ArticlePDF Available

Twentieth-century atmospheric river activity along the west coasts of Europe and North America: algorithm formulation, reanalysis uncertainty and links to atmospheric circulation patterns

Authors:

Abstract and Figures

A new atmospheric-river detection and tracking scheme based on the magnitude and direction of integrated water vapour transport is presented and applied separately over 13 regions located along the west coasts of Europe (including North Africa) and North America. Four distinct reanalyses are considered, two of which cover the entire 20th-century: NOAA-CIRES Twentieth Century Reanalysis v2 (NOAA-20C) and ECMWF ERA-20C. Calculations are done separately for the OND and JFM-season and, for comparison with previous studies, for the ONDJFM-season as a whole. Comparing the AR-counts from NOAA-20C and ERA-20C with a running 31-year window looping through 1900-2010 reveals differences in the climatological mean and inter-annual variability which, at the start of the 20th-century, are much more pronounced in western North America than in Europe. Correlating European AR-counts with the North Atlantic Oscillation (NAO) reveals a pattern reminiscent of the well-know precipitation dipole which is stable throughout the entire century. A similar analysis linking western North American AR-counts to the North Pacific index (NPI) is hampered by the aforementioned poor reanalysis agreement at the start of the century. During the second half of the 20th-century, the strength of the NPI-link considerably varies with time in British Columbia and the Gulf of Alaska. Considering the period 1950-2010, AR-counts are then associated with other relevant large-scale circulation indices such as the East Atlantic, Scandinavian, Pacific-North American and West Pacific patterns (EA, SCAND, PNA and WP). Along the Atlantic coastline of the Iberian Peninsula and France, the EA-link is stronger than the NAO-link if the OND season is considered and the SCAND-link found in northern Europe is significant during both seasons. Along the west coast of North America, teleconnections are generally stronger during JFM in which case the NPI-link is significant in any of the five considered subregions, the PNA-link is significant in British Columbia and the Gulf of Alaska and the WP-link is so along the U.S. West Coast. During OND, these links are significant in the Gulf of Alaska only. If AR-counts are calculated upon persistent (instead of instantaneous) ARs, the link to the NAO weakens over the British Isles and western Iberia. For the experimental set-ups most closely mirroring those applied in Lavers et al (2012)and Ramos et al (2015), the NAO-links are completely or partly insignificant indicating that the inclusion of the persistence criterion notably alters the results. Visual support for the present study is provided by an exhaustive historical atmospheric river archive built at http://www.meteo.unican.es/atmospheric-rivers.
Content may be subject to copyright.
Climate Dynamics manuscript No.
(will be inserted by the editor)
20th-Century Atmospheric River Activity along1
the West Coasts of Europe and North America:2
Algorithm Formulation, Reanalysis Uncertainty and3
Links to Atmospheric Circulation Patterns4
S. Brands ·J.M. Guti´errez ·D.5
San-Mart´ın6
7
Received: date / Accepted: date8
Abstract A new atmospheric-river detection and tracking scheme based on the9
magnitude and direction of integrated water vapour transport is presented and ap-10
plied separately over 13 regions located along the west coasts of Europe (including11
North Africa) and North America. Four distinct reanalyses are considered, two of12
which cover the entire 20th-century: NOAA-CIRES Twentieth Century Reanal-13
ysis v2 (NOAA-20C) and ECMWF ERA-20C. Calculations are done separately14
for the OND and JFM-season and, for comparison with previous studies, for the15
ONDJFM-season as a whole.16
Comparing the AR-counts from NOAA-20C and ERA-20C with a running 31-17
year window looping through 1900-2010 reveals differences in the climatological18
mean and inter-annual variability which, at the start of the 20th-century, are19
much more pronounced in western North America than in Europe. Correlating20
European AR-counts with the North Atlantic Oscillation (NAO) reveals a pattern21
reminiscent of the well-know precipitation dipole which is stable throughout the22
entire century. A similar analysis linking western North American AR-counts to23
the North Pacific index (NPI) is hampered by the aforementioned poor reanalysis24
agreement at the start of the century. During the second half of the 20th-century,25
the strength of the NPI-link considerably varies with time in British Columbia26
and the Gulf of Alaska.27
Considering the period 1950-2010, AR-counts are then associated with other28
relevant large-scale circulation indices such as the East Atlantic, Scandinavian,29
Pacific-North American and West Pacific patterns (EA, SCAND, PNA and WP).30
Along the Atlantic coastline of the Iberian Peninsula and France, the EA-link is31
stronger than the NAO-link if the OND season is considered and the SCAND-link32
found in northern Europe is significant during both seasons. Along the west coast33
of North America, teleconnections are generally stronger during JFM in which case34
the NPI-link is significant in any of the five considered subregions, the PNA-link35
S. Brands, 1. MeteoGalicia - Conseller´ıa de Medio Ambiente, Territorio e Infraestructuras -
Xunta de Galicia, Santiago de Compostela, Spain 2. Instituto de F´ısica de Cantabria (CSIC-
