Content uploaded by Joaquim G. Pinto
Author content
All content in this area was uploaded by Joaquim G. Pinto
Content may be subject to copyright.
Meteorologische Zeitschrift, Vol. 14, No. 6, 823-838 (December 2005)
c
by Gebrüder Borntraeger 2005 Article
Sensitivities of a cyclone detection and tracking algorithm:
individual tracks and climatology
JOAQUIM G. PINTO∗1, THOMAS SPANGEHL1,2, UWE ULBRICH1,2and PETER SPETH1
1Institut für Geophysik und Meteorologie, Universität zu Köln, Germany
2Institut für Meteorologie, Freie Universität Berlin, Germany
(Manuscript receivedNovember 1, 2004; in revised form August 1, 2005; accepted August 1, 2005)
Abstract
Northern Hemisphere cyclone activity is assessed by applying an algorithm for the detection and tracking of
synoptic scale cyclones to mean sea level pressure data. The method, originally developed for the Southern
Hemisphere, is adapted for application in the Northern Hemisphere winter season. NCEP-Reanalysis data
from 1958/59 to 1997/98 are used as input. The sensitivities of the results to particular parameters of the
algorithm are discussed for both case studies and from a climatological point of view. Results show that the
choice of settings is of major relevance especially for the tracking of smaller scale and fast moving systems.
With an appropriate setting the algorithm is capable of automatically tracking different types of cyclones
at the same time: Both fast moving and developing systems over the large ocean basins and smaller scale
cyclones over the Mediterranean basin can be assessed. The climatology of cyclone variables, e.g., cyclone
track density, cyclone counts, intensification rates, propagation speeds and areas of cyclogenesis and -lysis
gives detailed information on typical cyclone life cycles for different regions. The lowering of the spatial and
temporal resolution of the input data from full resolution T62/06h to T42/12h decreases the cyclone track
density and cyclone counts. Reducing the temporal resolution alone contributes to a decline in the number of
fast moving systems, which is relevant for the cyclone track density. Lowering spatial resolution alone mainly
reduces the number of weak cyclones.
Zusammenfassung
Die nordhemisphärische Zyklonenaktivität wird ausgehend von Feldern des auf Meeresniveau reduzierten
Luftdrucks durch Anwendung eines Algorithmus zur Positions- und Zugbahnbestimmung synoptischskaliger
Zyklonen erfasst. Das Verfahren, welches ursprünglich für die Südhemisphäre entwickelt wurde, wird hier-
zu an die Verhältnisse des nordhemisphärischen Winters angepasst. Als Eingangsdaten werden NCEP-
Reanalysen für den Zeitraum von 1958/1959–1997/1998 verwendet. Die Sensitivität der Ergebnisse bezüglich
verfahrensspezifischer Parameter wird sowohl mit Blick auf einzelne Zyklone, als auch in klimatologischer
Betrachtungsweise untersucht. Die Einstellung der Parameter ist besonders für kleinräumige und schnelle
ziehende Systeme von Bedeutung. Durch eine angemessene Einstellung dieser Parameter ermöglicht der
Algorithmus die simultane Zugbahnbestimmung verschiedener Zyklonentypen: es können sowohl schnell
ziehende und sich rapide entwickelnde Systeme über den großen Ozeanbecken als auch kleinskaligere Zy-
klonen über dem Mittelmeer erfasst werden. Die Klimatologie der Zyklonenvariablen wie Zugbahndichten,
Zyklonenhäufigkeiten, Intensivierungsraten, Zuggeschwindigkeit, Zyklogenese- und Zyklolyseraten liefert
detaillierte Informationen über den typischen Lebenszyklus der Systeme in unterschiedlichen Regionen. Der
Effekt einer von T62/06H auf T42/12H reduzierten räumlichen und zeitlichen Auflösung der Eingangsdaten
äußert sich in einer Abnahme von Zyklonenzahlen und Zugbahndichten. Die Reduktion der zeitlichen Auf-
lösung allein trägt vor allem zu einer Abnahme der Anzahl von schnell ziehenden Systemen bei, welche
sich insbesondere in der Zugbahndichte auswirkt. Eine Reduktion der räumlichen Auflösung führt zu einer
verminderten Anzahl von schwachen Systemen.
1Introduction
Extratropical cyclones are a feature of dominant im-
portance for mid-latitude climate (PEIXOTO and OORT,
1992). An assessment of typical spatial and tempo-
ral characteristics (e.g., cyclogenesis, cyclolysis, growth
rates, etc.) can serve the understanding of physical
mechanisms associated with cyclones both in global and
regional terms (e.g., HOSKINS and HODGES, 2002). In-
∗Corresponding author: Joaquim G. Pinto, Institut für Geophysik
und Meteorologie, Universität zu Köln, Kerpener Str. 13, 50923
Cologne, Germany, e-mail: jpinto@meteo.uni-koeln.de
dividual tracks and developments can be used, e.g., in
studies of extreme events (e.g., ULBRICH et al., 2001).
A consideration of surface cyclones is, of course,
a restrictive view. The systems and their development
should be regarded as three-dimensional features of the
atmosphere. This suggests a manual synoptic analysis,
which is, however, rather time consuming and cannot
be consistently performed for large numbers of events,
e.g., as they are present in reanalysis data and in gen-
eral circulation models (GCM) simulations. Thus, sev-
eral numerical algorithms have been developed and
DOI: 10.1127/0941-2948/2005/0068
0941-2948/2005/0068 $ 7.20
c
Gebrüder Borntraeger, Berlin, Stuttgart 2005
824 J. G. Pinto et al.: Sensitivities of a cyclone detection and tracking algorithm Meteorol. Z., 14, 2005
used in order to objectively identify cyclones in digi-
tal maps. Examples for such efforts are described in pa-
pers by LAMBERT (1988); LETREUT and KALNAY
(1990); ALPERT et al. (1990); MURRAY and SIM-
MONDS, (1991, hereafter MS91); KÖNIG et al. (1993);
HODGES (1994); SERREZE (1995); HAAK and UL-
BRICH (1996, hereafter HU96); BLENDER et al. (1997);
SINCLAIR (1997) and LIONELLO et al. (2002). In most
cases, cyclone cores are defined in terms of pressure
minima at sea level, or minima in 1000 hPa geopotential
heights. Alternatively, cyclones can be defined in terms
of maxima in low level vorticity (e.g., HODGES, 1994).
A combination of both criteria was also used (KÖNIG
et al., 1993). However, the procedures vary greatly with
respect to computational details and the degree of so-
phistication involved.
The simplest approaches are based on the defini-
tion of a cyclone as a local mean sea level pressure
(MSLP) grid point minimum (e.g., LAMBERT, 1988).
This leads to the identification of numerous non signif-
icant “systems” which can be eliminated later on using
a minimum intensity criterion. For example, LETREUT
and KALNAY (1990) only regard grid points that lie 4
hPa beneath the mean pressure of 20 surrounding grid
points, while UENO (1993) demands a minimum value
of pressure gradient calculated over 8 surrounding grid
points. Using coarse grid input data, core positions re-
sulting from the procedures cannot be located between
these grid points, even if this seems apparent from the
data. This problem has led to the idea of transforming
the data to a finer grid before starting the identification
procedure (e.g., ALPERT et al., 1990; HODGES, 1994;
HU96), in spite of the considerable computational ef-
fort. The number and depth of systems is strongly depen-
dent on the resolution of the original data (e.g., HU96;
BLENDER and SCHUBERT, 2000; ZOLINA and GULEV,
2002). This effect is especially important for the com-
parison of cyclone statistics based on datasets with dif-
ferent resolutions and will be considered more closely in
section 6 of the current paper.
