1. Introduction
Current standard observational datasets [i.e., aviation routine weather report (METAR), rawinsonde] provide useful meteorological information on synoptic scales, but may lack the spatial and temporal frequency to accurately sample mesoscale features such as frontal boundaries, sea breezes, drylines, and cold pools. This problem may be mitigated by increasing the spatial density of surface observations as could be provided by mesoscale networks, such as the Oklahoma Mesonet (McPherson et al. 2007), and by more effective access to currently available surface observations as accomplished by MesoWest (Horel et al. 2002). The National Mesonet, an effort to expand mesoscale surface networks in order to fill gaps in the current observing network, is fostering increases to the number of low-level meteorological and hydrologic observations available for high-resolution analyses and forecasts (National Research Council 2009).
One analysis that takes advantage of the finer spatial resolution of meteorological observations provided by the National Mesonet is the Real-Time Mesoscale Analysis (RTMA) produced by the National Oceanic and Atmospheric Administration's (NOAA) National Centers for Environmental Prediction (NCEP) in collaboration with the Earth System Research Laboratory and the National Environmental, Satellite, and Data Information Service (De Pondeca et al. 2011). The RTMA utilizes the 13-km Rapid Update Cycle (RUC; Benjamin et al. 2004)1 downscaled to 5 km as a background upon which to assimilate surface temperature, moisture, pressure, and wind observations using a two-dimensional variational data assimilation (2DVAR) version of the NCEP Gridpoint Statistical Interpolation (GSI; Wu et al. 2002). The RTMA provides hourly analyses of 2-m temperature, 2-m specific humidity, surface pressure, and 10-m winds over the United States as needed by National Weather Service forecasters and other end users of weather information (De Pondeca et al. 2011).
Surface observations have also been shown to be valuable when assimilated while using the ensemble Kalman filter method (EnKF; Evensen 1994; Houtekamer and Mitchell 1998), leading to improvements in mesoscale ensemble analyses and subsequent short-term ensemble forecasts generated from these analyses. Hacker and Snyder (2005) show that EnKF assimilation of simulated surface observations can reduce ensemble-mean errors of planetary boundary layer (PBL) structure across the diurnal cycle. With perfect and imperfect model experiments, Zhang et al. (2006) and Meng and Zhang (2007) demonstrate the ability of the EnKF to improve forecasts of a poorly predicted snowstorm event through assimilation of simulated surface and sounding data typical of the standard U.S. observational network. Meng and Zhang (2008a,b) extend the two prior studies under a more realistic model framework, with the assimilation of real surface and rawinsonde data, and find enhanced predictability for a mesoscale convective vortex (MCV) event and an extended warm-season period during June 2003. Fujita et al. (2007) also demonstrate the ability of EnKF surface data assimilation to improve the short-term predictability of mesoscale features on two high-impact severe weather days. More recent studies by Stensrud et al. (2009) and Wheatley and Stensrud (2010) have shown that ensemble analyses of the environments associated with mesoscale convective systems (MCSs), as well as their salient features, are enhanced by EnKF surface data assimilation. Wheatley et al. (2012) indicate that short-term (0–6 h) forecasts of severe weather parameters are improved through EnKF assimilation of surface observations from land and marine stations, rawinsondes, and aircraft.
While the standard suite of surface observations is very helpful to National Weather Service forecasters for monitoring atmospheric conditions, the impact of the higher spatial observation density provided by mesoscale networks on objective analysis accuracy has not been fully determined. With respect to forecast accuracy, Benjamin et al. (2007) conclude that additional mesonet observations add little to improve forecast accuracy after METARs are included. Tyndall and Horel (2013), via a 2DVAR analysis procedure, show that the impacts of mesonet observations are dependent on location through the relationship between the meteorological conditions at the observation location, a successful diagnosis of those conditions by the background analysis, and the density of the observations surrounding it. This study also seeks to determine the impact of mesonet observations on analysis accuracy, as in Tyndall and Horel (2013), but through changes in the number of mesonet observations assimilated using an EnKF approach.
The 2DVAR methods used by the RTMA and the EnKF are similar in that they are inherently Bayesian methods that use a background error covariance (BEC) matrix to help compute the analysis of a particular variable. However, the determination of the BEC matrix represents a significant difference between the two methods. The RTMA 2DVAR method employs a static BEC matrix for each variable that does not account for changes in the dynamical state of the atmosphere. Conversely, the EnKF computes its BEC matrix from model forecasts generated from the previous assimilation cycle, such that the BEC explicitly contains dynamical information. The BEC matrix in the EnKF constantly evolves through the assimilation period, which theoretically should lead to an analysis more consistent with the present atmospheric state when compared to the RTMA. The RTMA 2DVAR method also is a univariate analysis, whereas the EnKF is multivariate. If a relationship exists among the observation increments of different surface variables, then a multivariate method will have a larger response than a univariate method as a result of the multivariate method processing more observations (Daley 1991). It is hypothesized that the relationships among the observation increments of different surface variables will be correctly and consistently identified using an EnKF approach on a high-resolution domain, thereby leading to analysis improvements when compared to the RTMA.
The past success of EnKF data assimilation in providing high quality surface analyses and forecasts suggests that it is a good method to use in an attempt to clarify the impact of mesonet data on the accuracy of high-resolution surface analyses. To facilitate comparisons with the RTMA, the experiments are designed with the same horizontal grid spacing for the EnKF analysis as the RTMA and use the same observations as input. The characteristics of the EnKF with all available observations, including all mesonet observations, are compared directly with the RTMA to validate the approach. The RTMA serves as the standard for comparison owing to its status as a current operational system and its skillful performance in prior assessments (e.g., De Pondeca et al. 2011). Once the characteristics of the EnKF analyses are known, data denial experiments are conducted wherein increasing amounts of mesonet observations are removed from the data provided to the EnKF. The quality of the resulting analyses is compared to the original EnKF in order to quantify the added value of the mesonet data to mesoscale analyses produced.
Section 2 describes the ensemble design, the quality controls enacted on the observational datasets, the EnKF method, and the design of the data-denial experiments employed in this study. A comparison of the EnKF and RTMA analyses, and the data-denial and conterminous United States (CONUS) grid assimilation experiments, is presented in section 3. Section 4 provides a summary of the results and conclusions.
