Numerical modeling of a historic storm: Simulating the Blizzard of 1888
Allison C. Michaelis,
Department of Marine, Earth, and Atmospheric Sciences, North Carolina State University, Raleigh, North Carolina, USA
Corresponding author: A. C. Michaelis, Department of Marine, Earth, and Atmospheric Sciences, North Carolina State University, 1125 Jordan Hall, Box 8208, Raleigh, NC 27695-8208, USA. (email@example.com)
 The National Oceanic and Atmospheric Administration/Cooperative Institute for Research in Environmental Sciences Twentieth Century Reanalysis (20CR) is used to explore the feasibility of high-resolution simulation of a historic extratropical cyclone event: The New England Blizzard of 1888. Using the 20CR as initial and lateral boundary conditions for the Weather Research and Forecasting model, a reasonable depiction of the cyclone is obtained, albeit displaced significantly to the north of the observed cyclone during the later stages of the event. Despite the position error, the simulated storm produces heavy snowfall over parts of New England and intense offshore cyclogenesis.
 Before the advent of routine radiosonde launches in the 1940s [Durre et al., 2006], only sparse upper air data were available from specialized field experiments. Without upper air measurements, numerical simulations of storms prior to the development of the upper air network do not appear feasible. Reanalysis projects, such as the National Centers for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR) reanalysis [Kalnay and Coauthors, 1996] often begin in the late 1940s and 1950s, when rawinsonde data coverage first became adequate. However, the National Oceanic and Atmospheric Administration (NOAA)/Cooperative Institute for Research in Environmental Sciences (CIRES) Twentieth Century Reanalysis (20CR) project has utilized recent advances in data assimilation to reconstruct three-dimensional atmospheric states from surface observations for the late nineteenth century and early twentieth century. These reanalyses are of great value in of themselves, [Fischer et al., 2013] but they also introduce the possibility of high-resolution mesoscale model simulations of historic storms [Hart, 2010; Becker et al., 2010]. In this study, we present a simulation of the blizzard from March 1888, known as the Blizzard of ‘88, using the 20CR as initial and lateral boundary conditions for the Weather Research and Forecasting (WRF) model. This method shows promise for this case, encouraging further research into the dynamics of other historical storms. This approach may be particularly useful to extend the period of record in the context of past and present climate change.
 The Blizzard of ‘88 was a monumental winter storm that resulted in heavy snowfall across the northeastern United States. Beginning over Georgia on 11 March 1888, a cyclone traveled northward along the Atlantic Coast, accompanied by rain, and later, heavy snow. As the storm passed through the Carolinas and into the northeast, it underwent explosive cyclogenesis, reaching peak intensity of 978 hPa over New England on the evening of 12 March [Kocin, 1988] and dumping heavy snow over this region. On 13 March, the cyclone looped counterclockwise off the New England coast and began to rapidly weaken from 978 to 1004 hPa in 24 h. By 14 March, the storm had moved off the coast and dissipated, leaving the northeast buried under snowfall totals estimated up to 127 cm, with drifts up to 15 m, and over 400 people dead [Kocin, 1988; Hughes, 1981]. The Blizzard of ‘88 exerted an extreme impact on some major metropolitan areas, including Washington, D.C. and New York City. The storm halted transportation and telegraph transmission, cutting these areas off from outside communication. As a result of this storm, underground telegraph communication systems were implemented a few years later [Hughes, 1981].
 The absence of robust surface and upper air observational data sets causes difficulties in piecing together a complete and accurate analysis of the Blizzard of ‘88, thus restricting the ability to model this historic storm. Intrigued by this challenge, motivated by the studies of Truchelut and Hart  and Hart , and the given the availability of a data set extending back to before the time of radiosonde networks, the goal of this project is to determine the feasibility of high-resolution simulations of the Blizzard of ‘88 using the 20CR and WRF model. The 20CR has proven to be a useful tool for researching historic weather as shown by Hart , who used the 20CR to simulate historic tropical cyclones. Here we seek to simulate a historic extratropical cyclone. Given the critical importance of preexisting upper level disturbances to extratropical cyclone development, we speculated a priori that it would prove difficult to produce a credible high-resolution simulation of such an event. During periods of overlap, the strong correlation between the 20CR upper tropospheric geopotential heights with those of the ERA-40, which assimilates radiosonde data, supports the reliability of the 20CR upper air fields for testing of the type presented here [Compo et al., 2011].
