lead times up to 4–5 months from including climate indices in the data assimilation component. The paper concludes with a discussion of the relevance of the results to operational flow forecasting systems for other large lakes.

Automated Detection of Ice and Open Water From Dual-Polarization RADARSAT-2 Images for Data Assimilation

In this paper, we present a new technique for automated detection of ice and open water from RADARSAT-2 ScanSAR dual-polarization HH-HV images. Probability of the presence of ice within 2.05 km $\times2.05$ km areas is modeled using a form of logistic regression as a function of the difference between the wind speeds estimated from synthetic aperture radar (SAR) data and those obtained from numerical weather prediction short-term forecasts, the spatial correlation between HH and HV backscatter signals, and the spatial standard deviation of the wind speed estimated from SAR. The resulting ice probability model was built based on thousands of SAR images and corresponding Canadian Ice Service (CIS) Image Analysis products covering all seasons and all Canadian and adjacent Arctic regions being monitored by CIS. Extensive verification of the proposed technique was conducted for an entire year (2013) against independent Image Analysis products and Interactive Multisensor Snow and Ice Mapping System ice extent products. Using a probability threshold of 0.95, 72.2% of the retrievals were classified as either ice or open water with an accuracy of 99.2% in the most clean verification scenario against Image Analysis pure ice and water data. The ability to obtain such a large number of retrievals with a very high accuracy makes it feasible to assimilate the resulting retrievals in an ice prediction system. Consequently, the developed ice/water retrieval technique will be implemented as a part of the data assimilation component of the operational Environment and Climate Change Canada Regional Ice-Ocean Prediction System.

The CO5 configuration of the 7 km Atlantic Margin Model: large-scale biases and sensitivity to forcing, physics options and vertical resolution

Abstract. We describe the physical model component of the standard Coastal Ocean version 5 configuration (CO5) of the European north-west shelf (NWS). CO5 was developed jointly between the Met Office and the National Oceanography Centre. CO5 is designed with the seamless approach in mind, which allows for modelling of multiple timescales for a variety of applications from short-range ocean forecasting to climate projections. The configuration constitutes the basis of the latest update to the ocean and data assimilation components of the Met Office's operational Forecast Ocean Assimilation Model (FOAM) for the NWS. A 30.5-year non-assimilating control hindcast of CO5 was integrated from January 1981 to June 2012. Sensitivity simulations were conducted with reference to the control run. The control run is compared against a previous non-assimilating Proudman Oceanographic Laboratory Coastal Ocean Modelling System (POLCOMS) hindcast of the NWS. The CO5 control hindcast is shown to have much reduced biases compared to POLCOMS. Emphasis in the system description is weighted to updates in CO5 over previous versions. Updates include an increase in vertical resolution, a new vertical coordinate stretching function, the replacement of climatological riverine sources with the pan-European hydrological model E-HYPE, a new Baltic boundary condition and switching from directly imposed atmospheric model boundary fluxes to calculating the fluxes within the model using a bulk formula. Sensitivity tests of the updates are detailed with a view toward attributing observed changes in the new system from the previous system and suggesting future directions of research to further improve the system.

Demonstrating the value of larger ensembles in forecasting physical systems

Ensemble simulation propagates a collection of initial states forward in time in a Monte Carlo fashion. Depending on the fidelity of the model and the properties of the initial ensemble, the goal of ensemble simulation can range from merely quantifying variations in the sensitivity of the model all the way to providing actionable probability forecasts of the future. Whatever the goal is, success depends on the properties of the ensemble, and there is a longstanding discussion in meteorology as to the size of initial condition ensemble most appropriate for Numerical Weather Prediction. In terms of resource allocation: how is one to divide finite computing resources between model complexity, ensemble size, data assimilation and other components of the forecast system. One wishes to avoid undersampling information available from the model's dynamics, yet one also wishes to use the highest fidelity model available. Arguably, a higher fidelity model can better exploit a larger ensemble; nevertheless it is often suggested that a relatively small ensemble, say ~16 members, is sufficient and that larger ensembles are not an effective investment of resources. This claim is shown to be dubious when the goal is probabilistic forecasting, even in settings where the forecast model is informative but imperfect. Probability forecasts for a ‘simple’ physical system are evaluated at different lead times; ensembles of up to 256 members are considered. The pure density estimation context (where ensemble members are drawn from the same underlying distribution as the target) differs from the forecasting context, where one is given a high fidelity (but imperfect) model. In the forecasting context, the information provided by additional members depends also on the fidelity of the model, the ensemble formation scheme (data assimilation), the ensemble interpretation and the nature of the observational noise. The effect of increasing the ensemble size is quantified by its relative information content (in bits) using a proper skill score. Doubling the ensemble size is demonstrated to yield a non-trivial increase in the information content (forecast skill) for an ensemble with well over 16 members; this result stands in forecasting a mathematical system and a physical system. Indeed, even at the largest ensemble sizes considered (128 and 256), there are lead times where the forecast information is still increasing with ensemble size. Ultimately, model error will limit the value of ever larger ensembles. No support is found, however, for limiting design studies to the sizes commonly found in seasonal and climate studies. It is suggested that ensemble size be considered more explicitly in future design studies of forecast systems on all time scales.