Methods searching for pressure minima tend to over-
estimate deep and mature cyclones while they miss
small scale systems that are better identified from their
local maxima in relative vorticity, e.g., fast moving sys-
tems or cyclones in the early and late stages of their
life cycle (MS91; HOSKINS and HODGES, 2002). The
search of vorticity maxima in cyclone identification was
thus used by a number of authors, either explicitly or
implicitly (MS91; KÖNIG et al., 1993; HODGES, 1994;
SINCLAIR, 1997). In a geostrophic sense, vorticity fo-
cuses on the small-spatial-scale end of the synoptic
range. Thus, the number of systems identified from this
quantity is much larger than what is obtained using sea
level pressure (HOSKINS and HODGES, 2002). How-
ever, vorticity maxima are not always connected with lo-
cal pressure minima. For this reason, MS91 considered
a (closed) cyclone only if the associated vorticity maxi-
mum could be assigned to a local pressure minimum.
The algorithms are usually tested in terms of their
ability to produce patterns of cyclone climatologies
which agree with synoptic experience. As an exception,
HU96 discuss deviations of extreme cyclones in terms
of core depth and location comparing their identified
cyclones with a manual analysis of SCHINKE (1992,
1993). The cyclone climatologies derived from observa-
tional (re)analysis data sets (e.g., SERREZE et al., 1997;
SICKMÖLLER et al., 2000; GULEV et al., 2001; GENG
and SUGI, 2001; HOSKINS and HODGES, 2002; LAM-
BERT et al., 2002) generally agree in producing maxima
of cyclone counts over the Northern Hemisphere (here-
after NH) mid-latitude ocean basins. These areas cor-
respond closely to the storm tracks, regions of strong
height-field variance in a frequency band associated with
synoptic time scales (as defined by BLACKMON, 1976).
These areas of high synoptic activity over the North At-
lantic and the North Pacific are hereafter refereed to as
‘main storm tracks’. Most climatologies tend to iden-
tify rather few cyclones over secondary storm track re-
gions (e.g., the Mediterranean basin), which is in con-
trast with methods particularly tuned for these regions
(e.g., TRIGO et al., 1999, 2000) and synoptic experience.
Cyclones in these areas are, however, of major interest
as they can have large regional impacts.
The tracking of cyclones is usually performed by as-
signing an individual system identified at one particular
date to a successor at the subsequent date. The assign-
ment of the most likely successor involves a search in a
certain area. This area is sometimes chosen as a circle
around the original position (e.g., ALPERT et al., 1990;
KÖNIG et al, 1993; UENO, 1993; BLENDER et al, 1997);
it can also be defined from a local “climatological steer-
ing velocity” (e.g., MS91, HU96). Other authors (e.g.,
HODGES, 1994) have included acondition for “flat” tra-
jectories in order to agree with synoptic experience.
In the present study, we use an algorithm originally
developed for the identification and tracking of cyclones
in the Southern Hemisphere (cf. MS91; SIMMONDS et
al., 1999, hereafter SML99) and apply it to the NH. The
main goal is to obtain a cyclone climatology which suc-
cessfully includes the whole range of NH cyclones: the
large quasi stationary and transient oceanic cyclones,
fast moving storms and smaller systems over secondary
storm track areas. The original settings (optimised for
Southern Hemisphere long-living large scale cyclones)
are not suitable to identify all these types of systems,
and thus the new choices and the sensitivities of several
of the algorithm’s parameters areexplained in this study.
As a result, we present a new cyclone climatology and
show the quality of the algorithm’s parameter settings
by suitable case studies.
Meteorol. Z., 14, 2005 J. G. Pinto et al.: Sensitivities of a cyclone detection and tracking algorithm 825
The structure of this paper is as follows: The sec-
ond section gives a short description of the data used,
while the following ones deal with the different steps
of the method: cyclone identification (third) and feature
tracking (fourth). These two sections include a technical
description of the methods and several sensitivity tests.
The resulting cyclone climatology is presented in sec-
tion five. The sixth section deals with the impact of res-
olution of the input datasets on the cyclone climatology.
A short discussion concludes this paper.
2Data
We use the NCEP reanalysis (KALNAY et al., 1996)
for the period 1958/59–1997/98. One of the main ad-
vantages of reanalysis datasets is that model parame-
terisations and resolution are unchanged for the com-
plete time period. Another advantage compared to op-
erational analyses is that assimilated observations in-
clude late observational reports. Even though the NCEP
dataset is available from 1948 on, inhomogeneities have
been pointed out for thefirst decade (cf. KISTLER et al.,
2001). Thus, only the period 1958 to 1998 is consid-
ered. Within this period, some systematic problems in
monthly surface pressure have been detected for the time
period before 1967 (REID et al., 2001), but we found no
evidence of a strong impact on our results. The spec-
tral horizontal resolution of the numerical model (T62)
and the spatial and temporal resolution of the gridded
data (2.5◦x 2.5◦, available for 00, 06, 12, 18 UTC) does
not allow a good coverage of very small and short lived
systems, even if they were originally detected by the ob-
servational network.
In order to test the sensitivity of the identified cy-
clones with respect to data resolution, we created artifi-
cial MSLP datasets for NCEP with reduced a) spectral
(T42) b) temporal (12 hours) and c) both spectral and
temporal (T42/12h) resolutions. This choice of values is
motivated by the resolution of many current GCM simu-
lations. The reduction of the spatial resolution involved
an interpolation to the original Gaussian grid (T62) and
the removal of the high spectral wave numbers. Finally,
the truncated spectral data were retransformed onto the
regular 2.5◦x 2.5◦grid. We found that the effects of the
interpolation on MSLP could hardly be noticed when
subjectively comparing the respective maps. The reduc-
tion of temporal resolution was done by omitting the 06
and 18 UTC values.
The verification of cyclone positions and tracks
is based on operationally produced weather maps
(“Berliner Wetterkarte” issued by “Freie Universität
Berlin”, “Europäischer Wetterbericht” issued by the
GERMAN WEATHER SERVICE) with a focus on the At-
lantic and European/Mediterranean area. The maps give
information of sequences of cyclone positions based on
a synopsis of both surface and upper air data.
3Cyclone identification
The initial step of the feature identification is a transfor-
mation of the gridded MSLP fields to a regular 0.75◦x
0.75◦grid by a polar stereographic projection via bicu-
bic spline interpolation. This procedure does not add any
information to the original data, but it permits cyclones
with cores to be located between the original data grid
points. We found that the use of spline interpolation (see
also MS91 and HU96) improves the localisation of the
cyclones for the majority of cases. The positive effect is
particularly strong for mature cyclones cores with slack
gradients near their centre and for small scale cyclones.
In a second step, the high resolution grid is scanned for
local maxima of quasi-geostrophic relative vorticity (
ξ
)
via the laplacian of pressure (∇2
p).
ξ
is closely related to
∇2
paccording to
ξ
=1
ρ
·f∇2p(3.1)
where
ρ
is the air density, f the Coriolis parameter and
p stands in our case for MSLP. In a third step, the al-
gorithm iteratively searches the interpolated field for a
pressure minimum in the vicinity of the ∇2
pmaximum
in order to associate the latter with a “real” low pres-
sure core. If such a minimum is found, the cyclone is
classified as a closed system, with its core located at
the pressure minimum. If the search is not successful
within a distance of about 1200 km (the distance equiv-
alent to 12◦in latitude), a second search is performed
for the point with the minimum pressure gradient (in-
flection point), and the system is classified as an open
depression. More details on the method can be found in
MS91 and SML99.