2. Methodology
a. Ensemble design
A 36-member ensemble is constructed utilizing the Advanced Research Weather Research and Forecasting Model (ARW-WRF; Skamarock et al. 2008) with a (1000 km)2 domain with 5-km horizontal grid spacing centered over the central United States (Fig. 1a). To facilitate comparison with the ensemble, an identical grid is extracted from the full RTMA CONUS domain (Fig. 1b). An additional ARW-WRF CONUS grid with 18-km horizontal grid spacing (Fig. 1c) is created for use in an experiment discussed in section 3c. Fifty-one vertical grid levels are used, with spacing less than 100 m near the surface increasing to approximately 500 m at the model top pressure surface of 50 hPa. The model prognostic variables include the three wind components (u, υ, and w), perturbation variables (geopotential, potential temperature, and dry air surface pressure), water vapor, and hydrometeor species. The 2-m temperature and water vapor mixing ratio and 10-m winds, which are of particular interest to this study, are diagnosed from the surface layer schemes. All EnKF analyses of these diagnosed fields presented in section 3 depict the posterior ensemble mean, which is computed from an average over the 36-member ensemble following the update provided by the EnKF assimilation scheme described in section 2c.
(a) ARW-WRF domain used for full, data-denial, and no EnKF data assimilations for the 2-week warm- and cold-season experiments. (b) CONUS RTMA domain from which the comparison grid is extracted. (c) ARW-WRF domain used for the CONUS EnKF data assimilation experiments.
Citation: Weather and Forecasting 28, 3; 10.1175/WAF-D-12-00078.1
The ensemble-mean initial and boundary conditions are provided by the 12-km North American Mesoscale (NAM) model analyses, obtained from the NOAA National Operational Model Archive and Distribution System (NOMADS; http://nomads.ncdc.noaa.gov). To account for uncertainties in the initial and boundary conditions, the fixed covariance perturbation (FCP) method of Torn et al. (2006) is used whereby random samples from a default background error covariance file estimated by the National Meteorological Center (NMC, now known as NCEP) method (Parrish and Derber 1992) are generated by the WRF data assimilation (WRFDA) software and applied to all boundary points and fields for each ensemble member. These perturbations are on horizontal scales of tens to hundreds of kilometers, with magnitudes for the horizontal components of wind, water vapor mixing ratio, and temperature ranging from 5 to 10 m s−1, from 2 to 4 g kg−1, and from 2 to 4 K, respectively (e.g., Wheatley and Stensrud 2010, their Fig. 3).
To account for additional uncertainties induced by physical parameterization schemes, the ARW-WRF physics options are varied across the ensemble members (Table 1). Previous studies (e.g., Stensrud et al. 2000; Fujita et al. 2007; Stensrud et al. 2009; Wheatley et al. 2012) have demonstrated that the application of both initial condition and physical perturbations via model physics diversity to an ensemble system produces results that outperform those lacking physics diversity. The various schemes employed in this ensemble are as follow:
microphysics—Purdue Lin (Lin; Chen and Sun 2002), Thompson (Thompson et al. 2004);
cumulus—Kain–Fritsch (KF; Kain 2004), Betts–Miller–Janjić (BMJ; Betts and Miller 1986; Janjić 1994), Grell–Devenyi (Grell and Devenyi 2002);
PBL—Yonsei University (YSU; Hong et al. 2006), Mellor–Yamada–Janjić (MYJ; Janjić 1990, 1996, 2002), Asymmetrical Convective Model, version 2 (ACM2; Pleim 2007); and
shortwave radiation—fifth-generation Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model (MM5) shortwave (Dudhia 1989), Goddard shortwave (Chou and Suarez 1994).
Physical parameterizations used for each member of the 36-member EnKF ensemble. All ensemble members use the RRTM longwave radiation scheme and the Noah land surface model.
b. Assimilated data and quality control
All available observations of temperature, dewpoint, horizontal wind components, and altimeter settings from ~2100 METAR, ~80 rawinsonde, and ~17 000 integrated mesonet (Mesonet) sites (Fig. 2) are acquired through the Meteorological Assimilation Data Ingest System (MADIS; Miller et al. 2007). Additional upper-air observations of temperature, dewpoint, and horizontal wind components are obtained from the Aircraft Communications Addressing and Reporting System (ACARS) also through MADIS. Based on the work of Zapotocny et al. (2000) and Wheatley and Stensrud (2010), the magnitudes of error standard deviations for the wind components, temperature, dewpoint, and altimeter setting observations are set to 1.75 m s−1, 1.75 K, 1.75 K, and 1.5 hPa, respectively. All datasets are subject to the quality control procedure prescribed by MADIS, which is based primarily on the Advanced Weather Interactive Processing System (AWIPS) techniques specification package (TSP) 88–21-R2 (NWS 1994). This procedure consists of static and dynamic checks that include validity and internal, vertical, positional, temporal, and spatial consistencies (see Miller et al. 2005). Additional quality controls used in the RTMA are applied to the Mesonet dataset in the form of observation blacklists that remove stations with suspect pressure, temperature, dewpoint, and wind data [see De Pondeca et al. (2011) for list makeup] and wind “uselists” borrowed from the RUC system, which indicate stations that do not exhibit a low speed bias (Benjamin et al. 2007). The use of quality controls similar to those of the RTMA allows the current study to compare the results of EnKF data assimilation to the variational technique used by the RTMA without variability attributable to the use of significantly different observational datasets.
Mesonet (green dots), METAR (red dots), and rawinsonde (blue dots) observations available for use in the full-data EnKF assimilation experiments.
Citation: Weather and Forecasting 28, 3; 10.1175/WAF-D-12-00078.1
c. EnKF method
The observed variables mentioned in section 2b are assimilated at hourly intervals using the parallel version of the Data Assimilation Research Testbed (DART) software (Anderson and Collins 2007; Anderson et al. 2009) in conjunction with the ARW-WRF. The variables are cycled continuously throughout the periods of interest to produce hourly analyses of ensemble mean 2-m temperature, 2-m dewpoint, and 10-m winds for the 36-member ensemble. The DART EnKF is based upon the ensemble adjustment Kalman filter described in Anderson (2001), which is derived from the ensemble square root filter (Whitaker and Hamill 2002). An assimilation cycle begins with a 1-h forecast of all ensemble members. The following analysis step serially processes all available observations within a window ± 30 min surrounding the analysis time, rejecting observations flagged by the MADIS quality control procedure and greater than three standard deviations from the ensemble mean. This procedure provides the EnKF update of the ensemble-mean forecast and updates each ensemble member from its prior forecast state. To mitigate filter divergence and maintain ensemble spread, adaptive covariance inflation (Anderson 2009) is applied to the prior ensemble members for each model state vector. A spatial localization function (Gaspari and Cohn 1999) for assimilating observations is used with horizontal and vertical half-radii of 229 and 4 km, respectively. Finally, as part of the next hourly assimilation cycle, a 1-h forecast is computed from each of the updated ensemble members, and the cycle repeats.
