Abstract

The Weather Research and Forecasting Model (WRF) and its variational data assimilation system (WRFDA) are applied to the Chukchi–Beaufort Seas and adjacent Arctic Slope region for high-resolution regional atmospheric reanalysis study. To optimize WRFDA performance over the study area, a set of sensitivity experiments are carried out to analyze the model sensitivity to model background errors (BEs) and the assimilation of various observational datasets. Observational data are assimilated every 6 h and the results are verified against unassimilated observations. In the BE sensitivity analyses, the results of assimilating in situ surface observations with a customized, domain-dependent BE are compared to those using the WRF-provided global BE. It is found that the customized BE is necessary in order to achieve positive impacts from WRFDA assimilation for the study area. When seasonal variability is incorporated into the customized BE, the impacts are minor. Sensitivity analyses examining the assimilation of different datasets via WRFDA demonstrate that 1) positive impacts are always seen through the assimilation of in situ surface and radiosonde measurements, 2) assimilating Quick Scatterometer (QuikSCAT) winds improves the simulation of the 10-m wind field over ocean and coastal areas, and 3) selectively assimilating Moderate Resolution Imaging Spectroradiometer (MODIS) retrieved profiles under clear-sky and snow-free conditions is essential to avoid degradation of assimilation performance, while assimilation of Constellation Observing System for Meteorology, Ionosphere and Climate (COSMIC) retrievals has little impact, most likely due to limited data availability. Based on the sensitivity results, a 1-yr (2009) experimental reanalysis is conducted and consistent improvements are achieved, particularly in capturing mesoscale processes such as mountain barrier and sea-breeze effects.

1. Introduction

The Chukchi–Beaufort Sea region along the northern Alaskan coast is currently undergoing significant environmental changes, including the fastest rate of decline and maximum observed interannual variance of sea ice anywhere in the Arctic (Stroeve et al. 2007; Comiso et al. 2008), as well as increased 10-m wind speeds over recent decades as the sea ice retreats (Stegall and Zhang 2012). In addition, the potential for further oil industry development exists along the Chukchi–Beaufort coasts. The threat of oil spills comes with oil extraction these can have serious environmental consequences, leading to increased attention from the government, scientific community, and general public (e.g., Picou et al. 2009; Webler and Lord 2010). In particular, the coastal areas of the Chukchi–Beaufort Seas represent a vulnerable and fragile region, with an ecosystem and an environment that are sensitive to human impacts (Ford and Pearce 2010; Doney et al. 2012). It is therefore of critical importance to be able to accurately predict the dispersal and movement of oil spills and to assess the potential environmental impacts should a spill occur. Doing so requires a good understanding of surface wind, a crucial parameter for assessing and predicting oil spill transport (Reed et al. 1999).

Surface wind is primarily determined by the interaction of prevailing synoptic weather patterns with prominent underlying geographic features (e.g., Schwerdtfeger 1974; Kozo 1979, 1980; Bromwich 1989; Olsson and Harrington 2000; Parish and Cassano 2003; Liu et al. 2008; Moore and Pickart 2012). In the Chukchi–Beaufort Sea region, the Beaufort high and Aleutian low are the two dominant synoptic-scale weather patterns that most directly influence surface winds (Shulski and Wendler 2007; Overland 2009; Moore 2012); when the intensity and location of these systems change, the surface winds over the study area vary in response (Lynch et al. 2004; Stegall and Zhang 2012). Prevailing synoptic circulations further interact with local geographic features, generating mesoscale circulations (e.g., Dickey 1961; Lynch et al. 2001; Moore and Pickart 2012). The area’s geography is characterized by seasonally ice-covered ocean, along with the Brooks Range in northern Alaska and the Chukotka Mountains in eastern Siberia. During winter, the surface can become extremely cold and temperature inversions are common. Each of these local geographic features interacts with prevailing synoptic weather systems to generate characteristic mesoscale atmospheric circulations that correspondingly influence the region’s surface winds (Schwerdtfeger 1974; Kozo 1980).

Due to the complexity of weather and climate systems in the study area, accurate numerical modeling is a challenge. To improve performance in modeling the Arctic atmosphere, significant efforts have been made using the state-of-the-art regional Weather Research and Forecasting Model (WRF). Among these are the thorough evaluation of WRF performance over a large pan-Arctic domain (Cassano et al. 2011; Porter et al. 2011), as well as the development of Polar WRF (Bromwich et al. 2009; Hines et al. 2011), in which the model’s Arctic land surface processes and representation of sea ice were improved. The application of Polar WRF for the Arctic System Reanalysis (Bromwich et al. 2010; Wilson et al. 2011, 2012) indicated that its modeled surface temperature agreed well spatially with the forcing global reanalysis, while incorporating additional spatial details, particularly in regions of higher elevation. Wilson et al. (2011) suggested that the surface wind speed could be improved further if higher spatial resolution is used, as smoothed terrain on a coarse grid results in overestimated 10-m wind speed, encouraging us to choose a relatively high-resolution grid spacing for this study. In addition, mesoscale modeling of the Alaskan interior with WRF by Mölders (2008) suggested that biases can exist in such simulations due to inaccurate initial and boundary conditions. With the development of various satellite retrievals, as well as improvements in the in situ observational network, the accuracy of model initial conditions and forecast performance can be improved through the application of data assimilation techniques. Various reanalysis projects are taking advantage of such improved initial conditions to generate high quality data. We adopt the same approach and are applying WRF and its variational data assimilation system WRFDA (both version 3.2.1) to the study area to generate the long-term Chukchi–Beaufort High-Resolution Atmospheric Reanalysis (CBHAR). CBHAR will provide a unique opportunity for better understanding how the changing climate interacts with local finescale processes, and how these in turn impact the mesoscale meteorology of the surface wind field throughout the Chukchi–Beaufort Sea region. To achieve this goal, an optimized WRFDA will be configured for the area through a set of assimilation sensitivity experiments, in which the sensitivity of WRFDA to model background errors (BEs) and the assimilation of various observational datasets will be analyzed. With this customized data assimilation system in place, the high-resolution reanalysis CBHAR can be generated, toward the goal of more accurately capturing the finescale processes of the region’s surface wind field than has been possible in existing reanalysis products.

The remainder of this paper is structured as follows. Section 2 provides a brief description of the model and data used in this study; efforts toward optimizing the configuration of WRFDA, through generating model domain- and configuration-dependent model BEs and selectively assimilating various in situ and satellite observations, are introduced in section 3; a 1-yr experimental reanalysis produced with the optimized assimilation configuration is presented in section 4; and a summary of this study is given in section 5.

2. Model and data for constructing CBHAR

The Advanced Research core of WRF (ARW; Skamarock et al. 2008) is our vehicle for generating the high-resolution CBHAR reanalysis. ARW is a mesoscale model developed through a community collaboration effort and managed by the National Center for Atmospheric Research (NCAR). ARW has various dynamical and physical options and includes its own data assimilation system, WRFDA (Huang et al. 2009; Barker et al. 2012). WRFDA has been designed to be a flexible atmospheric data assimilation system, and includes options for both three-dimensional (3DVAR) and four-dimensional (4DVAR) variational data assimilation. In this study, the 3DVAR assimilation technique has been adopted to improve the model’s initial conditions and generate the final high-resolution analysis. We did not consider the 4DVAR due to its relatively high computational cost (Barker et al. 2004, 2012), though recent improvements made by Zhang et al. (2013a) that enhance its efficiency encourage further study with it. In WRFDA, the analysis field is generated by merging observations and background forecasts through iterative minimization of the cost function (Barker et al. 2004, 2012). The background forecasts in this study are generated by WRF simulations, and the observations are quality controlled before being assimilated by WRFDA. The modeling domain for this study (Fig. 1) is configured to encompass the Chukchi and Beaufort Seas, the entire Alaskan Arctic Slope and adjacent Brooks Range, portions of the Canadian Yukon, and the eastern tip of Siberia. The domain has a grid spacing of 10 km, with 49 vertical levels and a model top of 25 hPa. A summary of the model configuration parameters and physical schemes used is given in Table 1.

Fig. 1.

Modeling domain and distribution of in situ observation stations. Dots represent surface stations and diamonds are radiosonde stations. Shading over the ocean represents average sea ice concentration with the value 0.8 highlighted by the dashed line. Shading over land represents terrain height (km).

Fig. 1.

Modeling domain and distribution of in situ observation stations. Dots represent surface stations and diamonds are radiosonde stations. Shading over the ocean represents average sea ice concentration with the value 0.8 highlighted by the dashed line. Shading over land represents terrain height (km).

Table 1.

WRF configuration.

WRF configuration.
WRF configuration.