2 Data and Methods
2.1 Twentieth Century Reanalysis
 In order to represent the vertical structure of the atmosphere, most analysis and reanalysis techniques require the assimilation of rawinsonde, satellite, aircraft, radar, and other upper air measurements. Reanalysis data sets before the advent of routine upper air measurements are extremely difficult to produce.
 The Twentieth Century Reanalysis (20CR) is a pioneering global reanalysis data set that reconstructs the state of the atmosphere prior to the availability of radiosonde observations through the use of an ensemble Kalman filter data assimilation system in conjunction with a version of the Global Forecast System atmospheric model [Kanamitsu et al., 1991]. Surface and sea level pressure are used as input data to generate first guess fields, which are then used, with sea surface temperature and sea ice distribution fields, as boundary conditions [Compo et al., 2011]. Surface pressure observations have been recorded consistently since the late nineteenth century and can be used to deduce the barotropic flow, which is responsible for a large percentage of the total flow. Furthermore, surface pressure tendency provides information about the tropospheric flow. Therefore, the assimilation of surface pressure observations is a more reliable tactic for reproducing realistic tropospheric circulations than that of surface wind or temperature measurements [Compo et al., 2011; Whitaker et al., 2004]. Two versions of this data set were produced—the Twentieth Century Reanalysis version 1 and the Twentieth Century Reanalysis version 2 (20CRv2). This project uses 20CRv2 [Earth System Research Laboratory, NOAA, U.S. Department of Commerce, 2009], spanning the time period from 1871 to 2008 on a 2° by 2° spatial grid. This data set includes a 56 member filtered ensemble that provides meteorological fields and uncertainties every 6 h. Ensemble mean synoptic analysis and first guess forecast fields are also provided every 6 h and were used as initial and lateral boundary conditions for the model simulation. We utilized soil temperature and moisture, sea ice, and land use and vegetation categories from the 20CRv2. Other surface geophysical parameters were taken from the WRF preprocessing data sets, including terrain height, leaf area index, annual minimum and maximum vegetation fraction, and snow and background albedo.
2.2 Model Formulation
 The Weather Research and Forecasting (WRF) [e.g., Skamarock and Klemp, 2007] model version 3.2.1 was used to simulate the Blizzard of ‘88. Three domains were used, each integrated over 2.5 days with 3-hourly output. The outermost domain, with 54 km grid spacing, covers the region extending from 10°S–60°N and 30°–120°W, with one-way nested inner domains featuring 18 and 6 km grid spacing. The innermost (6 km) domain, featured in all graphics, extends from 23°–52°N and 62°–102°W.
 Experiments revealed improved results when the 20CRv2 synoptic first guess forecast sea surface temperature (SST) field was replaced with the Extended Reconstructed SST (ERSST) version 3. The ERSST recreates SST fields from 1854 to present using in situ SST data and statistics, allowing for reconstruction over regions with little data available [Smith and Reynolds, 2003]. While also on a 2° by 2° spatial grid, this data set exhibited a more sharply defined Gulf Stream, which improved the strength and location of the cyclone in our simulation.
 The model simulation utilized the Kain-Fritsch convective scheme, the Yonsei University boundary layer parameterization scheme, the Noah Land Surface Model [Harrold et al., 2011], and the WRF Single-Moment 6-Class Microphysics scheme. The simulation spanned 60 h, from 00:00 universal time coordinated (UTC) 12 March to 12:00 UTC 14 March in order to capture the development and intensity of the storm over New England. Model output was compared to manual analyses from Kocin and Uccellini  (hereafter KU04) and the March 1888 issue of Monthly Weather Review (American Meteorological Society [1888a, 1888b], hereafter MWR88). The analyses were based on surface data from 1888, which were collected 3 times a day—7:00 A.M. (12:00 UTC), 3:00 P.M. (20:00 UTC), and 10:00 P.M. (03:00 UTC the following day) [Kocin, 1983]; therefore, the simulation and analyses could only be directly compared at these times. The 20CRv2 analyses are available daily at 00:00, 06:00, 12:00, and 18:00 UTC. As a consequence, the position and intensity of the cyclone in the reanalyses, and therefore at the initial time of the simulation, could not be easily verified against existing manual analyses.