d. Data-denial experiments
Several data-denial experiments are conducted to assess the influence of Mesonet observations on the accuracy of surface analyses on a 5-km grid. Standard hourly METAR observations and all other non-Mesonet observations as described in section 2b are used in each experiment. Experiments that reduce the Mesonet data by 10%, 25%, 50%, and/or 75% are completed for several cases and for various time windows. The 10–11 May 2010 time frame is selected because of a dryline passage across Oklahoma and Kansas responsible for a large tornado outbreak in the region, while the 25 November 2010 time frame is selected because of the passage of a strong polar cold front, with cross-frontal temperature differences of 30°F across a large portion of the analysis domain. The Mesonet data removal is done using a random selection approach and the resulting dataset is added to the complete METAR, rawinsonde, and ACARS datasets from the respective data-denial time period. The randomization of wind observation stations is completed separately, owing to the additional quality controls provided by the wind uselist, before being recombined in the final observation datasets. Table 2 indicates the resultant number of Mesonet sites used in the suite of EnKF experiments on both model grids employed in this study. Figure 3 provides a qualitative indication of the Mesonet data density present in the 10% and 75% data-denial experiments for the temperature–dewpoint–altimeter observations. Wind observations for these experiments exhibit a similar pattern. Since the EnKF is a multivariate scheme, the decrease in data density is somewhat less than the percentages indicated in Table 2, as wind observations can influence the other model variables and vice versa.
Number of Mesonet stations available for the suite of EnKF data assimilation experiments on the CONUS and (1000 km)2 domains.
Temperature, dewpoint, and altimeter Mesonet observation sites available for assimilation in the (a) 10% and (b) 75% data-denial experiments.
Citation: Weather and Forecasting 28, 3; 10.1175/WAF-D-12-00078.1
3. Results
a. Characteristics of EnKF analyses and RTMA
Analyses from 0000 UTC 17 May (Fig. 4) and 0000 UTC 2 December 2010 (Fig. 5) are typical and illustrate that the two analyses produce the same general patterns and features. However, distinct qualitative differences between the structure of the EnKF and RTMA 2-m temperature, 2-m dewpoint temperature, and 10-m winds fields also are evident. The RTMA produces analyses with more small-scale structure than does the EnKF, as can be seen most clearly in the dewpoint temperature and wind analyses. (The localized warm/moist areas present in the EnKF analyses in December are indicative of a problem with the handling of inland lake temperatures in this version of the ARW-WRF. This may have some influence on the overall verification, but its impact is small owing to the similar verification results from warm and cold seasons and a minimal number of observation sites located near these areas.) A marked dipole in dewpoint temperature is indicated in central Oklahoma on 2 December, with values varying between 36° and 15°F over a small horizontal distance in the RTMA. The dewpoint field in the EnKF captures the same area of variable dewpoint values, but the field varies more smoothly. The 10-m wind field also exhibits more finescale structure in the RTMA across a large portion of the domain, but most significantly in Oklahoma, Kansas, Arkansas, and Missouri, where several regions have small-scale changes in both wind speed and direction that deviate from the overall flow pattern.
The 2-m temperatures (°F; see color bar) and 10-m winds from the (a) EnKF analysis and (b) RTMA valid at 0000 UTC 17 May 2010. The 2-m dewpoint temperatures (°F) and 10-m winds valid at the same time for the (c) EnKF analysis and (d) RTMA.
Citation: Weather and Forecasting 28, 3; 10.1175/WAF-D-12-00078.1
As in Fig. 4, but valid at 0000 UTC 2 Dec 2010.
Citation: Weather and Forecasting 28, 3; 10.1175/WAF-D-12-00078.1
The variability observed between the RTMA and EnKF analyses stems from the differences in their methodologies. The RTMA 2DVAR scheme employs a univariate analysis that assumes a specified distance-dependent Gaussian form for the covariances for all observations at all analysis times, differing only in the horizontal spatial correlation scale assumed for each observation variable. These covariances (in the absence of terrain gradients) are horizontally isotropic and are proportional to the square of the assumed background error standard deviation, which again varies by observation variable. When terrain gradients occur, the covariances are anisotropic and depend upon an assumed structure function that was determined via trial and error (De Pondeca et al. 2011). Conversely, the EnKF is a multivariate analysis, and flow-dependent covariances are calculated uniquely for each observation location and analysis time, such that the covariances are consistent with the nonlinear balance present in the ARW-WRF. The EnKF covariances tend to be anisotropic, their magnitudes vary both from point-to-point and in time, and the covariances can be of either sign. A comparison of the covariance structures assumed by the RTMA and calculated by the EnKF indicates that the covariances in the EnKF can be more strongly anisotropic than those from the RTMA (not shown). In addition, the assumed observation errors are smaller in the RTMA than in the EnKF for most variables. In general, the fixed covariances assumed by the RTMA lead to each innovation influencing the resulting analysis proportional to its magnitude, while the smaller assumed observation errors in the RTMA result in a tighter fit to the observations in comparison with the EnKF approach. The comparative smoothness in the EnKF analysis fields is a product of the flow dependence of its covariance structures and magnitudes, the larger assumed observational error standard deviations, as well as the allowance of cross covariances between the variables that is absent in the RTMA.
The accuracy of the EnKF and RTMA analyses is examined through calculation of the root-mean-square (RMS) innovations at each analysis hour as the difference between an observational value and the interpolated analysis value at the same location. Bilinear interpolation is used to interpolate the analysis to the observed location from the surrounding four grid points, and the posterior ensemble mean value is used for the EnKF analysis. Mesonet and METAR observations compose most of the observations available for assimilation, so RMS innovations computed from these datasets provide a robust indication of the performance of the EnKF and the variational technique employed by the RTMA. All available Mesonet and METAR observations within the domain are used in the calculations, if they pass the MADIS quality control procedure.
Several general characteristics are indicated by the RMS innovations for the four pertinent variables during 1 week of the warm season (Fig. 6) and 1 week of the cold season (Fig. 7). The evolution of the hourly RMS innovations for 2-m temperature and dewpoint are very similar over the 1-week periods. An average over the first weeks of the warm and cold seasons (Table 3) for all variables shows that the RMS innovations for the RTMA are between 0.2 and 0.3 K lower than the EnKF for 2-m temperature and dewpoint temperature. However, the averaged RMS innovations for the u- and υ-wind components indicate that the EnKF fits the observations more closely by an order of 0.1 m s−1 or less. The lower RMS innovations for the thermodynamic fields are related to the differences in covariance structures and the assumed observation errors (1.0 K in RTMA and 1.75 K in EnKF for temperature), with the RTMA producing a tighter fit to the observations. The similar wind component innovations are due in part to the similar observation errors assumed by the two methods (1.6 m s−1 in RTMA and 1.75 m s−1 in EnKF). Both analyses show a better fit to the observations when compared to an experiment in which no MADIS observations are assimilated and the model evolution is only governed by the NAM initial and lateral boundary conditions (Table 3). The value of the EnKF data assimilation is indicated by the lower RMS innovation for the posterior ensemble mean when compared to the prior. Similar characteristics are also found when the EnKF analyses are extended for another week in the warm and cold season (not shown).