The European Centre for Medium-Range Weather Forecasts (ECMWF) Interim Re-Analysis (ERA-I; Dee et al. 2011) is used to provide atmospheric forcing (initial and lateral boundary conditions) and initial conditions for soil moisture and temperature, as well as sea surface temperature. Daily National Aeronautics and Space Administration (NASA) Bootstrap ice concentration (Comiso 2012), as acquired from the National Snow and Ice Data Center, is used. ERA-I has been widely used in Arctic WRF simulations (Bromwich et al. 2013) and has also proved to help WRF generate accurate simulations in our study domain (Zhang et al. 2013b). In situ observational data used for assimilation and model verification in this study include surface observations from 122 stations distributed throughout the model domain, along with three radiosonde stations, as indicated in Fig. 1. These in situ observations were collected from different data sources, including the National Climatic Data Center, the interagency network of Remote Automated Weather Stations, and others. All collected data were quality controlled using three separate quality-control (QC) procedures, checking for observations that fall outside of a normal range (threshold test), consecutive values that differ too greatly (step change test), and instances of excessively high or low variability (persistence test) (Shulski et al. 2013). Criteria for the quality control were defined based on each station’s climatology. Satellite retrievals assimilated with WRFDA include Quick Scatterometer (QuikSCAT) ocean-surface winds, as well as temperature and moisture profiles from both the Moderate Resolution Imaging Spectroradiometer (MODIS) and Constellation Observing System for Meteorology, Ionosphere, and Climate (COSMIC) platforms.

The in situ surface observations contain hourly 2-m temperature and humidity, 10-m wind, and pressure. The radiosonde profiles include 12-hourly temperature, humidity, and wind at vertical levels extending from the surface up to about 10 hPa. The QuikSCAT winds are available over open water at a spacing of 12.5 km and cover the period from 1999 to 2009 (Long and Mendel 1990). The MODIS-retrieved moisture and temperature profiles (King et al. 2003) include 20 vertical levels from 1000 to 5 hPa at a horizontal spacing of 5 km, and are available since 2000. The COSMIC retrievals (Anthes et al. 2008), available since 2006, include temperature and moisture profiles, thinned here to 26 vertical levels from 1000 to 10 hPa.

3. Optimization of data assimilation configuration within WRFDA

Though data assimilation is a useful technique for constraining the model solution and reducing model forecast errors, assimilating an extra dataset into the model does not necessarily improve the quality of the subsequent simulation for all variables due to accuracy issues in both the assimilated data and error information (Barker et al. 2004, 2012). The variational assimilation system WRFDA determines the optimal analysis through the use of information that includes model BE and observational error. The assimilation scheme within WRFDA then combines the dynamic model results with the observations using weights inversely related to each of their errors (Barker et al. 2004, 2012). Each observation data type is characterized by its own error distribution, dependent on the accuracy of the individual observational system, but model errors can be affected by the choice of model domain and configuration. To best optimize WRFDA for the production of CBHAR, we conduct two sets of sensitivity experiments as described in the following two sections: section 3a describes tests for the evaluation of model BE sensitivity, and section 3b details those for evaluating the potential observational data types to be assimilated.

The selected simulation periods include January 2009 representing winter conditions and July 2009 for summer conditions. The oceanic portion of the domain experiences high variability throughout the year in its sea ice coverage and surface temperature, and the land similarly varies in its snow cover and thermal profile. This high level of variability provides a good opportunity for investigating the performance of WRFDA by examining these two climatologically extreme months.

In all of the sensitivity simulations, the model is reinitialized every 2 days with ERA-I providing initial and lateral boundary conditions, and run for 2 days and 6 hours (Fig. 2). Since forecast errors increase with time dramatically after 2–3 days, even with frequent data assimilation, a 2-day reinitialization is chosen and available observational data are assimilated every 6 h (including at the initial hour). Only data within ±1.5 h of the assimilation hour are used. The results from the first 6 h of each run are used as model spinup and removed from consideration, with the remaining 2-day periods pieced together to form a continuous time series. In addition to the assimilation runs, a control simulation (CTRL), in which no data are assimilated through WRFDA but which is otherwise identical to the assimilation runs, including the stop–restart cycle every 6 h, is also conducted for the purposes of comparing and evaluating the sensitivity results.

Fig. 2.

Cycling scheme for the assimilation experiments. A cold-start run is initialized from ERA-I at 1200 UTC and run for 2 days + 6 h, with observational data assimilated every 6 h.

Fig. 2.

Cycling scheme for the assimilation experiments. A cold-start run is initialized from ERA-I at 1200 UTC and run for 2 days + 6 h, with observational data assimilated every 6 h.

The simulations are evaluated by analyzing the root-mean-square errors (RMSEs) as verified against nonassimilated observations and satellite retrievals, including in situ surface observations, QuikSCAT winds, and radiosonde profiles. To maintain independence between observations used for assimilation and verification, only model results at the forecast (nonassimilation) hours of each assimilation cycle (Fig. 2) are included in the RMSE calculation. For verification against radiosondes, however, as only two profiles are normally produced each day, both at assimilation times, the model results 1 h before the sounding times are instead compared with observations. Verified variables include sea level pressure (SLP), 2-m temperature (T), 10-m wind vector (VEC), and upper-level temperature and wind vector. The RMSE of VEC is defined as the square root of the sum of the mean squared errors of the zonal (U) and meridional (V) wind components, in order that the metric represents errors in both wind speed and direction. For each simulation experiment, the RMSEs are first calculated at each observation station, then averaged over all stations within the model domain. A Student’s t test is used to assess the statistical significance of the RMSE differences between the sensitivity and CTRL simulations. An improved sensitivity simulation is defined to be one with RMSEs less than CTRL, and with the difference statistically significant at the 95% level.

a. Analysis of WRFDA sensitivity to model background error

WRFDA offers the option to use either built-in global model BE (BE-GFS) or user-customized model BE in its data assimilation process. The built-in BE is generated from Global Forecast System (GFS) forecasts, produced by NCAR with a horizontal grid spacing of about 80 km, and can be used for any domain due to its global coverage (Barker et al. 2004). Customized BE is calculated following the National Meteorological Center (NMC) method (Parrish and Derber 1992), in which the differences between 12- and 24-h model forecasts valid at the same time are first calculated, with the model BE estimated after averaging all forecast differences over a period of time and the error covariances calculated. In this study, the 12- and 24-h forecasts for the entire year of 2009 are first generated with the model configuration listed in Table 1. The differences between the 12- and 24-h forecasts are then averaged, both over the entire year and for each month individually, in order to produce both a yearly averaged BE (BE-1yr) and monthly averaged BEs (BE-Jan, BE-Jul), respectively.

A total of five control and sensitivity simulations (Table 2) are conducted in order to evaluate the sensitivity of WRFDA to model BE in the assimilation of in situ surface observations. In addition to the built-in model BE-GFS, customized model BEs for the entire year (BE-1yr) and months of January (BE-Jan) and July (BE-Jul) are independently tested in the sensitivity experiments BE-GFS, BE-1yr, BE-Jan, and BE-Jul. Simulation periods include both January and July 2009 for most simulation experiments, with the exceptions that BE-Jan is only used for January 2009 and BE-Jul only for July 2009. The BE variances and length scales used in BE-GFS are 0.25 and 1.0, respectively; 1.0 variance and 1.0 scale length are used in BE-1yr, BE-Jan, and BE-Jul. While tuning of these parameters can also affect assimilation performance, tests conducted with BE-GFS using different values of these parameters generally performed similarly in this study (not shown).

Table 2.

Experiments for evaluating model BE; the BE column describes the BE used in each experiment.

Experiments for evaluating model BE; the BE column describes the BE used in each experiment.
Experiments for evaluating model BE; the BE column describes the BE used in each experiment.

The impacts of model BEs on assimilation performance are analyzed using the RMSE of the model outputs relative to surface observations, radiosondes, and QuikSCAT winds (Table 3). When verifying against surface observations, the stations are divided into two groups––coastal and inland stations––in order to distinguish the performance of the assimilation between complex terrain and flat, coastal areas. Surface stations located within 30 km of the shoreline are designated as coastal stations, with those farther from shore are designated as inland stations. Using this criterion, a total of 56 coastal stations and 66 inland stations are available within the model domain.

Table 3.

Verification of model BE experiments against different observational sources for T (°C), SLP (hPa), and VEC (m s−1) in January and July 2009 (RMSEs are calculated against nonassimilated surface observations and QuikSCAT). When verified against surface observations, coastal and inland stations are verified separately. RMSEs calculated against radiosondes are averaged over all sounding levels. Improved or degraded RMSEs (compared to CTRL) that are statistically significant at the 95% level are shown in boldface.