 In order to be considered a success, the simulated storm needed to reproduce the analyses of KU04 and MWR88 with reasonable accuracy for several key aspects of the Blizzard of ‘88: a minimum central pressure of 978 hPa, a surface anticyclone to the northwest of the storm center, a looping track near southern New England before moving offshore, and heavy snowfall over New England.
3.1 Cyclone and Anticyclone Characteristics
 In the analyses of KU04 and MWR88, at 03:00 UTC 12 March, when the storm began rapidly intensifying, it was located off the central North Carolina and southern Virginia coastline with an average central pressure of 1002 hPa (Figure 1a). The simulation at this time shows the cyclone located off the coast of the Carolinas with a central pressure of 1004 hPa (Figure 1b). The storm reached its minimum central pressure of 978 hPa on 13 March between 03:00 UTC and 12:00 UTC while located off the coast of New York and Pennsylvania as depicted by KU04 (Figure 2a). At 03:00 UTC 13 March, the simulation has the storm off the coast of New England with a central pressure of 976 hPa (Figure 2b). The location of the storm in the simulation was shifted compared with KU04; it was too far south by ~340 km at 03:00 UTC 12 March and too far north by ~250 km at the time of minimum central pressure (Table 1). The strength and timing of the cyclone minimum central pressure, however, compared favorably to the historical analyses (Figure 3).
Table 1. Distances Between the Simulation Track and the Analyzed Tracks and Difference in Minimum Central Pressure Between the Simulation and Analyses
Distance from KU04 (km)
Distance from MWR88 (km)
Difference in Minutes Central Pressure from KU04 (hPa)
Difference in Minutes Central Pressure from MWR88 (hPa)
03:00 UTC Mar 12
12:00 UTC Mar 12
03:00 UTC Mar 13
12:00 UTC Mar 13
03:00 UTC Mar 14
 The combination of an anticyclone to the northwest and a cyclone offshore resulted in a pressure gradient favorable for the advection of cold air southward into eastern New York and New England, where snowfall totals were greatest [Kocin, 1988]. In the analyses, this high pressure system was well organized and located over the Great Lakes throughout the duration of the event with a maximum central pressure of 1040 hPa during the peak intensity of the storm [Kocin, 1988]. In the simulation, while there was a surface high pressure system present over the Great Lakes, it was 4 to 6 hPa weaker at the time of peak cyclone intensity (03:00 UTC 13 March) (Figure 2). The 500 hPa geopotential height field (Figure 4) cannot be verified by analyses given the absence of an upper air network; however, given the placement and intensity of the surface cyclone and anticyclone, the 500 hPa geopotential height field exhibits a realistic cyclogenetic pattern.
3.2 Storm Track and Snowfall
 The approximate cyclone tracks for the KU04, MWR88, and the simulation were plotted using NCAR Command Language (NCL) (Figure 5) and distances between the simulation and the analyses calculated using the Haversine Formula (Table 1). Near the start of the model run, the storm is farther south than the analyses show; it is located off the southern coast of North Carolina rather than the southern coast of Virginia. From 03:00 UTC 12 March to 03:00 UTC 13 March, the model-simulated cyclone generally follows the analyzed track. After 03:00 UTC 13 March, however, the simulated storm travels north through Maine and into Canada rather than looping down to the southwest and remaining off the coast of New England as did the observed system. The largest difference in the tracks occurs between 03:00 UTC 13 March and 12:00 UTC 13 March. During this time, the simulated storm travels northwest roughly 350 km. The analyses, on the other hand, show the cyclone traveling north approximately 50 km during this time interval. While the simulated storm does show a westward trend toward the end of the model run, it dissipates over southern Quebec rather than moving out to sea. None of the model simulations attempted was able to replicate the characteristic looping track of the observed cyclone.
