Regional climate of hazardous convective weather through high-resolution dynamical downscaling Robert J. Trapp • Eric D. Robinson • Michael E. Baldwin • Noah S. Diffenbaugh • Benjamin R. J. Schwedler Received: 15 January 2010 / Accepted: 10 April 2010 Ó Springer-Verlag 2010 Abstract We explore the use of high-resolution dynami- cal downscaling as a means to simulate the regional cli- matology and variability of hazardous convective-scale weather. Our basic approach differs from a traditional regional climate model application in that it involves a sequence of daily integrations. We use the weather research and forecasting (WRF) model, with global reanalysis data as initial and boundary conditions. Horizontal grid lengths of 4.25 km allow for explicit representation of deep con- vective storms and hence a compilation of their occurrence statistics over a large portion of the conterminous United States. The resultant 10-year sequence of WRF model integrations yields precipitation that, despite its positive bias, has a diurnal cycle consistent with observations, and otherwise has a realistic geographical distribution. Simi- larly, the occurrence frequency of short-duration, poten- tially flooding rainfall compares well to analyses of hourly rain gauge data. Finally, the climatological distribution of hazardous-thunderstorm occurrence is shown to be repre- sented with some degree of skill through a model proxy that relates rotating convective updraft cores to the pres- ence of hail, damaging surface winds, and tornadoes. The results suggest that the proxy occurrences, when coupled with information on the larger-scale atmosphere, could provide guidance on the reliability of trends in the observed occurrences. Keywords Severe thunderstorm Heavy rainfall Dynamical downscaling Reanalysis Weather research and forecasting model 1 Introduction The regional climatology, temporal variability, and long- term trends of hazardous, local-scale phenomena such as tornadoes, hail, and damaging thunderstorm winds have been difficult to determine reliably, owing to the often- subjective nature of their observations (e.g., Diffenbaugh et al. 2008). This is one of the leading reasons why climate change assessments to date have avoided definitive con- clusions about how anthropogenic global warming (AGW) will, or perhaps already has affected these phenomena (e.g., Intergovernmental Panel on Climate Change 2007; US Global Change Research Program 2009). The use of climate models toward this end is possible, despite the small scale of convective precipitating storms (CPSs) relative to the grid lengths of such models. Recent applications have exploited the established relation between local convective-storm formation and organiza- tion, and the larger-scale or environmental vertical distri- butions of temperature, humidity, and horizontal wind (e.g., Weisman and Klemp 1982). Hereinafter, quantifica- tions of these atmospheric conditions are referred to as environmental controls, and include the vertical shear of horizontal wind over the lower half of the troposphere, and convective available potential energy (CAPE). The spatial analysis of the product of these parameters computed from global reanalysis data has been shown to bear ‘‘a strong R. J. Trapp (&) E. D. Robinson M. E. Baldwin B. R. J. Schwedler Department of Earth and Atmospheric Sciences, Purdue Climate Change Research Center, Purdue University, 550 Stadium Mall Drive, West Lafayette, IN 47906, USA e-mail: [email protected]N. S. Diffenbaugh Department of Environmental Earth System Science, Woods Institute for the Environment Stanford University, 473 Via Ortega, Stanford, CA 94305, USA 123 Clim Dyn DOI 10.1007/s00382-010-0826-y
12
Embed
Regional climate of hazardous convective weather through high-resolution dynamical downscaling
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Regional climate of hazardous convective weatherthrough high-resolution dynamical downscaling
Robert J. Trapp • Eric D. Robinson •
Michael E. Baldwin • Noah S. Diffenbaugh •
Benjamin R. J. Schwedler
Received: 15 January 2010 / Accepted: 10 April 2010
� Springer-Verlag 2010
Abstract We explore the use of high-resolution dynami-
cal downscaling as a means to simulate the regional cli-
matology and variability of hazardous convective-scale
weather. Our basic approach differs from a traditional
regional climate model application in that it involves a
sequence of daily integrations. We use the weather research
and forecasting (WRF) model, with global reanalysis data
as initial and boundary conditions. Horizontal grid lengths
of 4.25 km allow for explicit representation of deep con-
vective storms and hence a compilation of their occurrence
statistics over a large portion of the conterminous United
States. The resultant 10-year sequence of WRF model
integrations yields precipitation that, despite its positive
bias, has a diurnal cycle consistent with observations, and
otherwise has a realistic geographical distribution. Simi-
larly, the occurrence frequency of short-duration, poten-
tially flooding rainfall compares well to analyses of hourly
rain gauge data. Finally, the climatological distribution of
hazardous-thunderstorm occurrence is shown to be repre-
sented with some degree of skill through a model proxy
that relates rotating convective updraft cores to the pres-
ence of hail, damaging surface winds, and tornadoes. The
results suggest that the proxy occurrences, when coupled
with information on the larger-scale atmosphere, could
provide guidance on the reliability of trends in the observed
occurrences.