Verification of model BE experiments against different observational sources for T (°C), SLP (hPa), and VEC (m s−1) in January and July 2009 (RMSEs are calculated against nonassimilated surface observations and QuikSCAT). When verified against surface observations, coastal and inland stations are verified separately. RMSEs calculated against radiosondes are averaged over all sounding levels. Improved or degraded RMSEs (compared to CTRL) that are statistically significant at the 95% level are shown in boldface.
Verification of model BE experiments against different observational sources for T (°C), SLP (hPa), and VEC (m s−1) in January and July 2009 (RMSEs are calculated against nonassimilated surface observations and QuikSCAT). When verified against surface observations, coastal and inland stations are verified separately. RMSEs calculated against radiosondes are averaged over all sounding levels. Improved or degraded RMSEs (compared to CTRL) that are statistically significant at the 95% level are shown in boldface.

Comparisons of RMSEs among the control and sensitivity simulations as verified against observations for both the January and July cases (Table 3) demonstrate that the built-in model BE (BE-GFS) significantly degrades the model performance in our study area, generating much greater errors than the control simulation in which no data are assimilated through WRFDA. This comparison suggests that in order to achieve positive impacts from data assimilation via WRFDA, usage of the built-in model BE should be avoided. On the other hand, the customized model BEs, both yearly and monthly averaged, generally serve to enhance the model performance, reducing the errors for most of the examined variables. A close look at the comparison further demonstrates that the improvements in SLP and the 10-m wind field by the customized BEs are seen continually throughout both the cold and warm months. For temperature, however, a slight negative effect is present in January, even with the use of customized BEs. This indicates that accurately modeling the strong surface temperature inversions characteristic of Arctic winter remains a challenge for WRF. Accordingly, the estimated model BEs might not fully represent the actual model performance in such conditions, resulting in a degradation of the data assimilation performance. The performance of the monthly and yearly averaged BEs is similar, indicating that including seasonal variability in the customized BEs is insignificant in this study.

Assimilation of in situ surface observations generates differing impacts in coastal and inland areas. In the BE-GFS experiment, simulation of the 2-m temperature is degraded by 2% (23%) in the coastal areas, and by 15% (23%) inland when compared to the CTRL simulation in July (January), while the RMSEs for the 10-m wind vector are 5% (11%) and 9% (4%) larger in July (January) for the coastal and inland stations, respectively (Table 3). This indicates that the built-in model BE generated using coarse-resolution (~80 km) GFS simulations is more problematic during winter and over complex terrain. Greater overall improvements are achieved in coastal relative to inland areas through the use of customized model BEs. During summer, 2-m temperature and SLP near the coast are improved by 9% and 9%–11%, respectively, while the impacts are very small inland. For the 10-m wind field, improvements are similar in coastal and inland areas, with 5% (4%) improvement at the coastal stations and 5% (4%) inland during July (January). Similar to the difficulty that the model has in simulating inversions, the inland area, characterized by complex topography, presents a challenge for WRF in accurately capturing many terrain-induced finescale details. Because model BE does not represent actual model error relative to the observations, but rather the average difference between one forecast time and another, BE can be underestimated in situations where the model performs inherently poorly. Thus, the performance of the assimilation can be hampered due to a less accurate estimation of the model BE.

To understand the reason why assimilation using BE-GFS degrades the results relative to CTRL, the observation-minus-analysis (OMA) and observation-minus-background (OMB) results at each assimilation time are compared. It is noted that benefits are achieved in the analysis relative to background no matter which model BE is used. However, forecast errors tend to grow faster when using BE-GFS. Therefore, during the five free-forecast hours (Fig. 2), the errors are gradually accumulated, even though assimilation serves to reduce them every 6 h. This demonstrates that model BE not only determines immediate assimilation effects, but also impacts forecast evolution.

Since the coastal area receives a greater benefit from assimilation, the averaged 48-h simulation cycle of the temperature and wind vector RMSEs, as averaged over 56 coastal stations, is compared for the month of July 2009 over all simulation tests (Fig. 3). Over the averaged simulation cycle, both monthly and yearly averaged customized model BEs perform very similarly in improving the model results throughout the entire diurnal cycle. On the other hand, the built-in model BE (BE-GFS) degrades the model results, particularly for 2-m temperature during the day.

Fig. 3.

Averaged RMSEs of (a) temperature (°C) and (b) surface wind vector (m s−1) for BE experiments in July 2009. RMSEs are averaged over 56 coastal stations and displayed over the average 48-h cold-start model cycle.

Fig. 3.

Averaged RMSEs of (a) temperature (°C) and (b) surface wind vector (m s−1) for BE experiments in July 2009. RMSEs are averaged over 56 coastal stations and displayed over the average 48-h cold-start model cycle.

The spatial distribution of assimilation impacts can be further revealed by the analysis increments generated by WRFDA. The increments are calculated by subtracting the model background fields from the resultant analysis fields. Comparisons of temperature and wind-field (divided into U and V components) increments at the lowest model level between the sensitivity experiments BE-GFS and BE-1yr demonstrate that increments generated using customized BE differ from those using the built-in BE in many respects, such as spatial scale and distribution, as well as the magnitudes of adjustment (Fig. 4). Increments produced by BE-GFS display a larger-scale distribution, as exemplified by the positive temperature increment centered on central Alaska that extends northeastward all the way to Banks Island and westward to the Chukotka Mountains (Fig. 4a). In contrast, increments in BE-1yr are essentially constrained within the land areas of the domain and characterized by several smaller mesoscale centers. Model BE can determine how assimilated information is distributed. BE-GFS, generated by GFS at a grid spacing of 80 km, has a larger spatial impact, which can propagate the effects of assimilated observations over a wider area. Increments in the customized-BE experiment seem to show a reasonable range of assimilation effects. The magnitudes of adjustment in the experiment BE-GFS are also generally greater relative to those in BE-1yr. The maximum temperature increase is about 2.8°C in BE-GFS, while only about 1.8°C in BE-1yr. Temperature increases are greater than 1°C over roughly half the model domain in BE-GFS, while the same area sees increases of only 0.4°C or greater in BE-1yr. Similarly, wind field adjustments are as high as 0.9 m s−1 in BE-GFS, but only 0.4 m s−1 in BE-1yr. Since the data assimilated here originate primarily from land areas, it is expected that the most significant increments will also occur over land. However, significant increments in the experiment BE-GFS also occur over ocean areas, where few observational data are available. This further implies that the built-in model BE cannot accurately represent the true model performance within the study area and will likely degrade the performance of the assimilation if used.

Fig. 4.

Monthly averaged analysis increments for (a),(d) temperature (T) and wind components (b),(e) U and (c),(f) V as introduced by WRFDA when using (a)–(c) BE-GFS and (d)–(f) BE-1yr for July 2009. The increments are calculated by subtracting background fields from the analysis fields. Values are denoted by solid lines (positive) and dashed lines (negative).

Fig. 4.

Monthly averaged analysis increments for (a),(d) temperature (T) and wind components (b),(e) U and (c),(f) V as introduced by WRFDA when using (a)–(c) BE-GFS and (d)–(f) BE-1yr for July 2009. The increments are calculated by subtracting background fields from the analysis fields. Values are denoted by solid lines (positive) and dashed lines (negative).

From the comparison of RMSEs calculated against radiosondes (Table 3), we see that although in situ surface observations alone were assimilated by WRFDA, the upper atmosphere can also be impacted nontrivially when the built-in model BE is used; the impacts from customized model BEs are very small, however. The built-in BE increases the RMSEs by more than 40%–100% relative to CTRL. Comparisons of the RMSE profiles among all the simulations for July 2009 show that the errors introduced by use of the built-in model BE gradually decrease with height, and are equivalent to values of the other simulations above 300 hPa (Fig. 5), indicating that the negative effect is three-dimensional even though only two-dimensional data are assimilated.

Fig. 5.

The monthly averaged profiles of RMSEs, averaged over all radiosonde stations for (a) temperature (°C) and (b) surface wind vector (m s−1) in July 2009 for the BE experiments. The RMSEs of different runs as verified against radiosonde data are calculated at various vertical levels and then averaged over the three available radiosonde stations for the entire month.

Fig. 5.

The monthly averaged profiles of RMSEs, averaged over all radiosonde stations for (a) temperature (°C) and (b) surface wind vector (m s−1) in July 2009 for the BE experiments. The RMSEs of different runs as verified against radiosonde data are calculated at various vertical levels and then averaged over the three available radiosonde stations for the entire month.