UC), Santander, Spain, E-mail: swen.brands@gmail.com
J. M. Guti´errez, Instituto de F´ısica de Cantabria (CSIC-UC), Santander, Spain
D. San-Mart´ın, Predictia Intelligent Data Solutions, Santander, Spain
2 S. Brands et al.
is significant in British Columbia and the Gulf of Alaska and the WP-link is so36
along the U.S. West Coast. During OND, these links are significant in the Gulf of37
Alaska only.38
If AR-counts are calculated upon persistent- instead of instantaneous ARs, the39
link to the NAO weakens over the British Isles and western Iberia. For the exper-40
imental set-ups most closely mirroring those applied in Lavers et al (2012) and41
Ramos et al (2015), the NAO-links are completely or partly insignificant indicat-42
ing that the inclusion of the persistence criterion notably alters the results. Visual43
support for the present study is provided by an exhaustive historical atmospheric44
river archive built at http://www.meteo.unican.es/atmospheric-rivers.45
Keywords Atmospheric Rivers ·Reanalysis Data ·20th century ·Atmospheric46
Circulation ·Europe ·North America47
1 Introduction48
The poleward transport of water vapour in the atmosphere is not organized homo-49
geneously in space and time. Rather, it is concentrated in narrow and elongated50
spatial structures of intense transport having a live-time of a few days at the51
utmost (Zhu and Newell, 1994, 1998). Due to their filamentary appearance remi-52
niscent of a river’s course seen from bird’s-eye perspective, these structures have53
been originally referred to as “tropospheric rivers” (Newell et al, 1992), a term54
which later on developed to “atmospheric rivers” (ARs). Two processes contribute55
to the formation and maintenance of the water vapour constituting these struc-56
tures. The first one is evapotranspiration in a remote source region followed by57
Lagrangian transport over thousands of kilometres, similar to the flow of a river,58
in which case evapotranspiration and condensation along the transport “route”59
play a minor role (Knippertz and Wernli, 2010; Gimeno et al, 2012; Sodemann60
and Stohl, 2013; Garaboa et al, 2015). The second process is small-scale mois-61
ture recycling (evapotranspiration, condensation and precipitation). In this case,62
water vapour is continuously lost and refreshed ahead of the cold front(s) of one63
or several extra-tropical cyclones, leading to a structure looking like a river but64
not sharing its transport properties (Bao et al, 2006). Recent studies point to the65
fact that, for most ARs, moisture recycling is more important than long-range66
transport (Newman et al, 2012; Dacre et al, 2015).1
67
ARs can be identified and tracked using either Eulerian or Langrangian meth-68
ods (Newell et al, 1992; Bao et al, 2006; Knippertz and Wernli, 2010; Gimeno et al,69
2012; Garaboa et al, 2015). The algorithms used within the Eulerian framework,70
which will be the focus of the present study, are capable to automatically detect71
and track AR-structures at a given point in time and usually operate on vertically72
integrated water vapour transport (Zhu and Newell, 1998; Lavers et al, 2012; Guan73
and Waliser, 2016). The corresponding data are ideally taken from dropsonde- or74
satellite observations which, however, have a limited spatial and temporal cover-75
age (Zhu and Newell, 1998; Ralph et al, 2004; Lavers et al, 2011). This is why76
1Author’s comment: Since the two aforementioned studies rely on reanalysis data, the cor-
responding results might be sensitive to the physics and parametrization schemes of the global
circulation model used for re-analysing. Note that the relative contribution of the two afore-
mentioned factors might change if other models and/or parametrization schemes are used.