The final step of the identification procedure is the
removal of systems on the basis of thresholds. Here-
after, the names of the corresponding parameters are
given in italics and correspond to those in the program
code and in SML99. The chosen values for these pa-
rameters (Tab. 1) have been carefully tested considering
more than one hundred individual cyclones in different
regions, but only a few examples can be presented here
for illustration.
– No cyclones over high ground: Vorticity max-
ima and lowsare sometimes artificially introduced
into the MSLP data due to extrapolation below
high ground. We found it useful toremoveall sys-
tems identified at grid points with surface heights
of more than 1500 m (zsmax, Tab. 1) above sea
level. On the other hand, cyclones close to the
high orography barriers often proved to be real
in areas with a dense observational network (e.g.,
Genoa Gulf, southern Alps) and thus we did not
follow SML99 in suppressing them systematically
(ftopeq).
826 J. G. Pinto et al.: Sensitivities of a cyclone detection and tracking algorithm Meteorol. Z., 14, 2005
a)
W1
W2
b)
Figure 1: Synoptic situation over the North Atlantic and Europe for February 28th, 1990, 06 UTC (a): weather chart taken from “Berliner
Wetterkarte”, including hand analysed MSLP contours (black lines, interval: 5.0 hPa), cyclones (marked L) and fronts. (b): MSLP (black
lines, interval: 5.0 hPa) and laplacian of MSLP(grey lines, interval: 1.0 hPa/(deg. lat.)2) from NCEP-Reanalysis data (T62) on a 0.75 deg.
lat. polar grid including identified systems. Large (small) symbols indicate strong (weak) identified system, and closed (open) ones are
indicated by circles (triangles). Systems referred to in the text are denoted by W (Fig. a) and W1, W2 (Fig. b).
– Minimum intensity: Spurious or artificial lows are
mostly characterised by comparatively low values
of the ∇2
p. The algorithm considers spatial aver-
ages of this quantity computed in an area around
the low pressure core (and not around the ∇2
p
maximum as suggested by SML99). The choice
of threshold values (cmnc parameters for closed
and for open systems) and of the averaging radius
(cvarad) must make sure that small scale lows
and cyclones in the initial stage of their life cycle
are kept. Note that Mediterranean cyclones have
a typical horizontal scale of 300 km (TRIGO et
al, 1999). Thus, a small averaging radius equiva-
lent to 4◦in latitude (about 400 km) was chosen
(Tab. 1). Smaller values were also tested, but ma-
ture cyclones with slack gradients near the centre
would no longer be well represented. On the other
hand, higher values imply a bias towards large
steering cyclones. With respect to the thresholds
for ∇2
p(see Tab. 2), we have chosen 0.2 hPa/(deg.
lat.)2for open systems and 0.1 hPa/(deg. lat.)2for
closed systems (cmnc1, Tab. 1).
– Minimum distance: The presence of vorticity
maxima along frontal zones of major cyclones
and also in the vicinity of orography frequently
leads to chains of open systems. Many of them
are not of any importance. We demand that only
the strongest system within a radius of 3 degrees
latitude is included (diflt1). This condition was
not necessary in SML99 because of the smooth-
ing they used in the input data.
Meteorol. Z., 14, 2005 J. G. Pinto et al.: Sensitivities of a cyclone detection and tracking algorithm 827
Table 1: Parameter settings for cyclone identification (upper) and tracking (lower) after SML99 and the present study. Names correspond
to those given in MURRAY and SIMMONDS 1991. The last column gives a short explanation of the parameter. (dg) units: deg. lat.; (hdg)
units: hPa/(deg. lat.)2; (#) Average radius calculated centred ∗around pressure minimum ∗∗ around vorticity maximum. Values without units
are non-dimensional. For further details see text.
Parameter
SML99
Current
Explanation
data set
GASP R53
NCEP T62
original data resolution
polar grid
2.0 (1.0) dg
0.75 dg
interpolated data resolution
smoothing
yes (2.0 a dg)
no
spatial smoothing (radius)
diflt1
no
3.0 a dg
min. distance between 2 cyclones
cvarad (#)
2.0 dg **
4.0 dg *
averaging radius for laplacian value
cmnc1
0.2 hdg
0.1 hdg
min. laplacian value, open systems
cmnc2
0.2 hdg
0.2 hdg
min. laplacian value, closed systems
cmncw
0.7 hdg
0.6 hdg
laplacian value, strong/weak
zsmax
1000 m
1500 m
max. height for cyclone
ftopeq
yes (0.005 hdg)
no
orography gradient condition
time step
6 /12 h
6 h
time resolution of input data
steering
velocity
background
flow
background
flow
type of steering velocity
rdpgrd
5.0 dg
4.0 dg
radius for averaging steer. velocity
fsteer
2.0
2.25
scaling factor for steer. velocity
wsteer
0.6
0.4
weighting steering/prev. movement
wpten
not used
0.8
weighting factor for pressure tendency
dequiv
not used
0.4 dg
deg. let. equivalent to 1mb pressure diff
rcprob
12.0 dg
12.5 dg
search radius for next position
rpbell
1.0
0.4
shape factor of cost function
qmxopn
0.75
0.6
devaluation value for open systems
qmxwek
0.50
0.6
devaluation value for weak systems
qmxnew
0.75
0.6
devaluation value for new systems
Examples illustrating successful cyclone identifica-
tion are given below. The first case (Fig. 1) refers to the
synoptic situation of February 28th, 1990, 06 UTC and
shows the importance of the identification of open sys-
tems. The weather situation is dominated by a mature
vortex centred over Scandinavia (Fig. 1a). Out of the two
secondary cyclones over the eastern North Atlantic the
latter (noted W) developed into a storm cyclone called
“Wiebke” by the German Weather Service. This storm is
of relevance as it produced major damage in central Eu-
rope later on the same day (e.g., KLAWA and ULBRICH,
2003). In the weather chart shown in Fig. 1a, cyclone W
is still rather shallow (closed 990 hPa isobar, near Ire-
land). Fig. 1b shows the MSLP and ∇2
pfields based on
the interpolated NCEP data for the same date, with the
objectively identified systems marked by symbols. With
respect to the storm cyclone W, no closed isobar is found
in the NCEP data, in contrast to the map drawing (Fig.
1a). Instead, there are two open systems W1 and W2 in
this area. Thus it would have been impossible to identify
the storm at this date when rejecting open systems.
The second example deals with smaller scale systems
over the Mediterranean and aims at demonstrating the
importance of the spline interpolation to a finer grid.
Fig. 2a shows the weather map for March 26th, 1993,
12 UTC. The synoptic situation is dominated by a cy-
clone in the Gulf of Lions, with frontal systems extend-
ing from North Africa over the Balkans up to the Black
Sea. The cyclone was formed in the lee of the Alps and
represents the type of a Genoa cyclone. Fig. 2b shows
the MSLP and ∇2
pfields for the same analysis time af-
ter spline interpolation. The identified Genoa cyclone’s
core (cyclone A) is in very good agreement with the
hand analysed cyclone position (cf. Fig 2a), while a band
of high ∇2
pvalues between the Gulf of Lions and the
Balkans coincide with the frontal system. Other systems
are identified within this zone, as the vorticity maxima
can be attributed to pressure minima (e.g., cyclone B)
and frontal wave (open) disturbances. When consider-
ing the coarse grid, the position of the Genoa cyclone
is not correctly identified, even though there is a maxi-
mum of ∇2
pclose to the observed cyclone core (Figure
not shown). A detailed analysis of this case reveals that
the algorithm identifies an open cyclone near the cor-
rect position (corresponding to position A in the finer
grid) in a first step, but it is later rejected as its position
is too close to the other cyclone core (corresponding to
position B in the finer grid) given the minimum distance
criterion.