As again shown in Table 3, improvements in wind speed and direction, as verified against QuikSCAT winds, are also achieved through the use of customized model BEs even though no QuikSCAT data are assimilated. The built-in model BE, however, degrades the performance of WRFDA, worsening both wind speed and direction. Improvements seen with the customized BEs are most likely due to improved 10-m winds along the coastal areas where coastal station observations are assimilated.

b. Performance analysis of WRFDA assimilation of multiple datasets

For a study area with a paucity of observational data coverage, multiple datasets, including both in situ observations and satellite retrievals from all available sources, have to be taken into consideration for use in WRFDA in order to produce a high quality regional reanalysis. As described in section 2, we have collected and quality controlled data from a total of three radiosonde locations and 122 in situ surface stations from different data networks. Satellite retrievals, including QuikSCAT ocean-surface winds and MODIS- and COSMIC-retrieved profiles, have also been acquired. Will assimilation of all these data via WRFDA enhance the model’s performance? To answer this question, a thorough evaluation is conducted through a series of data assimilation sensitivity experiments for July 2009 (Tables 4 and 7). All sensitivity experiments herein use the customized model BE from sensitivity experiment BE-1yr (Table 2). Simulation results, including 2-m temperature, SLP, and 10-m wind speed and direction, are verified against in situ surface and radiosonde observations, as well as QuikSCAT surface winds, and the RMSEs are calculated. The relative success achieved through assimilating each type of data is analyzed by comparing the RMSEs among the control and sensitivity simulations (Tables 5, 8, and 9). These will be detailed in the sections 3b(1)(4).

Table 4.

Experiments for evaluating different observational sources.

Experiments for evaluating different observational sources.
Experiments for evaluating different observational sources.
Table 5.

As in Table 3, but for experiments assimilating different observational sources and for July 2009 only. (In each experiment, only one observational source is assimilated, as denoted by the experiment name.)

As in Table 3, but for experiments assimilating different observational sources and for July 2009 only. (In each experiment, only one observational source is assimilated, as denoted by the experiment name.)
As in Table 3, but for experiments assimilating different observational sources and for July 2009 only. (In each experiment, only one observational source is assimilated, as denoted by the experiment name.)

1) Assimilation of in situ surface observations

Sensitivity experiment SFC (Table 4), in which the model BE from BE-1yr is used for the assimilation of surface station observations, is equivalent to the sensitivity experiment BE-1yr in Table 2 for the July case. As discussed in section 3a, assimilation of in situ surface observations generates improvements in most of the validated variables, particularly 10-m wind. Greater improvements are seen in the coastal region, where 2-m temperature is improved by 9% and 10-m wind by 5%, while the impact inland is relatively small. When the model results are verified against QuikSCAT winds, improvements are also seen. Therefore, assimilation of in situ surface observations does help to improve the model performance in simulating most near-surface variables, particularly 10-m winds.

2) Assimilation of radiosonde measurements

Radiosondes provide in situ measurements of atmospheric temperature, moisture, and wind at vertical levels ranging from 1000 hPa to about 10 hPa twice per day. In our study area, a total of three sounding stations (Barrow, Alaska; Inuvik, Northwest Territories, Canada; and Kotzebue, Alaska) made measurements during the study period of July 2009; two soundings per day from each station are assimilated in the assimilation experiment SONDE (Table 4). The comparison of RMSEs between the CTRL and SONDE simulations (Table 5) indicates that the assimilation of radiosonde measurements not only improves the model’s simulation of upper-air variables (where model output 1 h prior to the assimilation times is used for verification in order to maintain independence from the observations), but also affects the simulation of surface variables as verified against surface in situ observations and QuikSCAT winds. Due to the very limited number of soundings available in the study area, RMSEs of upper-air temperatures and winds are only slightly reduced in this experiment, by about 1%–2%. At the surface, 2-m temperature and SLP are improved by about 2% and 3%, respectively, in the coastal region. The 10-m wind over the ocean, as verified against QuikSCAT, is also slightly improved. Overall, this experiment indicates that a consistently positive impact can be achieved through the assimilation of radiosonde measurements.

3) Assimilation of QuikSCAT surface winds

The consistently positive impacts produced by WRFDA in assimilating in situ measurements give us the confidence to further extend our efforts by assimilating satellite retrievals. Assimilation of QuikSCAT surface winds is evaluated with the sensitivity experiment QSCAT (Table 4). QuikSCAT ocean-surface winds provide measurements of surface wind vectors over open water in the absence of sea ice (Fig. 6a). During the study period of July 2009, a meteorological buoy sponsored by the Bureau of Ocean Energy Management (BOEM) was deployed along the Beaufort Sea coast near Barrow, where it collected a month-long offshore meteorological dataset of 10-m winds, among other variables (http://knik.iarc.uaf.edu/buoy09). To better understand the accuracy of QuikSCAT winds in the study region, the wind speeds measured by the BOEM buoy are compared with QuikSCAT retrievals (Fig. 6b). The VEC correlation between buoy measurements and QuikSCAT winds for the period 5 August–15 September 2009 is 0.83, with an RMSE of just 0.82 m s−1, indicating that the quality of QuikSCAT winds in the study area is satisfactory. Here, the vector correlation is calculated using an equation written in terms of the orthogonal components of two vectors following Crosby et al. (1993).

Fig. 6.

(a) Snapshot of QuikSCAT surface wind coverage and speed (colors) in the model domain at 0551 UTC 18 Jul 2009. (b) QuikSCAT surface wind speed vs BOEM buoy (71.29°N, 152.14°W) observed wind speed in August–September 2009. Small red dot in (a) northeast of 68°N, 170°W represents the location of the BOEM buoy.

Fig. 6.

(a) Snapshot of QuikSCAT surface wind coverage and speed (colors) in the model domain at 0551 UTC 18 Jul 2009. (b) QuikSCAT surface wind speed vs BOEM buoy (71.29°N, 152.14°W) observed wind speed in August–September 2009. Small red dot in (a) northeast of 68°N, 170°W represents the location of the BOEM buoy.

During the open-water season (e.g., July 2009), when QuikSCAT winds are available in the study area, assimilation of QuikSCAT surface winds demonstrates a positive influence on the simulation of the offshore surface wind field. As before, model output from experiment QSCAT is verified against both in situ measurements and QuikSCAT winds at nonassimilation hours and the RMSE is calculated (Table 5). This demonstrates that assimilating QuikSCAT winds can significantly reduce the RMSE of surface wind vectors over open water as verified against unassimilated QuikSCAT winds. Wind vectors are improved by 11% relative to the control run.

The spatial distribution of the impacts of assimilating QuikSCAT surface winds can be depicted by the near-surface analysis increments, calculated by subtracting the model background field from the analysis field at the lowest model level. The increments of the wind components U and V, averaged over the simulation period (Fig. 7), demonstrate that the zonal wind (U) is increased over the Chukchi Sea by roughly 0.2 m s−1, and by 0.1–0.5 m s−1 in the Beaufort Sea coastal areas. Positive increments of the meridional wind (V) exhibit an adjustment of 0.1–0.3 m s−1, centered on the Alaskan Chukchi Sea coast and extending eastward along the coast. The increments also demonstrate how winds oriented parallel to the coast are improved. Larger increments for such winds are due to the surface wind regime in the coastal region being dominated by coast-parallel winds. Comparing the wind field increments introduced by assimilating QuikSCAT surface winds with in situ surface observations (Figs. 7 and 4e–f) demonstrates that areas where data are assimilated always receive greater impacts. The distribution of the 10-m wind field increments caused by the assimilation of QuikSCAT winds agrees well with the QuikSCAT data coverage (Fig. 6a) over the Chukchi Sea and Beaufort Sea coastal area. Thus, improved offshore 10-m winds are achieved in a region where few offshore in situ measurements are available via the assimilation of QuikSCAT surface winds, which is consistent with Fan et al. (2013).

Fig. 7.

Monthly averaged analysis increments for the wind components (a) U and (b) V introduced by assimilating QuikSCAT data during July 2009. The increment is calculated by subtracting background fields from the analysis fields. Values are denoted by solid lines (positive) and dashed lines (negative).

Fig. 7.

Monthly averaged analysis increments for the wind components (a) U and (b) V introduced by assimilating QuikSCAT data during July 2009. The increment is calculated by subtracting background fields from the analysis fields. Values are denoted by solid lines (positive) and dashed lines (negative).