Domain-averaged RMS differences calculated from observations and RTMA and the EnKF posterior ensemble-mean analysis (i.e., innovations) for (a) T, (b) Td, (c) 10-m u-wind component, and (d) 10-m υ-wind component for the first week of the warm season (0000 UTC 10 May–0000 UTC 17 May 2010).
Citation: Weather and Forecasting 28, 3; 10.1175/WAF-D-12-00078.1
As in Fig. 6, but for the first week of the cold season (0000 UTC 25 Nov–0000 UTC 2 Dec 2010).
Citation: Weather and Forecasting 28, 3; 10.1175/WAF-D-12-00078.1
Average RMS innovations for 2-m temperature T, 2-m dewpoint temperature Td, and the 10-m u- and υ-wind components for the prior and posterior EnKF full-data ensemble, no assimilation ensemble, and the RTMA during the first week (7-day period) of the warm-season (10–17 May 2010) and cold-season (25 Nov–2 Dec 2010) experimental periods.
To quantify the horizontal scales represented in the EnKF and RTMA analyses, kinetic energy spectra are computed for the 2-week warm and cold season periods (Fig. 8) using the technique described in Errico (1985). Results from the EnKF analyses follow close to the k−5/3 slope down to near 30-km wavelengths, and then follow a k−3 slope for smaller wavelengths. The EnKF spectrum during both periods is shown to be a better match to that computed by Skamarock (2004) from deterministic ARW-WRF forecasts. For wavelengths less than 20 km, there is more energy in the RTMA analyses compared to the EnKF (Fig. 8), signifying the presence of smaller-scale features in the RTMA fields as suggested visually in Figs. 4 and 5. However, the slopes of the RTMA spectra differ from those from the EnKF with more energy contained in wavelengths greater than 100 km and less energy in wavelengths between 100 and 30 km. The reasons for these differences have not been examined, but we hypothesize that they are as a result of the assumed covariance structures used by the RTMA modifying the model energy spectra. The EnKF spectra exhibit an effective resolution of approximately 35 km, which is consistent with the approximate 7Δx effective resolution found by Skamarock (2004) for ARW-WRF grid spacings of 22, 10, and 4 km.
Kinetic energy spectra for the EnKF ensemble-mean analysis and RTMA from the (a) 2-week warm- and (b) 2-week cold-season experimental periods.
Citation: Weather and Forecasting 28, 3; 10.1175/WAF-D-12-00078.1
Further confidence that the EnKF approach yields reasonable analyses compared to the RTMA is provided by an examination of specific events. Thus, the differences between the full-data assimilation EnKF analyses and the RTMA are examined more closely for events where significant mesoscale structure is present in the experimental domain. Three cases are selected; two during the warm season and one during the cold season, each characterized by the presence of a particular mesoscale feature.
Because of its well-defined structure within the experimental domain, the dryline that promoted a major tornado outbreak across the southern portion of the analysis domain on 10–11 May 2010 is examined more closely. The structures of the 2-m dewpoint temperatures and 10-m winds from the EnKF ensemble-mean and RTMA at 0000 UTC 11 May 2010 clearly illustrate the differences in the two analyses (Fig. 9). The dewpoint temperature gradient, as well as the dewpoint field behind and ahead of the dryline, exhibits a much smoother and coherent appearance in the EnKF analysis. The highly wavy dryline structure shown by the RTMA is not seen in the fine lines visible along the dryline in the Oklahoma City, Oklahoma (KTLX), and Wichita, Kansas (KICT), Doppler radars (not shown). The 10-m winds also exhibit some peculiar features in the RTMA to the east of the dryline, where the RTMA indicates westerly winds, and in the cyclonic flow around the surface low pressure area in southern Kansas. These features are not seen in the EnKF analysis.
The 2-m dewpoint temperature (°F; see color bar) and 10-m winds valid at 0000 UTC 11 May 2010 from the (a) EnKF posterior ensemble-mean analysis and (b) RTMA.
Citation: Weather and Forecasting 28, 3; 10.1175/WAF-D-12-00078.1
A mesoscale cold pool, the result of training convection associated with several waves of surface low pressure along a stalled frontal boundary, is the main feature of the next event that occurs on 14 May 2010. Six-hour total precipitation ending at 1800 UTC 14 May 2010 from NCEP's stage IV analysis (not shown) shows the heavy rainfall that occurred along the Interstate 44 corridor across southern Missouri and eastern Oklahoma. The result of this precipitation is an extensive area of cold 2-m temperatures across this region that is evident in both the EnKF analysis and RTMA at 1600 UTC 14 May 2010 (Fig. 10). Differences in 2-m temperature between the EnKF analysis and RTMA (Fig. 10c) reveal that the most significant differences on scales greater than 100 km are found in eastern Missouri and across southern Kansas into northern and western Oklahoma.
The 2-m temperature (°F; see color bar) and 10-m winds valid at 1600 UTC 14 May 2010 from the (a) EnKF posterior ensemble-mean analysis and (b) RTMA. (c) The 2-m temperature difference between the EnKF analysis in (a) and the RTMA in (b). Warm colors show regions where the EnKF temperature is warmer, whereas cold colors indicate areas where the RTMA temperature is warmer.
Citation: Weather and Forecasting 28, 3; 10.1175/WAF-D-12-00078.1
The cold-season event occurs on 25 November 2010, whereby a strong polar cold front is progressing through the experimental domain. Temperature changes across the boundary range from 30° to 40°F, with an abrupt wind shift from southerly to northerly observed as the front pushes south. Both analyses capture these features in the 2-m temperature and 10-m wind fields at 0600 UTC 25 November 2010 (Fig. 11). Comparison of the analyses shows general agreement across Oklahoma and Texas. However, the EnKF temperatures are warmer than the RTMA in a narrow band just ahead of the frontal boundary across southeast Missouri and eastern Oklahoma, and across an expansive postfrontal airmass area that covers northeastern Oklahoma, eastern Kansas, northwestern Missouri and western Iowa, including most of Nebraska (Fig. 11c).
As in Fig. 10, but valid at 0600 UTC 25 Nov 2010.