20th-Century Atmospheric Rivers 3
model-data from reanalyses, usually referred to as “quasi-observations” (Brands77
et al, 2012), are used if long time series and complete spatial coverage is required,78
e. g. for assessing the climatological aspects of atmospheric rivers (Higgins et al,79
2000; Neiman et al, 2008; Knippertz et al, 2013; Dacre et al, 2015).80
Recently, two long-term reanalyses covering the entire 20th century have be-81
come available: the NOAA-CIRES 20th century reanalysis [hereafter: NOAA-20C,82
Compo et al (2011), version 2 is used here], and the ECMWF ERA-20C reanal-83
ysis [ERA-20C, Poli et al (2013)]. In comparison to alternative reanalyses relying84
on surface-, upper-air- and satellite observations (Kalnay et al, 1996; Dee et al,85
2011), only surface observations were considered in the data-assimilation proce-86
dure opted for in these projects. This was done to reduce the risk of artificial shifts87
(or inhomogeneities) in the time-series simulated by the Global Circulation Models88
used for re-analysing. These “observational shocks” (Ferguson and Villarini, 2012)89
are caused by sudden increases in the number of assimilated observations; the in-90
troduction of satellite data in the late 1970s being the most prominent example91
(Sturaro, 2003; Sterl, 2004).92
On the regional scale, seasonal precipitation sums and extreme precipitation93
events have been associated with ARs and strong relationships were found for those94
regions characterized by specific topographic features such as mountain ranges95
near the coast or a coastline perpendicular to the main direction of the horizontal96
moisture flow (Neiman et al, 2004; Guan et al, 2012; Kim et al, 2013; Ramos et al,97
2015; Eiras et al, 2016). Therefore, ARs are on the one hand beneficial for a region’s98
water supply but on the other are potentially harmful since they can trigger heavy99
flooding and landslide events [hereafter jointly referred to as “hydrological extreme100
events”, Lavers et al (2011)], especially in case they coincide with a previously101
accumulated thick snow pack and/or water-saturated soils (Leung and Qian, 2009;102
Ralph et al, 2013).103
Whereas ARs triggering (extreme) precipitation have been studied extensively104
to-date, a spatially (and also temporally) complete picture on the large-scale at-105
mospheric conditions triggering ARs is yet “under construction” [see Gimeno et al106
(2014) and references therein]. Bao et al (2006) found that enhanced IVT over107
the United States (U.S.) West Coast originating from the tropics is favoured by108
a weakened subtropical ridge in the central Pacific. Kim and Alexander (2015)109
found the strength and position of the Aleutian low to be key for the spatial pat-110
tern of IVT anomalies. If this low pressure system is anomalously deep, IVT is111
above normal in the northwestern U.S. and if it is displaced to the south moist112
conditions are exhibited by the southwestern U.S. and Mexico. Guan et al (2012)113
found that the exceptional AR-activity over California’s Sierra Nevada during the114
2010/2011 snow season was linked to the negative phase of both the Pacific-North115
American pattern (PNA) and the Arctic Oscillation (Barnston and Livezey, 1987).116
The particular role of the PNA in “driving” AR-frequency counts in the region117
extending from the Canada-United States boarder to Alaska has been recently118
pointed out by Guan and Waliser (2016). Considering the 1979-2010 period, Jiang119
and Deng (2011) demonstrated that East-Asian cold surges increase the odds for120
ARs land-falling along the west coast of North America during the days following121
the peak of the cold-surge.122
Since the above mentioned modes of atmospheric variability are known to be123
influenced by low-frequency modes originating in the tropics —which, in principle,124
are predictable on intra-seasonal to seasonal time-scales— attempts have been125
4 S. Brands et al.
made to indirectly associate these tropical modes with IVT/AR-count anomalies126
along the west coast of North America (Bao et al, 2006; Guan et al, 2012; Payne127
and Magnusdottir, 2014; Kim and Alexander, 2015; Guan and Waliser, 2016).128
Similar studies for Europe are sparse and partly contradictory. Lavers et al129
(2012) found the number of extended-winter season (October-to-March) AR land-130
falls over the British Isles to be inversely related to the Scandinavian Pattern131
(Barnston and Livezey, 1987). In a follow-up study conducted on continental scale132
(Lavers and Villarini, 2013), the sea-level pressure composite maps associated with133
AR-arrivals in northern and southern Europe were found to resemble the positive134
and negative phase of the North Atlantic Oscillation [NAO, (Hurrell et al, 2003)]135
respectively. Ramos et al (2015) focused on October-through-March AR-arrivals136
over the Iberian Peninsula and found them to be positively related to the Scandi-137
navian pattern in first place. Unlike in Lavers and Villarini (2013), the “AR-NAO”138
link was found to be insignificant in Ramos et al (2015) which is perhaps some-139
what counter-intuitive given that the NAO is known to describe a large fraction140
of variability of the wintertime precipitation totals in this region (Hurrell, 1995;141
Trigo et al, 2004). Reasons for this disagreement might be found in differences in142
the considered datasets, time-periods and season-definitions.143
This study assesses the atmospheric river phenomenon along the west coasts144
of Europe and North America from a climatological point of view. Considering145
the October-through-December and January-through-March seasons (OND and146
JFM), a new AR-detection and tracking algorithm is proposed and applied to 6-147