The set of cyclone cores resulting from the identifi-
cation procedure includes a large number of artificial or
irrelevant systems. Such systems can largely be elimi-
nated by applying a smoothing procedure as suggested
by SML99. Smoothing, however, also eliminates smaller
828 J. G. Pinto et al.: Sensitivities of a cyclone detection and tracking algorithm Meteorol. Z., 14, 2005
a)
A
B
b)
Figure 2: Synoptic situation over the Mediterranean Basin for 26.03.1993, 12 UTC (a): weather chart taken from the “Europäischer
Wetterbericht”, including hand analysed MSLP contours (black lines, interval: 5.0 hPa), cyclones (L) and fronts. (b): MSLP (black lines,
interval: 5.0 hPa) and laplacian of MSLP (grey lines, interval: 1.0 hPa/(deg. lat.)2) from NCEP-Reanalysis data (T62) on 0.75 deg. lat. polar
grid including identified cyclones. Large (small) symbols indicate strong (weak) identified system. Additionally, closed (open) systems are
indicated by circles (triangles). Systems mentioned in the text are denoted in the lower panel by A and B.
scale cyclones over secondary storm track regions. We
decided not to smooth the data, accepting about 40 %
more identified systems than what was produced with
the SML99 settings. This enhanced number is useful in
respect to the subsequent cyclone tracking as discussed
below.
4Cyclone tracking
The algorithm determines cyclone tracks based on the
results of the identification scheme. For each identified
cyclone, the algorithm predicts a subsequent position
and core pressure. The identified cyclones in the follow-
ing time step which are located in the vicinity of the sug-
gested position are examined and the most likely candi-
date is chosen. The estimation of the subsequent position
uses a ‘‘prediction velocity” upred
upred = (1−wsteer)·uM +wsteer ·(fsteer·uS)(4.1)
where upred is an average of velocity deduced from the
“previous displacement” uM and a “geostrophic steer-
ing velocity” (fsteer*uS) term. The relative weight of
both terms is given by the factor wsteer. The steering
velocity at the surface level uS is calculated from an
averaged pressure gradient around the centre of the cy-
clone over a radius of 4◦of latitude (rdpgrd) for the ac-
tual date. This is a simplification against SML99 who
had also taken thesubsequent date into account. It is not
attempted to take into account steering by upper-level
flow explicitly (as suggested by HOSKINS and HODGES,
2002) in order to minimize data requirements. SML99
found that the additional inclusion of upper level data
does not significantly change the performance of the al-
gorithm. Instead, they multiply the value of uS with a
factor (fsteer) to account for the increasing wind speed
with height, for which we have chosen a value of 2.25.
The weighting factor wsteer is set to 0.4 (see Tab. 1).
This implies a larger contribution of the displacement
term than of steering velocity term to upred.
Meteorol. Z., 14, 2005 J. G. Pinto et al.: Sensitivities of a cyclone detection and tracking algorithm 829
Figure 3: Examples of individual cyclone tracks corresponding to known winter storms in early 1990. (a) and (b): Vivian; (c) and (d):
Wiebke; (e) and (f): Hertha. Left column shows tracks computed with the final settings, right column with settings after SML99. Large
(small) symbols indicate strong (weak) identified system. Additionally, closed (open) systems are indicated by circles (triangles). The
computed lifetime of the individual cyclone is indicated below each panel.
The probability of association between cyclones
identified in the current chart, and all possible candidates
as their successors in the subsequent chart, is calculated
using a cost function. This function (now with shape fac-
tor rpbell =0.4, see MS91), involves the distance from
the estimated position (looking for cyclones within a dis-
tance of rcprob =12.5◦of latitude), and the difference
in core pressures (including an estimate of core pres-
sures from the previous pressure tendency, controlled
by factors wpten and dequiv, see Tab. 1). The associa-
tions between predicted cyclones and possible succes-
sors are sorted into groups (cf MS91, Fig. 8). The sort-
ing into groups can be understood as a spatial cluster-
ing in order to help a correct assignment of the cyclones
and to minimize computational costs. For each group,
the most probable combination of associations is deter-
mined. Systems not paired up inthis process have either
just emerged (cyclogenesis) or ceased to exist (cycloly-
sis). Cyclone splitting and merging is not permitted in
the scheme. Instead, the cyclone track with the closest
similarity according to the cost function is continued.
Further details of this part of the tracking methodology
830 J. G. Pinto et al.: Sensitivities of a cyclone detection and tracking algorithm Meteorol. Z., 14, 2005
0
2
4
6
8
10
1 2 3 4 5 6 7 8 9
Lifetime [days]
[N]
a)
0
1
2
3
4
1 2 3 4 5 6 7 8 9
Lifetime [days]
[N]
b)
0
2
4
6
8
00 10 01 11
Cyclone Class
[N]
c)
Figure 4: Accumulated number (N) of tracks versus lifetime for (a):
all tracks; (b): remaining tracks after removal of short lived systems
that never reached a closed status and an intensity exceeding the
threshold of strong cyclones. Further explanations see text. (c): ac-
cumulated cyclone counts (identified systems) for different cyclone
classes (00: strong and closed, 10: weak and closed, 01: strong and
open, 11: weak and open, see Tab. 2). Grey bars correspond to (a),
black bars to (b).
are found in MS91 and SML99. Regarding the filtering
parameters for open and new systems (qmx parameters,
see Tab. 1, SML99), we have chosen a 10 % higher dis-
regard of weak systems, in order to favour a continuation
of weak cyclone tracks towards a strong system (instead
of a weak one)
Objectively identified cyclone tracks were exten-
sively compared with hand analysed tracks from the syn-
optic charts and original SLP fields. The results were
also compared to calculations performed with original
settings (SML99) (cf. Tab. 1). The overall result was
that sensitivities with respect to the choice of the set-
tings are generally very small for large oceanic cyclones
where the assignment is not ambiguous. They can, how-
ever, be very large for secondary small scale systems,
for fast moving systems, and for the early and late parts
of the systems’ life cycles. In these cases, the result is
also heavily dependent on the identification procedure,
in particular to the question if smoothing (as in SML99)
is involved or not. The tracks of three cyclones that even-
tually produced gales over Central Europe in early 1990
(Fig. 3) provide examples for the tracking’s sensitivities.
The track of the storm called “Vivian” (Fig. 3a) as
identified with the current setting is in agreement with
results obtained by a manual synoptic analysis. How-
ever, an identification of cyclone cores based on the
SML99 settings (Fig. 3b) leads to a association with
the track of an open system east of the Great Lakes.
This is due to the less restrictive probability function
(rpbell =1.0, see Tab. 1) and data smoothing in the
SML99 settings. The latter manipulation removes some
secondary systems which are needed for obtaining cor-
rect associations. Towards the end of the cyclone’s life-
time, smoothing causes the system to vanish too soon.
Very similar effects are found for the storm “Wiebke”
which was already introduced in section 3. Again, the
track obtained with the current setting (Fig. 3c) is in
good agreement with the hand analysis from synoptic
weather maps. The track based on the SML99 setting
begins earlier and also ends earlier because of problems
with the correct identification and subsequent associa-
tion of the system at weak stages ofits lifetime (Fig. 3d).
Wiebke’s track can only be identified over western Eu-
rope when taking into account that it is an open cyclone
in the middle of its life cycle, at February 28th, 1990,
18 UTC (see also Fig. 1). This is the case both with and
without application of smoothing.