Citation: Weather and Forecasting 28, 3; 10.1175/WAF-D-12-00078.1
The overall similar quality of the RTMA and EnKF analyses, along with the small differences between the analyses in the vicinity of significant mesoscale features and the agreement between the kinetic energy spectra of the EnKF and past studies, suggest that the EnKF method can be used to produce reasonable surface analyses. Thus, the EnKF approach is used to evaluate the influence of Mesonet surface observations in the creation of surface analyses.
b. Data-denial experiments
Results from 1-week data-denial experiments during both the cold and warm seasons show a surprising lack of degradation when 75% of the Mesonet data are removed. RMS innovations from this experiment during the periods 10–17 May (Fig. 12) and 25 November–2 December 2010 (Fig. 13) for four variables are shown to be nearly identical to those from the full EnKF assimilation experiment. Very modest improvement in the 10-m wind components when using all Mesonet observations is indicated by the slightly lower RMS innovations during the warm season, while less improvement is denoted during the cold season. The greatest differences between the EnKF and no-assimilation experiments are seen in the 2-m temperature and 10-m wind component RMS innovations when peaks are observed in the no-assimilation experiment. This is particularly evident in the 2-m temperature RMS innovations during a rapid push of cold air behind a strong polar cold front in the 29–30 November time frame (Fig. 13a), which are greater than 1 K lower for both the full-data and 75% Mesonet data-denial EnKF experiments when compared to the no-assimilation experiment, indicating the benefits of data assimilation during times of rapidly changing meteorological conditions. An additional set of RMS innovations computed for the 75% data-denial experiment between the nonassimilated observations and the prior and posterior ensemble means during both periods shows that the posterior mean more closely fits the observations, an indication that the EnKF system is improving the ensemble-mean analysis quality.
Domain-averaged RMS differences calculated from observations and the EnKF posterior ensemble-mean analysis (i.e., innovations) for (a) T, (b) Td, (c) 10-m u-wind component, and (d) 10-m υ-wind component for the first week of the warm season (0000 UTC 10 May–0000 UTC 17 May 2010) for the full-data assimilation, no data assimilation, and 75% Mesonet data-denial experiments.
Citation: Weather and Forecasting 28, 3; 10.1175/WAF-D-12-00078.1
As in Fig. 12, but for the first week of the cold season (0000 UTC 25 Nov–0000 UTC 2 Dec 2010).
Citation: Weather and Forecasting 28, 3; 10.1175/WAF-D-12-00078.1
A continued lack of analysis degradation is exhibited in the shorter-term data-denial experiments where 10%, 25%, 50%, and 75% of the Mesonet data are removed during periods in which important mesoscale features are present within the experimental domain. Dewpoint temperatures at 0000 UTC 11 May 2010 for the suite of data-denial experiments (Figs. 14a–d) all exhibit the same general pattern, with only subtle variability in the vicinity of the dryline and minimal discrepancies across the remainder of the domain. Differences between the data-denial experiments and the full EnKF experiment are generally consistent with those calculated from the 50% data-denial experiment (Fig. 14e), which depicts a narrow band of larger analysis differences near the dryline. The 2-m dewpoint temperatures are higher in the data-denial experiment just ahead of the dryline compared to the full-data experiment, while the dewpoints are lower in a narrow band just behind the dryline compared to the full-data experiment. Only two areas, one in west-central Oklahoma and another along the Texas–Oklahoma border, show larger differences between the experiments, where the 25% (not shown) and 50% (Fig. 14e) experiments have broader regions of 2-m dewpoint temperatures that are lower when compared to the full assimilation experiment. The similarity in the 2-m dewpoint temperature analyses from the data-denial experiments is verified by the RMS innovations (Fig. 15) during the 36-h experimental period, where the results again are extremely similar to one another and the full-data experiment. As in the full-data assimilation experiment, for temperature and dewpoint, the RMS innovations for the suite of EnKF experiments are generally higher than those of RTMA, but similar for the u- and υ-wind components.
The 2-m dewpoint temperatures (°F; see color bar) and 10-m winds from the (a) 10%, (b) 25%, (c) 50%, and (d) 75% Mesonet data-denial experiments valid at 0000 UTC 11 May 2010. (e) The 2-m dewpoint temperature differences (°F; see label bar) between the full EnKF data assimilation and the 50% Mesonet data-denial experiments. Warm colors indicate that the dewpoint temperature from the data-denial experiment is higher, whereas cold colors show regions where the full-data EnKF dewpoint temperature is higher.
Citation: Weather and Forecasting 28, 3; 10.1175/WAF-D-12-00078.1
Domain-averaged RMS differences calculated from observations and RTMA and the EnKF posterior ensemble-mean analysis (i.e., innovations) for (a) T, (b) Td, (c) 10-m u-wind component, and (d) 10-m υ-wind component for the period 0100 UTC 10 May–1200 UTC 11 May 2010 for the full-data assimilation experiment, the suite of data-denial experiments, and the RTMA.
Citation: Weather and Forecasting 28, 3; 10.1175/WAF-D-12-00078.1
Similar results are found in the suite of data-denial and full-data assimilation experiments using observations from 25 November 2010. The 2-m temperatures at 0600 UTC 25 November 2010 for the suite of data-denial experiments (Figs. 16a–d), as well as the differences between the 50% data-denial and full EnKF experiments (Fig. 16e) indicate that the greatest differences occur near the frontal boundary and within the colder air mass behind the front. The narrow band of relatively large negative values just behind the front, as well as across much of the western portion of the domain, shows regions where the data-denial 2-m temperatures are cooler relative to the full-data experiment. Conversely, 2-m temperatures are generally warmer in the data-denial experiments just ahead of the front. This overall difference pattern is consistent throughout the suite of data-denial experiments at this time. This similarity also is supported by the RMS innovations for the 2-m temperature, dewpoint, and the 10-m wind components (Fig. 17) for the entire experimental period on 25 November 2010, in which there are few notable differences between the EnKF experiments. Comparison of the RMS innovations between the suite of EnKF experiments and the RTMA indicates the presence of similar features as in the 10–11 May 2010 data-denial experiment.
As in Fig. 14, except that 2-m temperatures are shown valid at 0600 UTC 25 Nov 2010.
Citation: Weather and Forecasting 28, 3; 10.1175/WAF-D-12-00078.1
As in Fig. 15, but for the period 0100 UTC 25 Nov–0000 UTC 26 Nov 2010.