Keywords Severe thunderstorm � Heavy rainfall �Dynamical downscaling � Reanalysis �Weather research and forecasting model
1 Introduction
The regional climatology, temporal variability, and long-
term trends of hazardous, local-scale phenomena such as
tornadoes, hail, and damaging thunderstorm winds have
been difficult to determine reliably, owing to the often-
subjective nature of their observations (e.g., Diffenbaugh
et al. 2008). This is one of the leading reasons why climate
change assessments to date have avoided definitive con-
clusions about how anthropogenic global warming (AGW)
will, or perhaps already has affected these phenomena
(e.g., Intergovernmental Panel on Climate Change 2007;
US Global Change Research Program 2009).
The use of climate models toward this end is possible,
despite the small scale of convective precipitating storms
(CPSs) relative to the grid lengths of such models. Recent
applications have exploited the established relation
between local convective-storm formation and organiza-
tion, and the larger-scale or environmental vertical distri-
butions of temperature, humidity, and horizontal wind
(e.g., Weisman and Klemp 1982). Hereinafter, quantifica-
tions of these atmospheric conditions are referred to as
environmental controls, and include the vertical shear of
horizontal wind over the lower half of the troposphere, and
convective available potential energy (CAPE). The spatial
analysis of the product of these parameters computed from
global reanalysis data has been shown to bear ‘‘a strong
R. J. Trapp (&) � E. D. Robinson � M. E. Baldwin �B. R. J. Schwedler
Department of Earth and Atmospheric Sciences,
Purdue Climate Change Research Center, Purdue University,
550 Stadium Mall Drive, West Lafayette, IN 47906, USA
Supercells are not the only convective storms that possess
rotating updraft cores, however. Consider Fig. 4b, which
reveals a developing quasi-linear convective system. This
CPS morphology is: also well organized; frequently causes
wind damage and occasionally, tornadoes (Gallus et al.
2008; Trapp et al. 2005); and is known to possess multiple
rotating updraft cores, albeit not as intense and vertically
deep as those in supercells (Weisman and Trapp 2003).
Thus, we explore a proxy for severe thunderstorm
occurrence in the downscaled fields, given the existence of
April
May
June
d
e
f
occurrrence frequency of 25mm/1 hr rainfall
a
b
c
Fig. 3 Mean occurrence frequency of hourly rainfall [25 mm, a–c from the reanalysis-driven WRF model integrations (1991–2000), and d–ffrom hourly rain gauge data analyzed by Brooks and Stensrud (2000) (1948–1993), for the months of April, May, and June
R. J. Trapp et al.: Regional climate of hazardous convective weather
123
a rotating updraft core. Our occurrence proxy requires
gridpoint values of UH C 40 m2 s-2 (see Kain et al. 2008)
and Z [ 50 dBZ, which indicates convective precipitation.