Not unexpectedly, there are particular cases when our
settings perform worse than the original ones. In the case
of the storm called “Hertha”, the SML99 setting pro-
vides a more accurate beginning of the track (Fig. 3f). In
this case, the scheme without smoothing produces many
very weak systems, which turn out to be weaker than
the required threshold. Thus, no systems are identified
at January 31th, 1990, 18 UTC and February 1th, 1999,
00 UTC (Fig. 3c).
In spite of the problems in this particular case, we
found that the current settings produce fewer errors over
the North Atlantic area than the SML99 settings, which
is mostly caused by the effects of smoothing on the
Meteorol. Z., 14, 2005 J. G. Pinto et al.: Sensitivities of a cyclone detection and tracking algorithm 831
a) b)
Figure 5: Cyclone counts of cyclones for (a): remaining cyclones after the removal of short-lived systems that never reached a closed
status and an intensity exceeding the threshold of strong cyclones (interval: 6.0 cyclone days/winter, areas above 12.0 are shaded grey).
(b): removed cyclones (interval: 10.0 cyclone days/winter), areas with differences above 99 percentile value are shaded (t-test based on
monthly basis). Values are calculated for 10.0◦x 5.0◦lon.-lat. grid boxes. Isolines and shadings in areas with orography above 1500 m are
suppressed.
smaller scale cyclones. The performance of the cyclone
tracking algorithm is most sensitive to the shape of the
probability function. Sensitivity tests show that larger
values for the maximum radius (rcprob) combined with
a slow decreasing bell shaped function (rpbell) lead to
the best results for fast moving cyclones. Another focus
of the analysis was on the Mediterranean Basin, which
is an example of a secondary storm track region. Again,
most of the problems encountered with the tracking
were related with the identification of “raw” cyclones
in the previous step rather than with the chosen tracking
settings (results for different track settings not shown).
The most sensitive tracking parameters were qmx terms
(preference of open/closed systems, see Table 2) and
particularly the search radius (rcprob). A value not too
large for rcprob is helpful for the correct tracking of sys-
tems, especially for regions with weak SLP gradients.
In order to account for both these smaller scale systems
and the faster oceanic cyclones, an intermediate value of
12.5 degrees (referring to latitude) for the search radius
was chosen.
The list of tracks still contains a large number of spu-
rious systems. Cyclones with a lifetime of less than one
day are in fact dominating the statistics (Fig. 4a). We
have thus imposed a set of conditions as the final step
of the tracking procedure that serve to eliminate them.
A cyclone remains in the list of events if it (a) has a
lifetime of at least 24 hours and (b) has been classified
as a closed and intense system (class 00, ∇2
plarger than
0.6 hPa/deg.lat.2, see Table 2) at least once in its life-
time. The second restriction clearly reduces the number
of systems with short lifetimes (1–3 days, Fig. 4b), but
it has comparatively little effect on systems tracked for
Table 2: Classification of systems in terms of circulation
(open/close) and of values of ∇2
pin hPa/(deg. lat.)2. Identified
closed cyclones whose strength does not reach 0.1 hPa/(deg. lat.)2
are eliminated. For open cyclones, minimum allowed strength is 0.2
hPa/(deg. lat.)2.
Class Type ∇2
p
00 Closed x >0.6
01 Open x >0.6
10 Closed 0.1<x≤0.6
11 Open 0.2<x≤0.6
a larger number of days: For 6–10 days lifetime the re-
duction is about 10–20 %. Fig. 4c displays that the re-
duction hardly affects the number of closed and intense
systems, as they do not usually occur on time scales be-
low one day, which is in agreement with synoptic expe-
rience. Weak (strong) open systems are reduced by 76 %
(68 %), which is reasonable assuming that most identi-
fied open systems are spurious. Weak closed systems are
also reduced by 65 %. With respect to core pressure val-
ues, we found that the removed systems are generally
very shallow (core pressure above 1010 hPa).
The relevance of this elimination differs between
geographical regions. Fig. 5a displays the distribution
of cyclone counts that fulfilled the required conditions
and Fig. 5b of those which were removed. Many of
the latter are located over lower latitude continental re-
gions, which are regions of generally slack SLP gradi-
ents where orography can dramatically affect the reduc-
tion of pressure to sea level. Thus, the systems identified
close to orographic borders which were not explicitly
832 J. G. Pinto et al.: Sensitivities of a cyclone detection and tracking algorithm Meteorol. Z., 14, 2005
a) b)
[10 m/s]
c) d)
e) f)
Figure 6: Cyclone track statistics of all tracked cyclones with data at T62 / 06 h resolution for the period from 1958/59 to 1997/98. Isolines
refer to winter half year (October–March) mean fields of (a): cyclone track density (interval: 6.0 cyclone days/winter, areas above 12.0 are
shaded grey). (b): core pressure (interval: 5.0 hPa). (c): cyclone velocity (interval: 2.0 m/s; arrows are included to indicate the direction of
the average propagation). A normed vector for 10 m/s is shown in the lower left corner. (d): core pressure tendency (thin lines, interval: 2.5
hPa/day) and upper-tropospheric baroclinicity field calculated as the EADY growth rate at 400 hPa (bold lines, interval 0.1 1/day. For (b),
(c) and (d), isolines are enhanced by shading over areas with a cyclone track density of more than 6 cyclone days/winter. (e): cyclogenesis
(interval: 1.0 events/winter, areas above 3.0 are shaded grey) (f): cyclolysis (interval: 1.0 events/winter, areas above 3.0 are shaded grey).
Values are calculated for areas with a radius of 7.5 lat. around each grid point. Isolines and shadings in areas with orography above 1500 m
are suppressed.
removed in the previous identification step (other than
in SML99) are effectively eliminated here, except when
they are part of a longer cyclone track.
5Cyclone climatology
On the basis of the remaining tracks, it is possible to con-
struct a complete climatology. Cyclone activity is now
Meteorol. Z., 14, 2005 J. G. Pinto et al.: Sensitivities of a cyclone detection and tracking algorithm 833
NA
MM
NP
a) b)
[2 m/s]
c) d)
e) f)
Figure 7: Cyclone track statistics and differences of all tracked cyclones with data at T42 / 12 h resolution for the period 1958/59 to
1997/98. (a): winter half year mean fields of cyclone track density (interval: 6.0 cyclone day/winter). Included are boxes for the North
Atlantic (NA), North Pacific (NP) and Mediterranean (MM). Differences for data at T42 / 12 h versus T62 / 06 h for (b): cyclone track
density (interval: 2.5 cyclone days/winter). (c): for absolute amount of cyclone propagation velocity (interval: 0.5 m/s). Arrows indicate the
mean direction of changges in propagation velocity, vector length corresponds to the absolute field values. A normed vector for 2 m/s is
shown in the lower left corner. (d): core pressure tendency (interval: 0.4 hPa/day). (e): cyclogenesis (interval: 1.0 events/winter) (f): cyclone
counts for all tracked and manipulated systems (interval: 1.0 cyclone days/winter). For (b) to (f), areas with differences above 99 percentile
value are shaded (t-testbased on monthly basis). Valuesare calculated for areas with a radius of 7.5 lat. around each grid point in (a) to (e)
and for 10.0◦x 5.0◦lon.-lat. grid boxes in (f). Isolines and shadings in areas with orography above 1500 m are suppressed.
quantified in terms of track density, calculated over a
certain area for each gridpoint (here 7.5◦radius around
each grid point of the 2.5 lon. lat. grid). This variable is
primarily influenced by fast moving systems (as these
systems travel a long distance and therefore produce
long tracks), unlike cyclone counts, where slower sys-
834 J. G. Pinto et al.: Sensitivities of a cyclone detection and tracking algorithm Meteorol. Z., 14, 2005
Figure 8: Distribution of the number of cyclones (N) in terms of their core pressure for (a): the main ocean basins NA+NP (c): the
Mediterranean region. Dashed line for data with T62 / 06 h resolution, solid line for data with T42 / 12 h resolution. Cyclone counts are
calculated for 1 hPa segments. Distribution of cyclones in terms of their ∇2
pvalues for (b): the mean ocean basins (d): for the Mediterranean
region. Lines as a) and c). Cyclone counts are calculated for 0.1 hPa/(deg. lat.)2segments.