Citation: Weather and Forecasting 28, 3; 10.1175/WAF-D-12-00078.1
These results indicate that the addition of Mesonet data has little influence on the resulting EnKF analyses. To help understand why these additional data have limited impact, localized background covariance fields are computed, employing the spatial localization function of Gaspari and Cohn (1999) for temperature, dewpoint temperature, and the u- and υ-wind components from the EnKF ensemble. For the dryline event on 10–11 May 2010, the covariances are calculated between assumed observation locations ahead of and behind the dryline and all other points within the model grid. Results indicate that nonzero covariances have a large spatial extent, can be positive or negative, and may have detailed structure. Dewpoint temperature and υ-wind covariances are particularly illustrative for the point chosen ahead of the dryline, showing positive values nearby and weakly positive to negative values in the area behind the dryline (Figs. 18b,d). As for the point behind the dryline, dewpoint temperature and u-wind covariances are also large nearby whereas they are near zero in the region ahead of the dryline (Figs. 19b,c). Covariances with magnitudes above 0.1 extend beyond 200 km from the observation location for several variables at both locations, which is a greater distance than is found in the static covariances used by the RTMA for the same covariance value. Similar results are found in the suite of covariance values computed for the points inside and outside of the cold pool on 14 May 2010, and ahead of and behind the cold front on 25 November 2010. Although both large- and small-scale structures are evident in the covariance fields, the lack of impact shown in the RMS innovations and EnKF analyses when the Mesonet data are removed suggests that ensemble adjustments by the EnKF assimilation scheme are dominated by these larger-scale flow-dependent covariances that modify the background in a manner consistent with the model dynamics. This implies that when information from lower-density (i.e., METAR) observations is spread effectively over a large region, higher-density (i.e., Mesonet) observations are not needed to enhance the diagnosis of the surface conditions.
Localized background covariance values (see color bar) computed with respect to a location to the east of the dryline for (a) T (K), (b) Td (K), (c) 10-m u-wind component (m s−1), and (d) 10-m υ-wind component (m s−1) valid at 0000 UTC 11 May 2010. Black dot indicates the point where covariance values are computed.
Citation: Weather and Forecasting 28, 3; 10.1175/WAF-D-12-00078.1
As in Fig. 18, but computed with respect to a location to the west of the dryline. Black dot indicates point where covariance values are computed.
Citation: Weather and Forecasting 28, 3; 10.1175/WAF-D-12-00078.1
c. CONUS assimilation experiments
Current computational restraints limit the ability to use an EnKF approach on the CONUS scale with the finer grid resolution employed by the RTMA. However, grid spacings similar to that of the current North American operational model suite (i.e., NAM and RUC) are feasible to use with the EnKF data assimilation procedure utilized in this study. Thus, the CONUS grid described in section 2a is used to examine the impact of Mesonet data removal on a national scale and compare to the results from the original (1000 km)2 model domain.
RMS innovations for the period 0000 UTC 10 May–0000 UTC 13 May 2010 are again calculated for temperature, dewpoint temperature, and the u- and υ-wind components for experiments with full-data EnKF assimilation, 75% Mesonet data removal, and no data assimilation (Fig. 20). These results show that removal of a large percentage of Mesonet data still has minimal effect on the RMS innovations from the full and data-denial EnKF experiments, which largely overlap one another. The RMS innovations of both data assimilation experiments also show significant improvement over the experiment with no data assimilation. These results are consistent with those from the original model domain and continue to show that the removal of a large percentage of Mesonet data does not significantly impact the resulting analyses.
Domain-averaged RMS differences calculated from observations and the EnKF posterior ensemble-mean analysis (i.e., innovations) for (a) T, (b) Td, (c) 10-m u-wind component, and (d) 10-m υ-wind component for the period 0000 UTC 10 May–0000 UTC 13 May 2010 for the CONUS grid full-data assimilation, no assimilation, and 75% Mesonet data-denial experiment.
Citation: Weather and Forecasting 28, 3; 10.1175/WAF-D-12-00078.1
4. Summary and conclusions
The current expansion of mesonets across the United States provides a robust observational dataset for use in high-resolution meteorological analyses and prediction. Prior studies utilizing the EnKF have demonstrated the improvements in ensemble analyses and short-term ensemble forecasts gained through the assimilation of standard observational datasets, such as METAR and rawinsondes. The benefits of assimilating mesonet observations via a variational technique have been shown by the RTMA, but the impact of the mesonet data on the analysis accuracy remain largely undetermined. This study seeks to clarify the impact of mesonet data on analyses of 2-m temperatures, 2-m dewpoint temperature, and 10-m winds using an EnKF assimilation method on a (1000 km)2 domain centered over the central region of the United States for periods during the warm and cold seasons.
Results from the EnKF assimilation over 2-week experimental periods when compared to the RTMA reveal a general similarity between the analyses. The EnKF analyses are shown to be smoother with less small-scale variability than the RTMA, an observation supported by the lower energies in horizontal wavelengths less than 20 km indicated in the kinetic energy spectra. Inherent differences in the covariance structures and the assumed observational errors between the RTMA and EnKF analyses appear to be the primary factors in the variability observed between the analyses. Lower RMS innovations for the RTMA for temperature and dewpoint over most of the 2-week periods for both seasons imply a better overall fit to the observations. The comparable quality of the EnKF analyses versus the RTMA, coupled with kinetic energy spectra consistent with those of Skamarock (2004), point toward the feasibility of the EnKF method to test the impact of Mesonet observations on surface analyses. Three additional cases during the 2-week periods highlighted by a particular mesoscale feature are chosen to further compare analyses from the EnKF to those from the RTMA. Results from the three cases show that although differences within small regions exist between the EnKF analyses and the RTMA, both provide a reasonable depiction of specific mesoscale features and the overall surface environment.
Data-denial experiments completed for the first weeks of the warm and cold seasons reveal that removal of 75% of the Mesonet data imparts minimal degradation in the analyses, as evidenced from the comparison of the RMS innovations to the original full-data assimilation experiment. Both experiments, however, show significant improvement over the experiment with no data assimilation. Short-term data-denial experiments during two periods characterized by high mesoscale variability within the experimental domain continue to indicate little to no degradation in the analyses as increasing percentages of the Mesonet data are removed. Differences between the respective data-denial and full EnKF experiments are generally confined to the areas near the dryline in the warm season as well as near the cold front in the cold season and are very similar across the suite of data-denial experiments. Localized background covariances computed for the four pertinent variables within the environments associated with these features are shown to have large spatial scales, even in the presence of mesoscale features, that are much greater than the spatial separation between Mesonet stations. This may explain why the increased removal of these observations, even up to 75%, has little influence on the resultant analyses.