4) Assimilation of satellite-retrieved profiles

An extreme lack of radiosonde observations in the study area motivates us to include as many satellite-retrieved profiles as possible via WRFDA for the production of the CBHAR reanalysis. Assimilating satellite retrievals and/or radiance data has demonstrated the ability to improve model simulations (Pu et al. 2002; Fan and Tilley 2005; Huang et al. 2005). Directly assimilating radiance data is recommended in order to avoid potential negative impacts as a result of retrieved profiles providing insufficient accuracy and/or vertical resolution. However, for radiance assimilation to have a positive impact, significant bias correction, as well as sensor and channel selection efforts are required (Liu et al. 2011). Even after such efforts, impacts at the surface are still limited (Liu et al. 2011). Assimilation of satellite-retrieved profiles also displays the ability to improve model results, particularly in remote and data-sparse areas, and in some cases even outperforms the direct assimilation of radiance data (Singh et al. 2012). MODIS-retrieved temperature and humidity profiles, when assimilated by a forecast model, have been shown to improve the forecasts (Xavier et al. 2006; Zhang et al. 2007). Assimilation of COSMIC profiles has also produced improvements in model simulations through the reduction of model bias (Cucurull and Derber 2008). Thus, satellite-retrieved temperature and moisture profiles from the MODIS and COSMIC instruments are targeted for use in this study.

MODIS profiles have much higher spatial and temporal resolutions than do radiosonde measurements, with two to three satellite passes per day over some part of the study domain and the data available at a horizontal spacing of about 5 km. After the data are thinned to a spacing of approximately 120 km in order to prevent oversaturation, which can cause diminished results (e.g., Liu and Rabier 2002), there remain approximately 1900 MODIS profiles per day across the model domain available to be assimilated by WRFDA.

In the retrieval product, each MODIS profile is given flags that indicate the quality of the profile and the conditions under which it was collected. Two examples include the cloud mask flag (CM), which indicates the presence of clouds, and the snow–ice background flag (SI), which indicates if the background was covered by snow or sea ice (Table 6). By taking note of these flags, users can assess how the profile accuracy varies under different conditions. To characterize the quality of data marked by these flags in the study area, the MODIS profiles for 2009 are first divided into six groups according to their CM and SI flags [snow–ice background under all sky (SI = 0); snow–ice background under cloudy sky (SI = 0, CM = 0, 1); snow–ice background under clear sky (SI = 0, CM = 2, 3); snow–ice-free under all sky (SI = 1); snow–ice-free under cloudy sky (SI = 1, CM = 0, 1); and snow–ice-free under clear sky (SI = 1, CM = 2, 3)] and then verified against radiosonde observations at the three sounding stations within the model domain. MODIS profiles within each category are calibrated by calculating the RMSE against radiosonde profiles measured at approximately the same time (within a 1-h window). A comparison of temperature RMSE profiles among the six flag categories (Fig. 8) shows that the profiles retrieved under a snow or ice background (SI = 0) contain errors that are 30%–50% larger than those without (SI = 1). A similar contrast is present between profiles retrieved under cloudy conditions (CM = 0, 1) and under clear sky (CM = 2, 3). In addition, data errors are highly variable with height. The RMSE at 1000 hPa is up to twice larger than at upper levels (850–300 hPa).

Table 6.

MODIS retrieval QC flags.

MODIS retrieval QC flags.
MODIS retrieval QC flags.
Fig. 8.

Average RMSE profiles for temperature in 2009 over all radiosonde stations with the use of different quality-control flags. MODIS profiles are categorized into six groups according to their SI and CM flags shown in the legend and compared with radiosonde profiles observed within a 1-h time window.

Fig. 8.

Average RMSE profiles for temperature in 2009 over all radiosonde stations with the use of different quality-control flags. MODIS profiles are categorized into six groups according to their SI and CM flags shown in the legend and compared with radiosonde profiles observed within a 1-h time window.

Considering the variations in MODIS profile quality demonstrated in Fig. 8, a series of sensitivity experiments (Table 7), in which MODIS profiles are screened by the CM and SI flags and data errors examined in Fig. 8, have been conducted in order to determine the optimal configuration for assimilating MODIS profiles via WRFDA. In these experiments, MODIS data are either assimilated at all levels (MODIS) or between 850 and 300 hPa (MODIS-MID). The sensitivity simulations are evaluated in the same way as before, with model results (SLP, T, and VEC) verified against radiosondes, surface observations, and QuikSCAT winds. Comparisons of the results from the sensitivity experiments MODIS, MODIS-MID, and CTRL (Table 8) demonstrate that assimilation of unfiltered MODIS retrievals (experiment MODIS) makes the results worse relative to CTRL for almost all examined variables. The experiment MODIS produces 10-m wind (VEC) RMSEs more than 2% larger than for CTRL over land, and more than 7% larger over the ocean. Perhaps surprisingly, the experiment MODIS-MID, in which the retrieval levels are selectively assimilated, does not produce any improvement over the degraded performance of the MODIS experiment. The results from MODIS-MID are also worse than CTRL, with RMSEs about 3% larger for 10-m wind and 1% larger for 2-m temperature in the coastal region.

Table 7.

MODIS assimilation experiments.

MODIS assimilation experiments.
MODIS assimilation experiments.
Table 8.

As in Table 3, but for experiments assimilating MODIS profiles and for July 2009 only.

As in Table 3, but for experiments assimilating MODIS profiles and for July 2009 only.
As in Table 3, but for experiments assimilating MODIS profiles and for July 2009 only.

In addition to the selective experiment MODIS-MID, MODIS profiles can also be screened through the use of CM and SI flags. An additional experiment, MODIS-QA, is conducted to examine the performance when assimilating only MODIS profiles that have a snow/ice-free background (SI = 1) and were retrieved under either probably clear (CM = 2) or confidently clear sky (CM = 3) conditions. To understand the vertical effects of assimilating MODIS retrievals, model-simulated temperature profiles from the experiments CTRL, MODIS, and MODIS-QA are compared at all levels and averaged over the entire simulation period of July 2009 at Barrow (Fig. 9). Comparisons of the monthly averaged bias profiles among all sensitivity tests show that the unfiltered experiment MODIS overcorrects the warm bias generated by CTRL in the middle and lower atmosphere (below ~500 hPa), producing a cold bias of up to 0.5°C. When MODIS profiles are selectively assimilated on the basis of quality assurance flags (experiment MODIS-QA), an improvement is seen in temperature at almost all levels, reducing both the cold bias in the middle and lower atmosphere (below ~300 hPa) and the warm bias in the upper atmosphere (above ~300 hPa). Selectively assimilating MODIS profiles via WRFDA through the use of quality assurance flags, such as those indicating snow–ice-free background and/or clear sky conditions, is therefore essential for avoiding the degradation of assimilation performance. This further confirms the results of previous studies suggesting that, before using satellite-retrieved products, the data should be filtered in order to diminish the possible negative impacts from poor quality retrievals (Key et al. 2003; Powers 2007).

Fig. 9.

Monthly averaged profiles of temperature bias at Barrow during July 2009 for experiments CTRL, MODIS, and MODIS-QA, as verified against radiosonde data at various vertical levels.

Fig. 9.

Monthly averaged profiles of temperature bias at Barrow during July 2009 for experiments CTRL, MODIS, and MODIS-QA, as verified against radiosonde data at various vertical levels.

About 20 COSMIC-retrieved profiles are available per day in the study domain. The profiles originally have 400 vertical levels, but are here thinned to 26 vertical levels to prevent oversaturation. When evaluating the effect of assimilating COSMIC profiles, the filtering procedure used in the MODIS experiments is not applied. COSMIC products are not accompanied by QC-flag information, suggesting that third-party sources are required in order to conduct such filtering, which may result in some inconsistencies. In addition, the radio occultation technique used by COSMIC is known to be minimally affected by clouds (Anthes et al. 2000). Experimental results (COSMIC) are verified separately against surface station (coastal and interior stations separately), radiosonde, and QuikSCAT observations for temperature, SLP, and wind (Table 9). Despite the lack of quality assurance, the assimilation of COSMIC profiles produces no negative impacts for either the surface or upper-air variables, encouraging us to consider their assimilation via WRFDA in generating the final CBHAR reanalysis.

Table 9.

As in Table 3, but for the experiment assimilating COSMIC profiles and for July 2009 only.

As in Table 3, but for the experiment assimilating COSMIC profiles and for July 2009 only.
As in Table 3, but for the experiment assimilating COSMIC profiles and for July 2009 only.