Impacts of a 4-dimensional variational data assimilation in a coastal ocean model of southern Tyrrhenian Sea

The impact of the assimilation of satellite sea surface height, sea surface temperature and surface velocity fields observed by a set of high-frequency (HF) radars is studied using a three-dimensional ocean circulation model configured for the southern Tyrrhenian Sea. The study period is October–December 2010 covered by a large number of data. The nonlinear model is based on the Regional Ocean Modeling System (ROMS) and the data assimilation component on the four-dimensional variational (4D-Var) algorithm.Assimilation proceeds in a series of 7-day windows, providing an analysis solution in each window.The assimilation of surface velocity combined with other observations has more utility in recovering the density fields based on the theory of geostrophic adjustment and a strong impact both on near-surface horizontal volume fluxes and subsurface flows, constraining surface geostrophic currents in the area not covered by the HF radar data.The adjoint of the 4D-Var gain matrix was used to quantify the impact of individual observations and observation platforms on different aspects of the 4D-Var circulation estimates during both the analysis and subsequent forecast cycles. In this study, we focus on the alongshore transport of the surface and intermediate waters in the eastern zone of southern Tyrrhenian Sea. The majority of the observations available during any given analysis cycle are from HF radar, and on average these data, together with SSH data, exert the largest controlling influence on the analysis increments of coastal transport. Also, observations from satellite platforms in the form of SST have a considerable impact on analyses and forecasts of coastal transport, even though these observations represent a relatively small fraction of the available data at any particular time.During 4D-Var, the observations are used to correct for uncertainties in the model control variables, namely, the initial conditions, surface forcing, and open boundary conditions. It is found that correcting for uncertainties in the initial conditions and only secondarily in the boundary conditions has the largest impact on the analysis increments in alongshore transport.Finally, we note that both the control vector and the observation impact calculations are a useful way for monitoring the performance of the data assimilation system, as well as quantifying the impact of the observations on the circulation estimates.

Predicting the resilience and recovery of aquatic systems: A framework for model evolution within environmental observatories

Maintaining the health of aquatic systems is an essential component of sustainable catchment management, however, degradation of water quality and aquatic habitat continues to challenge scientists and policy-makers. To support management and restoration efforts aquatic system models are required that are able to capture the often complex trajectories that these systems display in response to multiple stressors. This paper explores the abilities and limitations of current model approaches in meeting this challenge, and outlines a strategy based on integration of flexible model libraries and data from observation networks, within a learning framework, as a means to improve the accuracy and scope of model predictions. The framework is comprised of a data assimilation component that utilizes diverse data streams from sensor networks, and a second component whereby model structural evolution can occur once the model is assessed against theoretically relevant metrics of system function. Given the scale and transdisciplinary nature of the prediction challenge, network science initiatives are identified as a means to develop and integrate diverse model libraries and workflows, and to obtain consensus on diagnostic approaches to model assessment that can guide model adaptation. We outline how such a framework can help us explore the theory of how aquatic systems respond to change by bridging bottom-up and top-down lines of enquiry, and, in doing so, also advance the role of prediction in aquatic ecosystem management.

Implementation of Deterministic Weather Forecasting Systems Based on Ensemble–Variational Data Assimilation at Environment Canada. Part II: The Regional System

Abstract The modifications to the data assimilation component of the Regional Deterministic Prediction System (RDPS) implemented at Environment Canada operations during the fall of 2014 are described. The main change is the replacement of the limited-area four-dimensional variational data assimilation (4DVar) algorithm for the limited-area analysis and the associated three-dimensional variational data assimilation (3DVar) scheme for the synchronous global driver analysis by the four-dimensional ensemble–variational data assimilation (4DEnVar) scheme presented in the first part of this study. It is shown that a 4DEnVar scheme using global background-error covariances can provide RDPS forecasts that are slightly improved compared to the previous operational approach, particularly during the first 24 h of the forecasts and in the summertime convective regime. Further forecast improvements were also made possible by upgrades in the assimilated observational data and by introducing the improved global analysis presented in the first part of this study in the RDPS intermittent cycling strategy. The computational savings brought by the 4DEnVar approach are also discussed.