Experiments examining the impact of Mesonet data removal on a CONUS domain with grid spacing similar to that of the current operational North American model suite show similar results to those from the original model domain. Negligible analysis degradation is shown by the lack of differences between the RMS innovations from the full-data EnKF and the 75% Mesonet data-denial experiments, although improvement is once again shown over the no-data assimilation experiment.
Implications of the results presented herein are of particular importance to those tasked with the responsibility of interpreting meteorological analyses and assessing the value of different data assimilation schemes. Despite the slightly higher domain-averaged RMS innovations, the smoother ensemble-mean analysis from the EnKF data assimilation method appears to provide a clearer delineation of the general mesoscale pattern present in the domain. This will have significant impacts on short-term predictions where the correct assessment of mesoscale features is the primary element. The ability of the EnKF to create analyses consistent with the observed environment in both quiescent and variable mesoscale environments across the respective 2-week experimental periods is clearly demonstrated.
The lack of any significant degradation in the EnKF analyses as increasing percentages of Mesonet data are removed is the most surprising result of this study. While Mesonet data are clearly valuable to human forecasters in assessing the current state of the atmosphere, Mesonet data do not improve EnKF analysis accuracy on the mesoscale. This lack of impact is due in part to the large spatial scales of the localized background covariances. It may be that the hourly assimilation period does not take full advantage of the high temporal resolution of the Mesonet data. Current results also suggest that compressing the dense Mesonet observations into groups of superobservations via a superobbing technique akin to that discussed in Purser et al. (2000) for radar observations should be explored. Xu (2011) indicates that this method can reduce the total number of observations without information loss provided the procedure uses an optimally tuned background resolution. Future work examining EnKF data assimilation on meso- and convective scales should look toward increasing the frequency of assimilation and the use of superobservations to better utilize the National Mesonet data in the creation of analysis and subsequent forecasts from these analyses.
Acknowledgments
The authors thank Qin Xu, David Dowell, and the four anonymous reviewers for helpful comments that greatly improved this manuscript. We also thank Nusrat Yussouf for archiving the RTMA data and Dustan Wheatley for aid with analysis scripts. Funding for this research was provided by NOAA's National Weather Service Office of Science and Technology and NOAA's Office of Oceanic and Atmospheric Research under NOAA–University of Oklahoma Cooperative Agreement NA11OAR4320072, U.S. Department of Commerce.
REFERENCES
Anderson, J. L., 2001: An ensemble adjustment Kalman filter for data assimilation. Mon. Wea. Rev., 129, 2884–2903.
Anderson, J. L., 2009: Spatially and temporally varying adaptive covariance inflation for ensemble filters. Tellus, 61A, 72–83.
Anderson, J. L., and Collins N. , 2007: Scalable implementations of ensemble filter algorithms for data assimilation. J. Atmos. Oceanic Technol., 8, 1452–1463.
Anderson, J. L., Hoar T. , Raeder K. , Liu H. , Collins N. , Torn R. , and Avellano A. , 2009: The Data Assimilation Research Testbed: A community facility. Bull. Amer. Meteor. Soc., 90, 1283–1296.
Benjamin, S. G., and Coauthors, 2004: An hourly assimilation–forecast cycle: The RUC. Mon. Wea. Rev., 132, 495–518.
Benjamin, S. G., Moniger W. R. , Sahm S. R. , and Smith T. L. , 2007: Mesonet wind quality monitoring allowing assimilation in the RUC and other NCEP models. Preprints, 22nd Conf. on Weather Analysis and Forecasting/18th Conf. on Numerical Weather Prediction, Park City, UT, Amer. Meteor. Soc., P1.33. [Available online at https://ams.confex.com/ams/pdfpapers/124829.pdf.]
Betts, A. K., and Miller M. J. , 1986: A new convective adjustment scheme. Part II: Single column tests using GATE wave, BOMEX, and arctic air-mass data sets. Quart. J. Roy. Meteor. Soc., 112, 693–709.
Brown, J., and Coauthors, 2012: Rapid Refresh replaces the Rapid Update Cycle at NCEP. Preprints, 2012 Canadian Meteorological and Oceanographic Society Congress/21st Conf. on Numerical Weather Prediction/Conf. on 25th Weather and Forecasting, Montreal, QC, Canada, CMOS and Amer. Meteor. Soc., 3B1.2.
Chen, F., and Dudhia J. , 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.
Chen, S.-H., and Sun W.-Y. , 2002: A one-dimensional time dependent cloud model. J. Meteor. Soc. Japan, 80, 99–118.
Chou, M.-D., and Suarez M. J. , 1994: An efficient thermal infrared radiation parameterization for use in general circulation models. NASA Tech. Memo. 104606, Vol. 3, 85 pp.
Daley, R., 1991: Atmospheric Data Analysis. Cambridge University Press, 457 pp.
De Pondeca, M. S. F. V., and Coauthors, 2011: The Real-Time Mesoscale Analysis at NOAA's National Centers for Environmental Prediction: Current status and development. Wea. Forecasting, 26, 593–612.
Dudhia, J., 1989: Numerical study of convection observed during the Winter Monsoon Experiment using a mesoscale two-dimensional model. J. Atmos. Sci., 46, 3077–3107.
Errico, R. M., 1985: Spectra computed from a limited area grid. Mon. Wea. Rev., 113, 1554–1562.
Evensen, G., 1994: Sequential data assimilation with a nonlinear quasigeostrophic model using Monte Carlo methods to forecast error statistics. J. Geophys. Res., 99 (C5), 10 143–10 162.
Fujita, T., Stensrud D. J. , and Dowell D. C. , 2007: Surface data assimilation using an ensemble Kalman filter approach with initial condition and model physics uncertainty. Mon. Wea. Rev., 135, 1846–1868.
Gaspari, G., and Cohn S. E. , 1999: Construction of correlation functions in two and three dimensions. Quart. J. Roy. Meteor. Soc., 125, 723–757.
Grell, G. A., and Devenyi D. , 2002: A generalized approach to parameterizing convection combining ensemble and data assimilation techniques. Geophys. Res. Lett., 29, 1693, doi:10.1029/2002GL015311.
Hacker, J. P., and Snyder C. , 2005: Ensemble Kalman filter assimilation of fixed screen-height observations in a parameterized PBL. Mon. Wea. Rev., 133, 3260–3275.
Hong, S.-Y., Noh Y. , and Dudhia J. , 2006: A new vertical diffusion package with an explicit treatment of entrainment processes. Mon. Wea. Rev., 134, 2318–2341.
Horel, J., and Coauthors, 2002: Mesowest: Cooperative mesonets in the western United States. Bull. Amer. Meteor. Soc., 83, 211–225.
Houtekamer, P. L., and Mitchell H. L. , 1998: Data assimilation using an ensemble Kalman filter technique. Mon. Wea. Rev., 126, 796–811.