These thresholds are based on a quantitative comparison
with observed reports of hail, severe convective winds, and
tornadoes (e.g., NCDC 2009). Comparisons are enabled by
assigning each report to the nearest WRF gridpoint (a
‘‘nearest-neighbor’’ interpolation), and then summing these
within coarsened grid cells of 38.25-km length. This coarse
grid length is approximately half that used in severe-
weather climatologies (Brooks et al. 2003a), and partially
compensates for known errors in report location (and time).
Occurrences of the model proxy are likewise coarsened.
Figure 5a–c depicts a monthly geographical progression
of simulated severe-thunderstorm frequency that is con-
sistent with the observed northward migration of severe
convective weather during these months. The maximum in
simulated severe-weather frequency generally envelops the
observed maximum (Fig. 5d, e). However, there is a ten-
dency for the model simulations to over-predict the report
frequencies during the months of May and June in the
Northern Great Plains (NGP), and then under-predict the
report frequencies during June in the eastern US. Indeed,
an evaluation of the RMSE between the simulated and
observed occurrences reveals a monthly increase in such
error within the warm season (Fig. 6). We attribute this in
part to the seasonal variation in the predominant CPS
forcing, from well-defined (and R1-resolved) synoptic-
scale weather systems down to weaker synoptic-scale
systems and also mesoscale systems that are not resolved in
the R1 data. Some of the discrepancies can also be attrib-
uted to the use of a constant UH–Z pair for the entire
period, which we did for the sake of consistency: Fig. 6
indicates that the UH–Z proxy has a monthly dependence,
with different pairs of these variables giving lower RMSE
values. We will continue to explore ways of improving this
severe-weather proxy; one possibility is suggested below.
Though helpful at a basic level, measures such as RMSE
can be misleading due to the sporadic nature of convective
weather occurrence and its fine-scale variations. For
example, a simulated occurrence maximum that is dis-
placed one grid cell from the observed maximum would be
Z UHa
b
Fig. 4 Example of a a supercell
thunderstorm and b quasi-linear
convective system included in
our set of simulations, as shown
by fields of simulated radar
reflectivity factor (Z), and by
markers where updraft helicity
(UH) locally exceeds 40 m2 s-2
R. J. Trapp et al.: Regional climate of hazardous convective weather
123
penalized with a large error, even though subjectively such
a small displacement would be considered acceptable and
reasonably accurate. Borrowing from the forecast verifi-
cation community, we employ the concept of ‘‘fuzzy
verification,’’ which relaxes the requirement of exact spa-
tial matches but still rewards proximity between the simu-
lations and verifying data (Ebert 2008). Our approach is to
map the modeled and observed severe weather occurrence
to increasingly coarse grids, and then compute a scale-
normalized similarity,
SNSn ¼ 1� RMSEn
RMSEn;ref
; ð2Þ
where n quantifies the coarseness of the grid relative to
native WRF grid, Oi,j and Mi,j are the observational and
modeled data at gridpoints (i, j), and thus
RMSEn ¼1
NxNy
XNx
i¼1
XNy
j¼1
On;i;j �Mn;i;j
� �224
35
1=2
; ð3Þ
and
RMSEn;ref ¼1
NxNy
XNx
i¼1
XNy
j¼1
O2n;i;j þ
XNx
i¼1
XNy
j¼1
M2n;i;j
!" #1=2
:
ð4Þ
Equation 2 is patterned after the fractions skill score
(FSS; see Roberts and Lean 2008), and like FSS, SNS
increases asymptotically as n and the corresponding grid
length (n 9 4.25 km) are increased over *2 orders of
magnitude (Fig. 7). SNS also has a dependence on month,
mean occurrrence frequency
JuneMayApril
Ob
serv
atio
ns
WR
F M
od
el
a b c
d e f
Fig. 5 Mean occurrence frequency of severe convective weather (hail, severe wind, and tornado) over the period 1991–2000, a–c from reported
occurrences, and d–f derived from the reanalysis-driven WRF model integrations, for the months of April, May, and June