 KU04 and MWR88 showed the areas of maximum snowfall over portions of eastern New York and southern Connecticut with totals exceeding 50 in. (127 cm)(Figure 6a). Our model-simulated snowfall totals (Figure 6b) exhibit maximum snowfall over Vermont and New Hampshire with totals reaching only 40 in (101 cm). While these areas are located to the north of the analyzed areas of greatest snowfall, they are consistent with the simulated storm tracking farther to the north. Therefore, the difference in snow depths and coverage could be attributed to this error in track.
 This study presents a simulation of a historic storm from the presatellite, prerawindsonde era. The availability of the 20CRv2, extending back to before an upper air network was established, provided the possibility of simulating the Blizzard of ‘88 using a mesoscale model. While the storm was not perfectly depicted by the model, the simulation with the WRF mesoscale model did produce a credible storm mimicking many aspects of the Blizzard of ‘88, such as a minimum central pressure close to 978 hPa and heavy snow over parts of New England. Another aspect of the storm reproduced by the model was a substantial surface temperature gradient over the northeast United States with temperatures ranging from −17 °C over New York to 7 °C off the coast of New Hampshire at the time of peak intensity (not shown). The cyclone track exhibited large errors, featuring an average deviation of ~370 km from KU04 and ~430 km MWR88 (Table 1). Intensity errors relative to these analyses were smaller, averaging 5 and 4 hPa, respectively. These differences show that while the model simulation did not reproduce the correct track, it did depict the cyclone intensity throughout the event reasonably well. A limitation of our approach is that it does not take advantage of the ensemble information available in the 20CR; we note, however, that the ensemble spread in the 20CR is relatively large along the western flank of the upper trough associated with the cyclone (not shown).
 We theorize that the upper level wave associated with both the surface anticyclone and the primary cyclone contained errors that prevented the simulated storm from stalling over southern New England, New York, and Pennsylvania. As a consequence, simulated snowfall totals were not as high over these regions [Fischer et al., 2013]. The simulated anticyclone over the Great Lakes was weaker than analyzed, which is consistent with a weaker-than-observed upper wave amplitude and reduced chances of the upper trough becoming closed over the surface center. The data assimilation method used takes into account surface pressure, but the lack of upper air data, the coarse resolution of the data set, and other imperfections with our model configurations could have contributed to the model error. In future work, the 56 member ensemble utilized in the construction of the 20CRv2 could be utilized to run the full spectrum of simulations. Nevertheless, it is remarkable that the 20CRv2 ensemble mean was able to provide sufficiently accurate initial and lateral boundary conditions to allow a mesoscale simulation of the event that did reproduce an intense cyclone in approximately the right location and strength during the early portion of the cyclone event.
 If the Blizzard of ‘88 were to happen today, it is likely that it would not be as deadly of a storm due to advanced weather prediction technology [Kocin, 1988]. It is unclear, however, if the storm would be of similar strength or exhibit the same track under present day climate conditions. Further research could provide insight into the effects climate change has had on extratropical cyclone track and intensity over the past century. The method outlined in this study proved promising for future endeavors in reproducing and studying presatellite era storms, thus advancing our knowledge of past weather patterns.
 This research was supported by NSF grant 1007606, awarded to North Carolina State University. We thank NOAA/CIRES for developing and distributing the Twentieth Century Reanalysis. Support for the Twentieth Century Reanalysis Project data set is provided by the U.S. Department of Energy, Office of Science Innovative and Novel Computational Impact on Theory and Experiment (DOE INCITE) program, and Office of Biological and Environmental Research (BER), and by the National Oceanic and Atmospheric Administration Climate Program Office. The authors would also like to thank Bob Hart and one anonymous reviewer for their constructive comments on an earlier version of the manuscript, as well as Matthew Gilmore of University of North Dakota for guidance and suggestions. The WRF model was made available by National Center for Atmospheric Research (NCAR), which is sponsored by the National Science Foundation.
 The Editor thanks two anonymous reviewers for their assistance in evaluating this paper.