tems are locally over-represented (as they are counted
more often for the same grid box than faster moving
ones). The cyclone track density for the NH is shown
in Fig. 6a, displaying the two main storm tracks with
high track density (North Atlantic, hereafter abbrevi-
ated NA, and North Pacific, the latter hereafter referred
to NP). Note that track density maxima are also found
over the secondary storm track areas, e.g., the Mediter-
ranean Basin and Siberia. Cyclones in these regions are
generally shallower (Fig. 6b) and have shorter lifetimes
than their oceanic counterparts (lifetime data not shown;
for further reference see TRIGO et al., 1999). The track
density is increased using our settings in comparison to
SML99, especially over the secondary storm track ar-
eas. This can be partly attributed to both the smoothing
of the input data and the exclusion of cyclones near oro-
graphic barriers used in SML99. Mean cyclone veloci-
ties (Fig. 6c) show maximum values close to the max-
imum jet regions over the western NA and NP where
cyclones travel predominantly east-north-eastward over
the ocean basins. Secondary maxima are found in the lee
of the Ural Mountains (Siberian storm track) and over
the Mediterranean.
Fig. 6d shows large negative pressure tendencies (i.e.
deepening of lows) over the western parts of the ocean
basins, while the rising of core pressures is found over
northern Europe and near the northern part of North
America’s Pacific coastline. Maxima in baroclinicity
(defined as the EADY growth rate at 400 hPa, dark lines
in Fig. 6d) are upstream of the maxima of deepening
of cyclones, which is consistent with the notion that re-
gionally high values of baroclinicity are generally as-
sociated with enhanced cyclone growth (HOSKINS and
VALDES, 1990). The main areas of cyclogenesis are lo-
cated downstream of steep orography, e.g., over the lee
of the Rocky Mountains and over Japan, but also over
the North American east coast and the Gulf of Genoa
(Fig. 6e). The main cyclolysis areas are located on the
eastern coasts of the oceanic basins (Fig. 6f). The ob-
tained cyclogenesis and -lysis areas are in very good
agreement with results by other authors (e.g., SICK-
MÖLLER et al., 2000; HOSKINS and HODGES, 2002).
We find that the accurate detection of individual cy-
clone positions and tracks and the realistic representa-
tion of the climatological cyclone characteristics provide
a good assessment of cyclone activity of the NH. More-
over, the method has the ability of being adequate for
both the main and the secondary storm track regions.
6Impact of reduced spectral and time
resolution
In the following section, the eventual impact of spectral
(S) and temporal (T) resolution of the basic data on the
cyclone climatology is assessed. We look at the two fac-
tors separately, as well as their combined effect (C). It
should be noted that a complete separation of the two
effects is not possible, as a negative synergy effect (ST)
in terms of C=S+T+ST must be taken into account. The
cyclone track density for T42/12 hour resolution (Fig.
Meteorol. Z., 14, 2005 J. G. Pinto et al.: Sensitivities of a cyclone detection and tracking algorithm 835
a) b)
c) d)
Figure 9: Cyclone statistics considering time (left) and spectral (right) reduction separately. (a): Difference for cyclone track density
with reduced time resolution (12h vs 06h; interval: 2.5 cyclone day/winter). (b): difference for cyclone track density with reduced spectral
resolution (T42 vs T62; interval: 2.5 cyclone days/winter). (c) As (a) but for cyclone counts of tracked systems (interval: 1.0 cyclone
days/winter). (d): as (b) but for cyclone counts of tracked systems (interval: 1.0 cyclone days/winter). Values for (c) and (d) are calculated
for 10.0◦x 5.0◦lon.-lat. grid boxes. Areas with significant anomalies (at 99 % confidence level) are shaded (t-test based on monthly basis).
Isolines and shadings in areas with orography above 1500 m are suppressed.
7a) displays a similar distribution in comparison to the
full resolution (Fig. 6a), with reduced absolute values
over the oceanic storm track areas (relative change of
minus 20 to 30 %) and Mediterranean basin, the Caspian
Sea and Siberia (with relative changes reaching minus
50 %, see Fig. 7b). Deviations in the obtained spatial
distributions are also found for other cyclone charac-
teristics: Average propagation speeds are lowered (en-
hanced south-westerly component compared to north-
easterly component in the climatological mean) over the
southern parts of the main storm tracks (Fig. 7c), indi-
cating that fast moving systems are severely affected by
the change in resolution in these areas.
In terms of core pressure tendencies (Fig. 6d), the ef-
fect is mainly a reduction of deepening rates (Fig. 7d).
Only small changes in these variables are observed over
the regions where the large cyclones decay (Fig. 7c and
7d). Our interpretation of these results is that the main
effect of reduced resolution is a loss of not only weaker
(which could be expected from the resolution change)
but also faster growing and propagating systems. The
loss of mainly weaker systems implies deeper mean core
pressures for the reduced dataset, as can be observed
over the exit regions of the main storm tracks (especially
for the NP) even though the changes are not significant
(figure not shown).
The effect of simultaneously reduced spatial and
temporal resolution of the input data is now investigated
in terms of the frequencies of different core depth (from
MSLP) and intensity classes (measured as ∇2
p). These
are investigated as spatial sums of the respective counts
using grid points over the main ocean basins (NP plus
NA boxes in Fig. 7a) and for the Mediterranean Basin
(MM box shown in Fig. 7a). From the ∇2
pstatistics,
it becomes clear that combining the reduction in both
the temporal and spatial domain generally leads to a
836 J. G. Pinto et al.: Sensitivities of a cyclone detection and tracking algorithm Meteorol. Z., 14, 2005
loss of many of the weaker systems while only mini-
mal changes are detected for stronger systems (Fig. 8b,
d). This effect is observed over all ocean basins consid-
ered. On the other hand, the relative changes for depth
statistics are similar for all depth classes (Fig. 8a,c). This
exhibits a clear contrast between results based on core
pressure and on vorticity.
A coarser temporal resolution (12 hourly instead of
6 hourly) without a change in spatial resolution leads
to a significantly reduced cyclone track density over the
whole storm track domain (Fig. 9a). The reduction in
track density amounts roughly half of the combined ef-
fect. An analysis of individual tracks shows that the loss
of tracks is frequently related to very large displace-
ments (fast moving systems) or to changes in the propa-
gation speed of the cyclones. With respect to the result-
ing climatologies, this implies that the mean eastward
propagation velocities and the deepening rates are re-
duced, in particular near the entrance and central regions
of the storm tracks over the oceans (not shown).
Reduction of the spatial resolution alone mostly
affects shallow systems. Cyclone track densities are
mostly reduced between Greenland and Scandinavia,
and between southern Alaska and northwest Canada
(Fig. 9b), while only a moderate loss of tracks occurs
over the main storm tracks. The reason for the small
sensitivity to reduced spatial resolution over this latter
region is simply that there are only few weak systems
that could be lost. There is also little effect on cyclone
velocities and cyclone deepening rates over these ar-
eas (not shown). On the other hand, sensitivity is high
over continental areas, as there are many shallower sys-
tems embedded in slack SLP gradients. The loss of these
weak and non-developing systems at lowered resolution
results in enhanced core pressure tendencies over these
regions (not shown). The growing and deeper cyclones
are still identified and tracked at reduced spatial resolu-
tion of the data. This also becomes evident from the fact
that the mean core pressure is lowered.