An Ensemble Algorithm Based Component for Geomagnetic Data Assimilation

Geomagnetic data assimilation is one of the most recent developments in geomagnetic studies. It combines geodynamo model outputs and surface geomagnetic observations to provide more accurate estimates of the core dynamic state and provide accurate geomagnetic secular variation forecasting. To facilitate geomagnetic data assimilation studies, we develop a standalone data assimilation component for the geomagnetic community. This component is used to calculate the forecast error covariance matrices and the gain matrix from a given geodynamo solution, which can then be used for sequential geomagnetic data assimilation. This component is very flexible and can be executed independently. It can also be easily integrated with arbitrary dynamo models.

The impact of the assimilation of Aquarius sea surface salinity data in the GEOS ocean data assimilation system

AbstractOcean salinity and temperature differences drive thermohaline circulation. These properties also play a key role in the ocean‐atmosphere coupling. With the availability of L‐band space‐borne observations, it becomes possible to provide global scale sea surface salinity (SSS) distribution. This study analyzes globally the along‐track (Level 2) Aquarius SSS retrievals obtained using both passive and active L‐band observations. Aquarius along‐track retrieved SSS are assimilated into the ocean data assimilation component of Version 5 of the Goddard Earth Observing System (GEOS‐5) assimilation and forecast model. We present a methodology to correct the large biases and errors apparent in Version 2.0 of the Aquarius SSS retrieval algorithm and map the observed Aquarius SSS retrieval into the ocean model's bulk salinity in the topmost layer. The impact of the assimilation of the corrected SSS on the salinity analysis is evaluated by comparisons with in situ salinity measurements from Argo. The results show a significant reduction of the global biases and RMS of observations‐minus‐forecast differences at in situ locations. The most striking results are found in the tropics and southern latitudes. Our results highlight the complementary role and problems that arise during the assimilation of salinity information from in situ (Argo) and space‐borne SSS retrievals.

A dilemma of the uniqueness of weather and climate model closure parameters

Parameterisation schemes of subgrid-scale physical processes in atmospheric models contain so-called closure parameters. Their precise values are not generally known; thus, they are subject to fine-tuning for achieving optimal model performance. In this article, we show that there is a dilemma concerning the optimal parameter values: an identical prediction model formulation can have two different optimal closure parameter value settings depending on the level of approximations made in the data assimilation component of the prediction system. This result tends to indicate that the prediction model re-tuning in large-scale systems is not only needed when the prediction model undergoes a major change, but also when the data assimilation component is updated. Moreover, we advocate an accurate albeit expensive method based on so-called filter likelihood for the closure parameter estimation that is applicable in fine-tuning of both prediction model and data assimilation system parameters. In this article, we use a modified Lorenz-95 system as a prediction model and extended Kalman filter and ensemble adjustment Kalman filter for data assimilation. With this setup, we can compute the filter likelihood for the chosen parameters using the output of the two versions of the Kalman filter and apply a Markov chain Monte Carlo algorithm to explore the parameter posterior distributions.

Variational assimilation of HF radar surface currents in a coastal ocean model off Oregon

The impact of assimilation of sea surface velocity fields observed by a set of high-frequency (HF) radars is studied using a three-dimensional ocean circulation model configured along the Oregon coast. The study period is June–July 2008 featuring upwelling and separation of the coastal currents into the adjacent interior ocean. The nonlinear model is based on the Regional Ocean Modeling System (ROMS) and the data assimilation (DA) component on the AVRORA system utilizing the representer-based variational algorithm. Assimilation proceeds in a series of 3-day windows, providing an analysis solution in each window and a 3-day forecast into the next window. Experiments with two different initial condition error covariances are compared (one is dynamically balanced, based on the linearized equation of state, temperature-salinity relation, and geostrophic and thermal wind balance relations and the other is multivariate unbalanced). While the assimilation impact is statistically better in the case of the balanced covariance, the case with the unbalanced covariance also provides sensible improvement in terms of surface velocity and sea surface temperature (SST) model-data forecast statistics. The analysis of representer functions shows that even if the initial condition error covariance is unbalanced, the correction fields at the model initial time are partially balanced after each dynamical field is smoothed independently, due to inherent dynamical properties of the adjoint model. Assimilation of the HF radar surface currents improves not only surface velocity forecasts, but also geometry of the upwelling SST front and the sea surface height (SSH) slope near the coast, as verified against unassimilated satellite SSH and SST data. The assimilation also alters the latitudinal distribution of the time-averaged offshore transport. Combined HF radar velocity and other observations, e.g., altimetry, is needed to better constrain surface geostrophic currents in the entire model domain, including the area not covered by the HF radar data.