4. One-year experimental reanalysis

Synthesizing the results achieved in the previous sensitivity experiments, a 1-yr experimental reanalysis is generated in order to ensure that the selected observation types function as expected when combined within a single assimilation system. The experimental reanalysis is generated for 2009, using the modeling configuration listed in Table 1. The 1-yr experiment uses the same cycling scheme as the previous experiments (Fig. 2). In addition, spectral nudging to the forcing dataset (ERA-I) is employed in the experimental reanalysis in order to maintain consistency between this modeling configuration and the one ultimately used to generate the final CBHAR reanalysis. It should be noted that spectral nudging was not employed in any of the sensitivity experiments discussed above to avoid the damped assimilation effect by nudging. However, nudging sensitivity tests conducted by Zhang et al. (2013b) demonstrated that additional improvements can be achieved from the combination of nudging and assimilation, and thus they were both utilized in the reanalysis configuration. Spectral nudging is configured to use wavenumber 3 and nudging is performed at all vertical levels and for all variables. Results are again analyzed using the RMSE of the simulation output as calculated against the observations. In this section, the near-surface variables T and VEC are solely verified at in situ surface stations, as these provide the most accurate observations throughout the entire year (Fig. 10). RMSEs of T and VEC are separately averaged over coastal stations, defined as those located within 30 km of the coastline, and inland stations, composing the remainder. The RMSEs of the same fields from ERA-I are calculated in the same manner for purposes of comparison. Relative to ERA-I, the experimental reanalysis demonstrates consistent improvements for all variables examined. Temperature is improved to a larger degree in colder months than in the warm season. In contrast, improvements to the 10-m wind field do not exhibit a significant seasonal variation. The results demonstrate that errors are generally lower in the warmer months, when the variability in atmospheric conditions is at a minimum, and greatly increase in winter. During May–October, the RMSE of VEC is 17% less than in the cold season from November through April. Differences between coastal and inland stations are not significant. For temperature, similar seasonal variations are observed, with the minimum RMSE occurring in September, while the model performs differently for inland and coastal areas, with 26%–45% smaller RMSEs in coastal areas throughout the year. This differential performance in inland versus coastal regions is consistent with the results shown in previous sections. In addition to the effects of higher atmospheric variability, the larger errors seen in winter can also be attributed to the model’s relatively poor performance in simulating strong, shallow surface-based inversions.

Fig. 10.

Seasonal variations of RMSE, averaged over coastal stations (blue lines) and inland stations (orange lines) for (a) temperature (°C) and (b) surface wind vector (m s−1) in the 1-yr experimental reanalysis (solid) and ERA-I (dashed).

Fig. 10.

Seasonal variations of RMSE, averaged over coastal stations (blue lines) and inland stations (orange lines) for (a) temperature (°C) and (b) surface wind vector (m s−1) in the 1-yr experimental reanalysis (solid) and ERA-I (dashed).

An accurate surface wind field is one of the greatest concerns in modeling oil spills and ocean currents. How well does the experimental reanalysis capture the detailed mesoscale processes in the surface winds? To answer this, the monthly 10-m wind rose for observations, the experimental reanalysis, and ERA-I are compared for three coastal stations, as shown in Fig. 11. Stations Mys Shmidta in far eastern Russia and Barter Island, Alaska, are located close to the Chukotka and Brooks mountain ranges, respectively; their wind roses during the cold season can thus be used to demonstrate the mountain barrier effect. The Kivalina station is located on the Chukchi coast of Alaska, receiving little topographic impact from the distant Brooks Range; the wind rose at this location during the warm season can be used to show the sea-breeze effect. Overall, the 1-yr experiment tends to have better agreement with observations, producing a more accurate distribution of wind directions than does ERA-I. At Mys Shmidta in April, the dominant observed 10-m wind regime, oriented parallel to the Chukotka Mountains due to the mountain barrier effect that turns synoptic northeasterly winds to northwesterly (Stegall and Zhang 2012), is well captured in the 1-yr experiment, while ERA-I includes a greater cross-range 10-m wind component. At Barter Island in March, the high frequency of westerly winds, again due to the mountain barrier effect, which turns synoptic northerly winds to westerly (Stegall and Zhang 2012), is also missing in ERA-I. At Kivalina, during the period 16–21 June, a weak synoptic condition (not shown) favors the generation of westerly sea breezes along the coast. This feature is well captured by the 1-yr experimental reanalysis. This comparison demonstrates the ability of the experimental reanalysis to provide added information to the surface wind field, particularly for areas of highly variable topography, and how the simulation of such winds can benefit from higher modeled spatial resolution.

Fig. 11.

Comparison of 10-m wind roses between (top to bottom) observations (OBS), the 1-yr experimental reanalysis (1-yr Exp.), and ERA-I (ERA-I), and at (a)–(c) Mys Shmidta (April), (d)–(f) Barter Island (March), and (g)–(i) Kivalina (June) in 2009 [station locations marked with red dots in the domain map insets in (a),(d),(g)].

Fig. 11.

Comparison of 10-m wind roses between (top to bottom) observations (OBS), the 1-yr experimental reanalysis (1-yr Exp.), and ERA-I (ERA-I), and at (a)–(c) Mys Shmidta (April), (d)–(f) Barter Island (March), and (g)–(i) Kivalina (June) in 2009 [station locations marked with red dots in the domain map insets in (a),(d),(g)].

5. Summary

WRFDA is optimized for investigating the 10-m wind field over the Chukchi and Beaufort Seas and the Arctic Slope of Alaska. A series of assimilation sensitivity experiments, in which different BEs are used or different observational data are assimilated, are conducted and analyzed by comparing the resultant RMSEs as calculated against unassimilated observations with those from a control experiment (CTRL).

Experiments evaluating model BE suggest that usage of the built-in global BE (BE-GFS) included in the WRFDA package should be avoided, as its use contributes to significantly degraded simulations in both winter and summer, and for all examined variables checked, relative to CTRL. In contrast, usage of customized BEs (BE-1yr and BE-Jul) serves to improve the simulations. On average, customized model BE improves 10-m wind and 2-m temperature in coastal areas throughout the diurnal cycle, while usage of BE-GFS degrades the results, particularly for daytime 2-m temperatures. Analysis increments generated using the built-in BE exhibit larger spatial-scale features, as well as unrealistic adjustments over the ocean. The degraded performance of BE-GFS can be attributed to its presumably inaccurate representation of errors for the regional model used here, generated as it was from global GFS simulations with horizontal grid spacing of about 80 km, compared to the 10 km used by WRF in this study.

Assimilating in situ surface observations and radiosonde measurements produces highly responsive results. Verified surface variables are significantly improved, particularly in coastal areas during summer, through the assimilation of in situ surface observations. Assimilating radiosonde profiles improves the simulation of upper-air variables, and slightly improves ocean-surface winds.

QuikSCAT data provide distinctive coverage of surface wind fields over open water and, when assimilated, contribute to significant improvements in the simulation of sea surface winds, with a reduction in error of as much as 11%. Assimilating QuikSCAT data enhances zonal winds over the Chukchi Sea and the coastal areas of the Beaufort Sea. Meridional winds are enhanced along the Chukchi and Beaufort coasts of Alaska. The spatial pattern of the analysis increments is collocated with QuikSCAT data coverage.

Experiments assimilating MODIS data reveal the need for the careful selection of the particular profiles to be assimilated in order to avoid the negative impacts that use of the entire dataset otherwise introduces. Assimilating all MODIS data, regardless of quality, makes most examined variables worse than in CTRL. When the profiles are instead screened through the use of cloud mask and snow–ice background flags, the negative impacts are mostly dispelled and the cold bias in the middle and lower atmosphere (below ~300 hPa) and warm bias in the upper atmosphere (above ~300 hPa) are significantly improved.

To ensure that the performance of the combination of selected model background error and observation types is consistent with the individual sensitivity tests, a 1-yr experimental reanalysis for 2009 is generated. The experimental reanalysis further confirms that consistent improvements are achieved for all examined variables. Relative to ERA-Interim, larger improvements in temperature are demonstrated in colder months than in the warm season. Errors in the experimental reanalysis exhibit significant seasonal variation, with lower values in the warmer months while greatly increasing in the winter. Mesoscale processes, including mountain barrier and sea-breeze effects, are also well captured in the experimental surface wind field reanalysis.

Acknowledgments

This work was sponsored by the Bureau of Ocean Energy Management (BOEM) of the U.S. Department of the Interior under Contract M06PC00018. Computing resources were provided by the Arctic Region Supercomputing Center at the University of Alaska Fairbanks. We wish to thank Martha Shulski for the observational data collection and quality control and Robert Pickart for assisting with buoy deployment in 2009. Appreciations also go to Yuh-Lang Lin, Jianjun Xu, John Roop, and Yevgeniy Restigejev for helpful discussions and comments. We appreciate editor and anonymous reviews, which significantly improved the manuscript. ERA-I reanalyses were obtained from the CISL Research Data Archive managed by NCAR. Sea ice data were obtained from the National Snow and Ice Data Center. QuikSCAT winds were obtained from NASA’s Physical Oceanography Distributed Active Archive Center. MODIS profiles were obtained from NASA’s Atmosphere Archive and Distribution System. COSMIC profiles were obtained from the COSMIC Data Analysis and Archive Center.