Janjić, Z. I., 1990: The step-mountain coordinate: Physical package. Mon. Wea. Rev., 118, 1429–1443.
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 surface layer in the NCEP Eta Model. Preprints, 11th Conf. on Numerical Weather Prediction, Norfolk, VA, Amer. Meteor. Soc., 354–355.
Janjić, Z. I., 2002: Nonsingular implementation of the Mellor–Yamada level 2.5 scheme in the NCEP Meso Model. NCEP Office Note 437, 61 pp.
Kain, J. S., 2004: The Kain–Fritsch convective parameterization: An update. J. Appl. Meteor., 43, 170–181.
McPherson, R. A., and Coauthors, 2007: Statewide monitoring of the mesoscale environment: A technical update. J. Atmos. Oceanic Technol., 24, 301–321.
Meng, Z., and Zhang F. , 2007: Tests of an ensemble Kalman filter for mesoscale and regional-scale data assimilation. Part II: Imperfect model experiments. Mon. Wea. Rev., 135, 1403–1423.
Meng, Z., and Zhang F. , 2008a: Tests of an ensemble Kalman filter for mesoscale and regional-scale data assimilation. Part III: Comparison with 3DVAR in a real-data case study. Mon. Wea. Rev., 136, 522–540.
Meng, Z., and Zhang F. , 2008b: Tests of an ensemble Kalman filter for mesoscale and regional-scale data assimilation. Part IV: Comparison with 3DVAR in a month-long experiment. Mon. Wea. Rev., 136, 3671–3682.
Miller, P. A., Barth M. F. , and Benjamin L. A. , 2005: An update on MADIS observation ingest, integration, quality control and distribution capabilities. Preprints, 21st Int. Conf. on Interactive Information and Processing Systems, San Diego, CA, Amer. Meteor. Soc., J7.12. [Available online at https://ams.confex.com/ams/pdfpapers/86703.pdf.]
Miller, P. A., Barth M. F. , Benjamin L. A. , Artz R. S. , and Pendergrass W. R. , 2007: MADIS support for UrbaNet. Preprints, 14th Symp. on Meteorological Observation and Instrumentation/16th Conf. on Applied Climatology, San Antonio, TX, Amer. Meteor. Soc., JP2.5. [Available online at http://ams.confex.com/ams/pdfpapers/119116.pdf.]
Mlawer, E. J., Taubman S. J. , Brown P. D. , Iacono M. J. , and Clough S. A. , 1997: Radiative transfer for inhomogeneous atmosphere: RRTM, a validated correlated-k model for the longwave. J. Geophys. Res., 102 (D14), 16 663–16 682.
National Research Council, 2009: Observing Weather and Climate from the Ground Up: A Nationwide Network of Networks. National Academies Press, 234 pp.
NWS, 1994: Technique specification package 88-21-R2 for AWIPS-90 RFP. Appendix G requirements numbers: Quality control incoming data, AWIPS Doc. TSP-03201992R2, NOAA/National Weather Service/Office of Systems Development, 39 pp.
Parrish, D. F., and Derber J. C. , 1992: The National Meteorological Center's Spectral Statistical–Interpolation Analysis System. Mon. Wea. Rev., 120, 1747–1763.
Pleim, J. E., 2007: A combined local and non-local closure model for the atmospheric boundary layer. Part I: Model description and testing. J. Appl. Meteor. Climatol., 46, 1383–1395.
Purser, R. J., Parrish D. F. , and Masutani M. , 2000: Meteorological observation data compression: An alternative to conventional “super-obbing.” NCEP Office Note 430, 12 pp.
Skamarock, W. C., 2004: Evaluating mesoscale NWP models using kinetic energy spectra. Mon. Wea. Rev., 132, 3019–3032.
Skamarock, W. C., Klemp J. B. , Dudhia J. , Gill D. O. , Barker D. M. , Wang W. , and Powers J. G. , 2008: A description of the Advanced Research WRF version 3. NCAR Tech. Note NCAR/TN-475+STR, 125 pp. [Available online at http://www.mmm.ucar.edu/wrf/users/docs/arw_v3.pdf.]
Stensrud, D. J., Bao J.-W. , and Warner T. T. , 2000: Using initial condition and model physics perturbations in short-range ensemble simulations of mesoscale convective systems. Mon. Wea. Rev., 128, 2077–2107.
Stensrud, D. J., Yussouf N. , Dowell D. C. , and Coniglio M. C. , 2009: Assimilating surface data into a mesoscale model ensemble: Cold pool analyses from spring 2007. Atmos. Res., 93, 207–220.
Thompson, G., Rasmussen R. M. , and Manning K. , 2004: Explicit forecasts of winter precipitation using an improved bulk microphysics scheme. Part I: Description and sensitivity analysis. Mon. Wea. Rev., 132, 519–542.
Torn, R. D., Hakim G. J. , and Snyder C. , 2006: Boundary conditions for limited-area ensemble Kalman filters. Mon. Wea. Rev., 134, 2490–2502.
Tyndall, D., and Horel J. , 2013: Impacts of mesonet observations on meteorological surface analyses. Wea. Forecasting, 28, 254–269.
Wheatley, D. M., and Stensrud D. J. , 2010: The impact of assimilating surface pressure observations on severe weather events in a WRF mesoscale ensemble system. Mon. Wea. Rev., 138, 1673–1694.
Wheatley, D. M., Stensrud D. J. , Dowell D. , and Yussouf N. , 2012: Application of a WRF mesoscale data assimilation system to springtime severe weather events 2007–09. Mon. Wea. Rev., 140, 1539–1557.
Whitaker, J. S., and Hamill T. M. , 2002: Ensemble data assimilation without perturbed observations. Mon. Wea. Rev., 130, 1913–1924.
Wu, W.-S., Purser R. J. , and Parrish D. F. , 2002: Three-dimensional variational analysis with spatially inhomogeneous covariances. Mon. Wea. Rev., 130, 2905–2916.
Xu, Q., 2011: Measuring information content from observations for data assimilation: Spectral formulations and their implications to observational data compression. Tellus, 63A, 793–804.
Zapotocny, T. H., and Coauthors, 2000: A case study of sensitivity of the Eta Data Assimilation System. Wea. Forecasting, 15, 603–621.
Zhang, F., Meng Z. , and Aksoy A. , 2006: Tests of an ensemble Kalman filter for mesoscale and regional-scale data assimilation. Part I: Perfect model experiments. Mon. Wea. Rev., 134, 722–736.
The RUC was replaced by Rapid Refresh (RAP; Brown et al. 2012) on 1 May 2012.
