please also see protocol.pdf
Experimental design and data
In order to test the effect of snow line position on extratropical cyclones, 20
North American cold season cyclones between 1986-2005 were simulated using the
Advanced Research core of the NCAR Weather Research and Forecasting model (WRF-ARW)
version 4.0.3 with perturbed snow cover extent (SCE). Four cyclone cases were
subjectively selected from each of the months from November through March based on
manual observational evaluation of all mid-latitude cyclones identified by
low-pressure centers through this period in daily surface and upper-level weather
charts. The criteria of selected cases required storm trajectories over or adjacent to
the Great Plains study area which resemble either the Alberta Clipper track or that of
the Colorado Low with lifetimes of at least 2 days, based on presence of well-defined
central minimum pressure. Cases were chosen until a sufficient variety of differences
in the lifetime minimum sea-level pressure (SLP) and magnitude of upper level forcings
in the form of 500 hPa height curvature and vorticity advection by the thermal wind
were found. Cases were simulated with observed initial conditions and validated
against observations using the 32-km spatial resolution North American Regional
Reanalysis (NARR) to ensure that WRF could accurately simulate each case.
Alterations to the SCE of each case were made by applying average poleward snow
line retreat (PSLR) from the 20-year periods of 1986-2005 (historical) to 2080-2099
(projected) for each of the five months examined in this study. Projected PSLR was
determined by examination of the grid cell snow mass change in 14 models of the 5th
phase of the Coupled Model Intercomparison Project (CMIP5; Taylor et al. 2012; doi in
Data Provenance table) wherein daily snow mass data were available and experiments
were conducted with two Representative Concentration Pathway forcings: RCP4.5 and
RCP8.5. Grid cells were identified as snow-covered if their simulated snow mass was at
least 5 kg m-2, which corresponds to typically 5 cm of snow depth (assuming a 10:1
snow to water ratio), sufficient to cover the surface. We did test other thresholds
and did not find a strong sensitivity to this choice in the projected snow cover maps.
The southernmost such grid cells were considered to comprise the snow line if the 5
degree span to the north of a cell had an average snow mass exceeding that threshold.
This search radius was employed in order to exclude outlying isolated southern patches
of snow. To limit artifacts that arise from small-scale variability in snow cover, a
600 km moving window average was then applied to all derived southern extent of snow
cover, hereafter referred to as the snow line. For each month, the 20-year average
snow line of the historical and projected periods was calculated, and the amount of
projected PSLR was determined from west to east in 30 km-wide bins across North
America. Different iterations, realizations, and physics options belonging to
experiments of the same model were combined in a one model, one vote scheme. With PSLR
calculated for both RCP forcings for each of the 14 models, each month contained 28
PSLR values from which the 10th, 50th, and 90th percentiles were determined.
The modeling effort involved simulating each of the 20 mid-latitude cyclone cases
with five degrees of snow line perturbation, each at five different initialization
times, from zero to four days prior to cyclogenesis, yielding a total of 500 distinct
simulations. One hundred simulations were generated without changes made to snow cover
(control). The remaining 400 runs imposed projected snow line changes of varying
magnitude (10th, 50th, and 90th percentiles) or complete snow removal across the
domain in order to determine the degree to which the position of the snow line
influences storms as opposed to that attributable solely to snow removal. Snow lines
for perturbed simulations were determined by applying values of PSLR to corresponding
30 km bins of the snow lines, as determined based on the method above, for each case
and removing all snow south of the new snow line except at altitudes greater than
2,000 m, where snowpack may persist even in warmer climates. It should be noted that
the removal of all snow south of the assigned snow line creates a discontinuous step
function in snow depth, a hard margin which is not necessarily characteristic of real
snow extent boundaries.
WRF model configuration
All WRF-ARW simulations were executed in the same domain comprised of the
continental United States (CONUS), central and southern Canada, northern Mexico, and
the surrounding oceans. 30 km horizontal resolution was used to best capture synoptic
scale transport with a 150 km buffer zone on each side and 45 vertical levels. Initial
and lateral boundary conditions were acquired from 3-hour NARR data provided in grib
format by NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, at
https://www.esrl.noaa.gov/psd/. Version 4.0 of WRF offers a CONUS suite of physics
options which was used in this experiment. The NOAH Land Surface Model (Noah LSM) was
altered to reduce surface snow accumulation to zero during simulation in order to
avoid snow deposition prior to the arrival of the cyclone of interest into the area
without removing precipitation in the atmosphere. The Noah LSM uses a single layer
snow model which calculates the albedo of the snow-covered portion of a grid cell
as
∝_snow= ∝_max A^(t^B )
where αmax is the maximum albedo for fresh snow in the given grid cell, t is the
age of the snow in days, and A and B are coefficients which are, respectively, 0.94
and 0.58 (0.82 and 0.46) during periods of accumulation (ablation). Coefficients A and
B were set to accumulation phase for simulations in every month except for March, when
the snow was considered to be ablating.
Data Information
Desciptions of variables within each netCDF file are available in the attached
file, WRF_output_variables.txt. The file may also be found at
http://co2.aos.wisc.edu/data/snowcover/outputs/WRF_output_variables.txt.
Alternatively, if users have netCDF installed, data for each file can be viewed in a
Linux terminal with the command:
$ ncdump –h filename
Filename Convention
case number_T-days initialized prior_perturbation degree_datetime string
e.g. Case00_T-1_fif_1993-01-30_03:00:00