REFERENCES

REFERENCES
Anthes
,
R. A.
,
C.
Rocken
, and
Y.
Kuo
,
2000
:
Applications of COSMIC to meteorology and climate
.
Terr. Atmos. Oceanic Sci.
,
11
,
115
156
.
Anthes
,
R. A.
, and
Coauthors
,
2008
:
The COSMIC/FORMOSAT-3 mission: Early results
.
Bull. Amer. Meteor. Soc.
,
89
,
313
333
.
Barker
,
D. M.
,
W.
Huang
,
Y.-R.
Guo
,
A. J.
Bourgeois
, and
Q. N.
Xiao
,
2004
:
A three-dimensional variational data assimilation system for MM5: Implementation and initial results
.
Mon. Wea. Rev.
,
132
,
897
914
.
Barker
,
D. M.
, and
Coauthors
,
2012
:
The Weather Research and Forecasting Model’s Community Variational/Ensemble Data Assimilation System: WRFDA
.
Bull. Amer. Meteor. Soc.
,
93
,
831
843
.
Bromwich
,
D. H.
,
1989
:
An extraordinary katabatic wind regime at Terra Nova Bay, Antarctica
.
Mon. Wea. Rev.
,
117
,
688
695
.
Bromwich
,
D. H.
,
K. M.
Hines
, and
L.-S.
Bai
,
2009
:
Development and testing of Polar Weather Research and Forecasting model: 2. Arctic Ocean
.
J. Geophys. Res.
,
114
,
D08122
,
doi:10.1029/2008JD010300
.
Bromwich
,
D. H.
,
Y.-H.
Kuo
,
M.
Serreze
,
J.
Walsh
,
L.-S.
Bai
,
M.
Barlage
,
K.
Hines
, and
A.
Slater
,
2010
:
Arctic System Reanalysis: Call for community involvement
.
Eos, Trans. Amer. Geophys. Union
,
91
(
2
),
13
14
,
doi:10.1029/2010EO020001
.
Bromwich
,
D. H.
,
F. O.
Otieno
,
K. M.
Hines
,
K. W.
Manning
, and
E.
Shilo
,
2013
:
Comprehensive evaluation of Polar Weather Research and Forecasting model performance in the Antarctic
.
J. Geophys. Res. Atmos.
,
118
,
274
292
, doi:10.1029/2012JD018139.
Cassano
,
J. J.
,
M. E.
Higgins
, and
M. W.
Seefeldt
,
2011
:
Performance of the Weather Research and Forecasting (WRF) model for month-long pan-Arctic simulations
.
Mon. Wea. Rev.
,
139
,
3469
3488
.
Chen
,
F.
, and
J.
Dudhia
,
2001
:
Coupling an advanced land surface–hydrology model with the Penn State–NCAR MM5 modeling system. Part I: Model implementation and sensitivity
.
Mon. Wea. Rev.
,
129
,
569
585
.
Comiso
,
J. C.
, cited
2012
: Bootstrap Sea Ice Concentrations from Nimbus-7 SMMR and DMSP SSM/I-SSMIS. National Snow and Ice Data Center. [Available online at http://nsidc.org/data/nsidc-0079.html.]
Comiso
,
J. C.
,
C. L.
Parkinson
,
R.
Gersten
, and
L.
Stock
,
2008
:
Accelerated decline in the Arctic sea ice cover
.
Geophys. Res. Lett.
,
35
,
L01703
, doi:10.1029/2007GL031972.
Crosby
,
D. S.
,
L. C.
Breaker
, and
W. H.
Gemmill
,
1993
:
A proposed definition for vector correlation in geophysics: Theory and application
.
J. Atmos. Oceanic Technol.
,
10
,
355
367
.
Cucurull
,
L.
, and
J. C.
Derber
,
2008
:
Operational implementation of COSMIC observations into NCEP’s Global Data Assimilation System
.
Wea. Forecasting
,
23
,
702
711
.
Dee
,
D. P.
, and
Coauthors
,
2011
:
The ERA-Interim reanalysis: Configuration and performance of the data assimilation system
.
Quart. J. Roy. Meteor. Soc.
,
137
,
553
597
.
Dickey
,
W. W.
,
1961
:
A study of a topographic effect on wind in the Arctic
.
J. Atmos. Sci.
,
18
,
790
803
.
Doney
,
S. C.
, and
Coauthors
,
2012
: Climate change impacts on marine ecosystems. Annu. Rev. Mar. Sci.,4, 1137, doi:10.1146/annurev-marine-041911-111611.
Fan
,
X.
, and
J. S.
Tilley
,
2005
:
Dynamic assimilation of MODIS-retrieved humidity profiles within a regional model for high-latitude forecast applications
.
Mon. Wea. Rev.
,
133
,
3450
3480
.
Fan
,
X.
,
J. R.
Krieger
,
J.
Zhang
, and
X.
Zhang
,
2013
:
Assimilating QuikSCAT ocean surface winds with the Weather Research and Forecasting model for surface wind-field simulation over the Chukchi/Beaufort Seas
.
Bound.-Layer Meteor.
,
148
,
220
226
.
Ford
,
J. D.
, and
T.
Pearce
,
2010
:
What we know, do not know, and need to know about climate change vulnerability in the western Canadian Arctic: A systematic literature review
.
Environ. Res. Lett.
,
5
,
014008
, doi:10.1088/1748-9326/5/1/014008.
Hines
,
K. M.
,
D. H.
Bromwich
,
L.-S.
Bai
,
M.
Barlage
, and
A. G.
Slater
,
2011
:
Development and testing of Polar WRF. Part III: Arctic land
.
J. Climate
,
24
,
26
48
.
Huang
,
C.-Y.
,
Y.-H.
Kuo
,
S.-H.
Chen
, and
F.
Vandenberghe
,
2005
:
Improvements in typhoon forecasts with assimilated GPS occultation refractivity
.
Wea. Forecasting
,
20
,
931
953
.
Huang
,
X.
, and
Coauthors
,
2009
:
Four-dimensional variational data assimilation for WRF: Formulation and preliminary results
.
Mon. Wea. Rev.
,
137
,
299
314
.
Iacono
,
M.
,
J.
Delamere
,
E.
Mlawer
,
M.
Shephard
,
S.
Clough
, and
W.
Collins
,
2008
:
Radiative forcing by long-lived greenhouse gases: Calculations with the AER radiative transfer models
.
J. Geophys. Res.
,
113
,
D13103
, doi:10.1029/2008JD009944.
Janjić
,
Z. I.
,
1994
:
The step-mountain Eta coordinate model: Further developments of the convection, viscous sublayer and turbulence closure schemes
.
Mon. Wea. Rev.
,
122
,
927
945
.
Janjić
,
Z. I.
,
1996
: The Mellor–Yamada level 2.5 scheme in the NCEP Eta Model. Preprints, 11th Conf. on Numerical Weather Prediction, Norfolk, VA, Amer. Meteor. Soc.,
333
334
.
Janjić
,
Z. I.
,
2002
: Nonsingular implementation of the Mellor–Yamada level 2.5 scheme in the NCEP Meso Model. NCEP Office Note 437, National Centers for Environmental Prediction, 61 pp.
Kain
,
J. S.
,
2004
:
The Kain–Fritsch convective parameterization: An update
.
J. Appl. Meteor.
,
43
,
170
181
.
Key
,
J. R.
,
D.
Santek
,
C. S.
Velden
,
N.
Bormann
,
J.-N.
Thépaut
,
L. P.
Riishojgaard
,
Y.
Zhu
, and
W. P.
Menzel
,
2003
:
Cloud-drift and water vapor winds in the polar regions from MODIS
.
IEEE Trans. Geosci. Remote Sens.
,
41
,
482
492
.
King
,
M. D.
, and
Coauthors
,
2003
:
Cloud and aerosol properties, precipitable water, and profiles of temperature and water vapor from MODIS.
IEEE Trans. Geosci. Remote Sens.,
41
,
442
458
.
Kozo
,
T.
,
1979
:
Evidence for sea breezes on the Alaskan Beaufort Sea coast
.
Geophys. Res. Lett.
,
6
,
849
852
.
Kozo
,
T.
,
1980
:
Mountain barrier baroclinity effects on surface winds along the Alaskan coast
.
Geophys. Res. Lett.
,
7
,
377
380
.
Liu
,
H.
,
P. Q.
Olsson
, and
K.
Volz
,
2008
:
SAR observation and modeling of gap winds in the Prince William Sound of Alaska
.
Sensors (Basel Switzerland)
,
8
,
4894
4914
.
Liu
,
Z.-Q.
, and
F.
Rabier
,
2002
:
The interaction between model resolution, observation resolution and observation density in data assimilation: A one-dimensional study
.
Quart. J. Roy. Meteor. Soc.
,
128
,
1367
1386
.
Liu
,
Z.-Q.
,
H.-C.
Lin
,
Y.-H.
Kuo
,
D.
Bromwich
, and
L.
Bai
,
2011
:
Data assimilation over northern polar region using WRF and WRFDA
. Extended Abstracts, 12th WRF Users’ Workshop, Boulder, CO, NCAR, 7B.7.
Long
,
D. G.
, and
J. M.
Mendel
,
1990
:
Model-based estimation of wind fields over the ocean from wind scatterometer measurements. I. Development of the wind field model
.
IEEE Trans. Geosci. Remote Sens.
,
28
,
349
360
.
Lynch
,
A. H.
,
A. G.
Slater
, and
M.
Serreze
,
2001
:
The Alaskan Arctic frontal zone: Forcing by orography, coastal contrast, and the boreal forest
.
J. Climate
,
14
,
4351
4362
.
Lynch
,
A. H.
,
J. A.
Curry
,
R. D.
Brunner
, and
J. A.
Maslanik
,
2004
:
Toward an integrated assessment of the impacts of extreme winds events on Barrow, Alaska
.
Bull. Amer. Meteor. Soc.
,
85
,
209
221
.
Mölders
,
N.
,
2008
:
Suitability of the Weather Research and Forecasting (WRF) model to predict the June 2005 fire weather for interior Alaska
.
Wea. Forecasting
,
23
,
953
973
.
Moore
,
G. W. K.
,
2012
:
Decadal variability and a recent amplification of the summer Beaufort Sea high
.
Geophys. Res. Lett.
,
39
,
L10807
, doi:10.1029/2012GL051570.
Moore
,
G. W. K.
, and
R. S.
Pickart
,
2012
:
Northern Bering Sea tip jets
.
Geophys. Res. Lett.
,
39
,
L08807
, doi:10.1029/2012GL051537.
Morrison
,
H. C.
,
G.
Thompson
, and
V.
Tatarskii
,
2009
:
Impact of cloud microphysics on the development of trailing stratiform precipitation in a simulated squall line: Comparison of one- and two-moment schemes
.
Mon. Wea. Rev.
,
137
,
991
1007
.
Olsson
,
P. Q.
, and
J. Y.
Harrington
,
2000
:
Dynamics and energetics of the cloudy boundary layer in simulations of off-ice flow in the marginal ice zone
.
J. Geophys. Res.
,
105
, 11 889–11 899.
Overland
,
J. E.
,
2009
:
Meteorology of the Beaufort Sea
.
J. Geophys. Res.
,
114
,
C00A07
, doi:10.1029/2008JC004861.
Parish
,
T. R.
, and
J. J.
Cassano
,
2003
:
The role of katabatic winds on the Antarctic surface wind regime
.
Mon. Wea. Rev.
,
131
,
317
333
.
Parrish
,
D.
, and
J.
Derber
,
1992
:
The National Meteorological Center’s spectral statistical-interpolation analysis system
.
Mon. Wea. Rev.
,
120
,
1747
1763
.
Picou
,
J. S.
,
C.
Formichella
,
B. K.
Marshall
, and
C.
Arata
,
2009
: Community impacts of the Exxon Valdez oil spill: A synthesis and elaboration of social science research. Synthesis: Three Decades of Research on Socioeconmic Effects Related to Offshore Petroleum Development in Coastal Alaska, S. R. Braund and J. Kruse, Eds., Alaska: Steven R. Braund and Associates, 279–310.
Porter
,
D. F.
,
J. J.
Cassano
, and
M. C.
Serreze
,
2011
:
Analysis of the Arctic atmospheric energy budget in WRF: A comparison with reanalyses and satellite observations
.
J. Geophys. Res.
,
116
,
D22108
, doi:10.1029/2011JD016622.
Powers
,
J. G.
,
2007
: Numerical prediction of an Antarctic severe wind event with the Weather Research and Forecasting (WRF) model. Mon. Wea. Rev.,135, 3134–3157.
Pu
,
Z.
,
W.-K.
Tao
,
S.
Braun
,
J.
Simpson
,
Y.
Jia
,
J.
Halverson
,
W.
Olson
, and
A.
Hou
,
2002
:
The impact of TRMM data on mesoscale numerical simulation of Supertyphoon Paka
.
Mon. Wea. Rev.
,
130
,
2448
2458
.
Reed
,
M.
,
Ø.
Johansena
,
P. J.
Brandvika
,
P.
Dalinga
,
A.
Lewisb
,
R.
Fioccob
,
D.
Mackayc
, and
R.
Prentkid
,
1999
:
Oil spill modeling towards the close of the 20th century: Overview of the state of the art
.
Spill Sci. Technol. Bull.
,
5
,
3
16
.
Schwerdtfeger
,
W.
,
1974
: Mountain barrier effect on the flow of stable air north of the Brooks Range. Proc. 24th Alaskan Science Conf., Fairbanks, AK, Geophysical Institute, University of Alaska Fairbanks, 204–208.
Shulski
,
M.
, and
G.
Wendler
,
2007
:
Climate of Alaska.
University of Alaska Press, 216 pp.
Shulski
,
M.
,
J.
You
,
J. R.
Krieger
,
W.
Baule
,
J.
Zhang
,
X.
Zhang
, and
W.
Horowitz
,
2013
:
Quality assessment of meteorological data for the Beaufort and Chukchi Sea coastal region using automated routines
.
Arctic
,
in press
.
Singh
,
R.
,
C. M.
Kishtawal
,
S. P.
Ojha
, and
P. K.
Pal
,
2012
:
Impact of assimilation of Atmospheric InfraRed Sounder (AIRS) radiances and retrievals in the WRF 3D-Var assimilation system
.
J. Geophys. Res.
,
117
,
D11107
, doi:10.1029/2011JD017367.
Skamarock
,
W. C.
, and
Coauthors
,
2008
: A description of the Advanced Research WRF version 3. NCAR Tech Note NCAR/TN–475+STR, 113 pp.
Stegall
,
S. T.
, and
J.
Zhang
,
2012
:
Wind field climatology, changes, and extremes in the Chukchi–Beaufort Seas and Alaska North Slope during 1979–2009
.
J. Climate
,
25
,
8075
8089
.
Stroeve
,
J.
,
M. M.
Holland
,
W.
Meier
,
T.
Scambos
, and
M.
Serreze
,
2007
:
Arctic sea ice decline: Faster than forecast
.
Geophys. Res. Lett.
,
34
,
L09501
, doi:10.1029/2007GL029703.
Webler
,
T.
, and
F.
Lord
,
2010
:
Planning for the human dimensions of oil spills and spill response
.
Environ. Manage.
,
45
,
723
738
.
Wilson
,
A. B.
,
D. H.
Bromwich
, and
K. M.
Hines
,
2011
:
Evaluation of Polar WRF forecasts on the Arctic System Reanalysis domain: Surface and upper air analysis
.
J. Geophys. Res.
,
116
,
D11112
, doi:10.1029/2010JD015013.
Wilson
,
A. B.
,
D. H.
Bromwich
, and
K. M.
Hines
,
2012
:
Evaluation of Polar WRF forecasts on the Arctic System Reanalysis domain: 2. Atmospheric hydrologic cycle
.
J. Geophys. Res.
,
117
,
D04107
, doi:10.1029/2011JD016765.
Xavier
,
A. C.
,
B. F. T.
Rudroff
,
L. M. S.
Berka
, and
M. A.
Moreira
,
2006
:
Multi-temporal analysis of MODIS data to classify sugarcane crop
.
Int. J. Remote Sens.
,
27
,
755
768
.
Zhang
,
X.
, and
J.
Zhang
,
2001
:
Heat and freshwater budgets and pathways in the Arctic Mediterranean in a coupled ocean/sea-ice model
.
J. Oceanogr.
,
57
,
207
237
.
Zhang
,
X.
,
Q.
Xiao
, and
P. J.
Fitzpatrick
,
2007
:
The impact of multisatellite data on the initialization and simulation of Hurricane Lili’s (2002) rapid weakening phase
.
Mon. Wea. Rev.
,
135
,
526
548
.
Zhang
,
X.
,
X.-Y.
Huang
, and
N.
Pan
,
2013a
: Development of the upgraded tangent linear and adjoint of the Weather Research and Forecasting (WRF) model. J. Atmos. Oceanic Technol.,30, 1180–1188.
Zhang
,
X.
, and
Coauthors
,
2013b
: Beaufort and Chukchi Seas Mesoscale Meteorology Modeling Study, final report. OCS Study BOEM 2013-0119, Alaska OCS Region, Bureau of Ocean Energy Management, U.S. Dept. of the Interior, Anchorage, AK, 204 pp. [Available online at http://www.boem.gov/BOEM-2013-0119/.]
Mellor
,
G. L.
, and
T.
Yamada
,
1982
:
Development of a turbulence closure model for geophysical fluid problems
.
Rev. Geophys. Space Phys.
,
20
,
851
875
.