Considering the effects of reduced resolution in
terms of cyclone counts, the signal is relatively small
for the temporal effect (Fig. 9c), but large for the spa-
tial effect (Fig. 9d). This result is in clear contrast to the
signal in cyclone track densities. The different impacts
are mainly due to the alternate nature of both variables:
While slower systems are over-represented in compari-
son to fast moving systems for the cyclone counts, the
track density is primarily influenced by fast movingsys-
tems.
The tracking of systems based on daily temporal
resolution data was also tested. However, the obtained
results indicate an unsatisfactory performance of the
tracking scheme (using the present settings) and show a
high percentage of incorrect and broken tracks in com-
parison to the hand analysed and 6 hourly-tracks (not
shown). Of course, the large steering cyclones with lin-
ear tracks are less affected than smaller scale systems.
This assessment is in agreement with results from other
schemes (BLENDER and SCHUBERT, 2000; ZOLINA
and GULEV, 2002). As a consequence, the resulting
number of cyclones is considerably reduced, in partic-
ular over the secondary storm track regions (not shown).
As suggested by MS91, improved results at low tem-
poral resolution of the basic input data (e.g., daily data)
could be obtained by using climatological steering ve-
locities instead of geostrophic velocities. This could pro-
vide better results for the climatology, even though indi-
vidual tracks are frequently not caught correctly. Data
smoothing (which we rejected in the present study)
could also contribute positively to a tracking that is re-
stricted to deep cyclones.
7Discussion
The aim of the present study was to obtain a cyclone cli-
matology which successfully includes the whole range
of NH cyclones: the large quasi stationary and transient
oceanic cyclones, fast moving storms and smaller sys-
tems over secondary storm track areas. Furthermore, in-
sight was provided into the relevance of algorithm pa-
rameters and impact of the choice of their values for the
identification and tracking of cyclones. Using a set of
values suitable for the NH, the algorithm proved to be a
very powerful tool not only in the generation of cyclone
climatologies but also in the assessment of individual
tracks. It was shown that cyclones may change between
a state with or without closed isobars during their life-
time (closed and open systems, respectively) so that the
inclusion of the open cyclones is a necessary ingredi-
ent in the scheme. The method is capable of identifying
cyclones in a range of locations and with different char-
acteristics, including small scale systems over secondary
storm track regions and fast moving storms that produce
extreme events like winter storms over Europe later in
their lifetime. The characteristics of the individual cy-
clones were assembled at each time step, e.g., their in-
tensity (∇2
p), core pressure, intensity tendency and prop-
agation speed. These results provide a basis for assess-
ing both particular life cycles of single cyclones and cli-
matological aspects based on large cyclone ensembles,
both from reanalysis and model data.
We have chosen to work with MSLP data, even
though several authors have reported some problems for
cyclone identification, often recommending the use of
lower level relative vorticity instead (e.g., HOSKINS and
HODGES, 2002; LIONELLO et al., 2002). Arguments
against our technique include the difficult identification
of smaller scale systems and the dependence on below-
ground extrapolation techniques with MSLP. Moreover,
the pressure minimum is considered a bad indicator for
Meteorol. Z., 14, 2005 J. G. Pinto et al.: Sensitivities of a cyclone detection and tracking algorithm 837
the evaluation of the strength of the circulation associ-
ated with a cyclone. On the other hand, our study gives
evidence that a cyclone identification and tracking may
be based on MSLP fields alone by using the ∇2
pfield
(which is proportional to relative vorticity) in the iden-
tification process. ∇2
palso provides a measure for the
strength of the systems largely independent from the
background flow. This approach proved to be extremely
helpful for the correct identification of the systems, es-
pecially at the initial and final stages of a cyclone’s life
cycle. The comparatively small data requirements make
the algorithm suitable for an investigation of long time
series as they are available in ensembles of GCM simu-
lations.
With respect to the main storm tracks over the
oceans, our results are in good agreement with other
studies (e.g., HODGES, 1996; SINCLAIR, 1997; SER-
REZE et al., 1997; SICKMÖLLER et al., 2000; GULEV
et al, 2001; HOSKINS and HODGES, 2002; LAMBERT
et al., 2002), even though a direct comparison of the
obtained cyclone climatologies is often hampered by
several differences between quantities considered and
datasets used. LEONARD et al. (1999) point out that
the algorithm used in the present paper is able to de-
tect a larger number of cyclones and tracks in compar-
ison to other methodologies. This is especially impor-
tant for secondary storm track regions where many other
schemes identify comparatively few systems (e.g., SER-
REZE et al., 1997; GULEV et al., 2001). Similar results
in these areas were found by HOSKINS and HODGES
(2002) when using the 850 hPa relative vorticity fields.
Our results are also in good agreement with work
performed specifically for the Mediterranean region
(TRIGO et al. 1999), e.g., in terms of cyclogenesis and
preferred cyclone tracks. Moreover, our results for indi-
vidual cyclones reveal a good agreement with hand anal-
ysed synoptic weather maps.
We related the cyclone deepening rates with baro-
clinicity and confirmed that enhanced regional baro-
clinicity is associated with an intensified development
of the systems. This corroborates results presented by
HOSKINS and HODGES (2002) and by ULBRICH et al.
(2001) for individual cyclones producing extreme winds
over Europe. Additional work based on the analysis of
individual tracks confirmed that locally enhanced baro-
clinicity often leads toa higher weather relevance of sys-
tems in the Mediterranean Basin.
With respect to the spatial and temporal resolution
of the input data, our results confirmed the findings of
BLENDER and SCHUBERT (2000) that the number of
cyclones detected is reduced when spatial and tempo-
ral resolution are lowered. In terms of core pressure, it
appears that bothweak and intense systems are affected,
but a consideration of ∇2
preveals that it is in fact a loss of
systems with small vorticity anomalies. Cyclone counts
are predominantly affected by reduced spatial resolu-
tion, while track density changes are equally affected
by both effects. On the other hand, a reduced tempo-
ral resolution (from 6 to 12 hours) has hardly any im-
pact on cyclone counts. This is due to the different na-
ture of these variables: Track density has an emphasis
on faster moving and rapidly developing systems while
slowly moving mature systems are comparatively less
important due to short displacements between individ-
ual dates. The loss of fast systems at lowered temporal
resolution is also confirmed by a reduction of mean cy-
clone velocities.
The cyclone climatology obtained with T42/12h res-
olution reproduces the main features of NH cyclone ac-
tivity observed at full resolution. This conclusion sug-
gests the use of this algorithm among others to validate
model data at this resolution against observational data.
An application to the ECHAM4/OPYC3 model is pre-
sented in the companion paper (PINTO et al, 2005).
Acknowledgements
The cyclone identification and tracking code was kindly
made available by Ross MURRAY and Ian SIMMONDS
(University of Melbourne, Australia), to whom we
would like to thank for the possibility to use and modify
it. We would also like to thank Prof. Dayton VINCENT
and the anonymous reviewers for many helpful sugges-
tions, and Mark REYERS for help in the preparation of
some of the figures.
J.G. PINTO was supported by the Portuguese Office
for Science and Technology (Program PRAXIS XXI,
Grant BD/15775/98). T. SPANGEHL was supported by
the German Research Foundation (DFG, Projects SP
191/25-1 and SP 191/25-2).
References
ALPERT, P., U. NEEMAN, Y. SHAL-EL, 1990: Climatologi-
cal Analysis of the Mediterranean Cyclones using ECMWF
data. – Tellus 42A, 65–77.
BLACKMON, M.L., 1976: A climatological spectral study
of the 500 mb geopotential height of the Northern Hemi-
sphere. – J. Atmos. Sci. 33, 1607–1623.
BLENDER, R., M. SCHUBERT, 2000: Cyclone Tracking in
Different Spatial and Temporal Resolutions. – Mon. Wea.
Rev. 128, 377–384.
BLENDER, R., K. FRAEDRICH, F. LUNKEIT, 1997: Identi-
fication of cyclone-track regimes in the North Atlantic. –
Quart. J. Roy. Meteor. Soc. 123, 727–741.
GENG, Q., M. SUGI, 2001: Variability of the North Atlantic
Cyclone Activity in Winter Analysed from NCEP-NCAR
Reanalysis Data. – J. Climate 14, 3863–3873.
GULEV,S.K., O. ZOLINA, S. GRIGORIEV, 2001:Extratropi-
cal Cyclone Variability in the Northern Hemisphere Winter
from the NCEP/NCAR Reanalysis Data. – Climate Dyn.
17, 795–809.
838 J. G. Pinto et al.: Sensitivities of a cyclone detection and tracking algorithm Meteorol. Z., 14, 2005
HAAK, U., U. ULBRICH, 1996: Verification of an objective
cyclone climatology for the North Atlantic. – Meteorol. Z.
N.F., 5, 24–30.
HODGES, K.I. , 1994: A General Method for Tracking Analy-
sis and its Application to Meteorological Data. – Mon. Wea.
Rev. 122, 2573–2586.
—, 1996: Spherical nonparametric estimators applied to the
UGAMP model integration for AMIP. – Mon. Wea. Rev.
124, 2914–2932.
HOSKINS, B.J., K.I. HODGES, 2002: New Perspectives on
the Northern Hemisphere Winter Storm Tracks. – J. Atmos.
Sci. 59, 1041–1061.
HOSKINS, B.J., P.J. VALDES, 1990: On the existence of
storm tracks. – J. Atmos. Sci. 47, 1854–1864.
KALNAY, E., M. KANAMITSU, R. KISTLER, W. COLLINS,
D. DEAVEN, L. GANDIN, M. IREDELL, S. SAHA, G.
WHITE, J. WOOLLEN, Y. ZHU, M. CHELLIAH, W.
EBISUZAKI, W. HIGGINS, J. JANOWIAK, K.C. MO, C.
ROPELEWSKI, J. WANG, A. LEETMA, R. REYNOLDS, R.
JENNE, D. JOSEPH, 1996: The NCEP/NCAR 40-Year Re-
analysis Project. – Bull. Amer. Meteor. Soc. 77, 437–471.
KISTLER, R., E. KALNAY, W. COLLINS, S. SAHA, J.
WOOLLEN, M. CHELLIAH, W. EBI-SUZAKI, M. KANA-
MITSU, V. KOUSKY, H. VAN DEN DOOL, R. JENNE,
R. FIORINO, 2001: The NCEP/NCAR 50-Year Reanaly-
sis: Monthly-Means CD-ROM and Documentation. – Bull.
Amer. Meteor. Soc. 82, 247–267.
KLAWA, M., U. ULBRICH, 2003: A model for the estima-
tion of storm losses and the identification of severe winter
storms in Germany. – Natural Hazards and Earth System
Sciences 3, 725–732.
KÖNIG, W., R. SAUSEN, F. SIELMANN, 1993: Objective
Identification of cyclones in GCM Simulations. – J. Climate
6, 2217–2231.
LAMBERT, S.J., 1988: A Cyclone Climatology of the Cana-
dian Climate Centre General Circulation Model. – J. Cli-
mate 1, 109–115.
LAMBERT, S.J., J. SHENG, J. BOYLE, 2002: Winter cyclone
frequencies in thirteen models participating in the Atmo-
spheric Model Intercomparison Project (AMIP1). – Climate
Dyn. 19, 1–16.
LETREUT, H., E. KALNAY, 1990: Comparison of observed
and simulated cyclone frequency distribution as determined
by an objective method. – Atmosphera 3, 57–71.
LEONARD, S.R., J. TURNER, A. VAN DER WAL, 1999: An
assessment of three automatic depression tracking schemes.
– Meteor. Appl. 6, 173–183.
LIONELLO, P., F. DALAN, E. ELVINI, 2002: Cyclone in the
Mediterranean Region: the present and the doubled CO2
climate scenarios. – Climate Res. 22, 147–159.
MURRAY, R.J., I. SIMMONDS, 1991: A numerical scheme
for tracking cyclone centres from digital data. Part I: devel-
opment and operation of the scheme. – Aust. Met. Mag. 39,
155–166.
PEIXOTO, J.P., A.H. OORT, 1992: Physics of Climate. –
American Institute of Physics, New York, 520 pp.
PINTO, J.G, T. SPANGEHL, U. ULBRICH, P. SPETH,
2005: Assessment of winter cyclone activity in a transient
ECHAM4-OPYC3 GHG experiment. – Submitted to Mete-
orol. Z.
REID, P.A., P.D. JONES, O. BROWN, C.M. GOODESS, T.D.
DAVIES, 2001: Assessments of the reliability of NCEP cir-
culation data and relationships with surface climate by di-
rect comparisons with station based data. – Climate Res.
17, 247–261.
SCHINKE, H., 1992: Zum Auftreten von Zyklonen mit Kern-
drücken <990 hPa im atlantisch-europäischen Raum von
1930 bis 1991. – Spezialarbeiten aus der Arbeitsgruppe Kli-
maforschung des Meteorologischen Instituts der Humboldt-
Universität zu Berlin, 1.
—, 1993: On the occurrence of deep cyclones over Europe
and the North Atlantic in period 1930–1991. – Beitr. Phys.
Atmos. 66, 223–237.
SERREZE, M.C., 1995: Climatological aspects of cyclone
development and decay in the Artic. – Atmosphere-Ocean
33(1), 1–23.
SERREZE, M.C., F. CARSE, R.G. BARRY, J.C. ROGERS,
1997: Icelandic Low Cyclonic Activity: Climatological
Features, Linkages with the NAO, and Relationship with
Recent Changes in Northern Hemisphere Circulation. – J.
Climate 10, 453–464.
SICKMÖLLER, M., R. BLENDER, K. FRAEDRICH, 2000:
Observed winter cyclone tracks in northern hemisphere in
re-analysed ECMWF data. – Quart. J. Roy. Meteor. Soc.
126, 591–620.
SIMMONDS, I., R.J. MURRAY, R.M. LEIGHTON, 1999: A
refinement of cyclone tracking methods with data from
FROST. – Aust. Met. Mag., Spec. Ed., 35–49.
SINCLAIR, M.R.,1997: Objective Identification of Cyclones
and Their Circulation Intensity and Climatology. – Weather
and Forecasting 12, 595–612.
TRIGO, I.F., T.D. DAVIES, G.R. BIGG, 1999: Objective Cli-
matology of Cyclones in the Mediterranean Region. – J.
Climate 12, 1685–1696.
—, —, —, 2000: Decline in Mediterranean Rainfall Caused
By Weakening of Mediterranean Cyclones. – Geophys.
Res. Lett. 27 (18), 2913–2916.
UENO, K., 1993: Inter-annual Variability of Surface Cyclone
Tracks, Atmospheric Circulation Patterns and Precipitation
in Winter. – J. Climatol. 4, 297–310
ULBRICH, U., A.H. FINK, M. KLAWA, J.G. PINTO, 2001:
Three extreme storms over Europe in December 1999. –
Weather 56, 70–80.
ZOLINA, O., S.K. GULEV, 2002: Improving the Accuracy of
Mapping Cyclone Numbers and Frequencies. – Mon. Wea.
Rev. 130, 748–759.