hourly instantaneous data from four distinct reanalyses, two of which date back148
to the year 1900 (NOAA-20C and ERA-20C). This is done separately for 13 re-149
gions located along the west coasts of Europe and North America (see Figure 1).150
After pointing out the advantages of the new algorithm, the similarity between151
the year-to-year AR-count series from the two long-term reanalyses is assessed152
backwards in time to the early 20th-century. Similarity is measured 1) in terms of153
the climatological mean (represented by the bias) and 2) in terms of inter-annual154
variability (represented by the rank correlation coefficient, rs). Both measures are155
calculated for a sliding 31-year window moving forward by one year in a loop rang-156
ing from 1900 to 2010. In the absence of any “true” dataset dating back to 1900,157
and following the ideas of Sterl (2004), a comparison of two distinct reanalyses158
provides an estimate of their degree of realism. If similar results are obtained from159
the two for a given time-series aspect (i.e. inter-annual variability as documented160
by rs), this is likely due to a strong observational constrain, indicating that the161
result is a realistic estimation of the “truth”. A large discrepancy, in contrast,162
indicates a loose observational constrain and an unreliable result. Encouraged by163
the finding that the inter-annual variability of the AR-counts for Europe is similar164
in the two reanalyses even at the start of the 20th century, the time-dependence165
of their association with the NAO is traced backwards until 1900 by means of a166
running correlation analysis. Results are then contrasted with a similar analysis167
relating AR-counts in western North America with the strength of the Aleutian168
low as described by the North Pacific index (Trenberth and Hurrell, 1994).169
In a second working step, the search for atmospheric drivers of regional AR-170
activity is extended to other relevant indices describing the East Atlantic, Scandi-171
navian, Pacific-North American and West Pacific patterns (Barnston and Livezey,172
1987). In this case, we focus on the more reliable period 1950-2010. For ease of173
comparison with previous studies (Lavers et al, 2012; Ramos et al, 2015), seasonal174
20th-Century Atmospheric Rivers 5
AR-counts are additionally derived from NCEP/NCAR reanalysis 1 (Kalnay et al,175
1996) and ECMWF ERA-Interim (Dee et al, 2011) and the entire extended winter176
season (October-through-March), as well as the persistence criterion described in177
Lavers et al (2012) are considered in addition. It will be shown that the application178
of the persistence criterion can weaken the strength of the statistical relationships179
to the degree that the significant link to the NAO is lost for the experimental180
set-ups that most closely match those applied in Lavers et al (2012) and Ramos181
et al (2015).182
Finally, an exhaustive historical archive of AR-events in the above mentioned183
13 regions was built and made publicly available at http://www.meteo.unican.184
es/atmospheric-rivers. This archive will hereafter be referred to as the “At-185
mospheric River Archive”. It documents the behaviour of the proposed detection186
and tracking algorithm for thousands of cases and permits to openly discuss its187
advantages and disadvantages.188
The remainder of this article is outlined as follows. The applied datasets are de-189
scribed in Section 2. The AR-detection and tracking scheme as well as the applied190
reanalysis similarity measures are described in Section 3. Results are presented in191
Section 4 and a discussion and some concluding remarks are provided in Section192
5.193
2 Data194
For the purpose of AR detection and tracking, 6-hourly instantaneous data from195
the four reanalyses specified in Table 1 are used (the respective URLs are provided196
in the Acknowledgements).197
The algorithm operates on the magnitude (I V T , in kg m1s1) and direc-198
tion (Din degrees) of the vertically integrated water vapour transport which are199
calculated as follows:200
IV T =pIV Tu2+I V Tv2(1)
D=atan2IV Tu
IV T ,I V Tv
IV T 180
π+ 180 (2)
where IV Tuand I V Tvare the vertical integrals of the zonal and meridional201
water vapour transport components respectively. The atan2 function returns the202
four-quadrant inverse tangent ranging in between πand πwhich is then trans-203
formed to degree values ranging in between 0and 360.204
IV Tuand I V Tvwere calculated from 2-dimensional pressure-level data be-205
tween 1000 and 300 hPa (Lavers et al, 2012).206
IV Tu=1
gZ300
1000
qu dp(3)
and207
IV Tv=1
gZ300
1000
qv dp(4)
6 S. Brands et al.
where q,uand vrefer to specific humidity (in kg kg1), zonal and meridional208
wind (in m s1) at pressure level p,gto acceleration due to gravity and dpto the209
difference between adjacent pressure levels (in Pa).210
For NCEP/NCAR and NOAA-20C, 7 and 15 vertical pressure levels between211
1000 and 300 hPa were available from the data providers respectively. Vertical212
integration is achieved by multiplying qu and qv at the pressure level pby a213
multiplier describing its contribution (as represented by the number of pressure214
levels in P a) to the entire column extending from 1000 to 300 hP a (see Table215
2), followed by summing up the resulting products. Since ECMWF’s public server216
already provides IV Tuand I V Tvas vertical integrals between the pressure level at217
model surface and the top of the atmosphere (ECMWF, personal communication),218
it was not necessary to apply Equations 3 and 4 for ERA-20C and ERA-Interim.219
Note that q,uand vfrom NOAA-20C are ensemble-mean data.220
In addition to the reanalysis datasets, monthly values of the large-scale atmo-221
spheric circulation indices relevant for the North Atlantic and North Pacific sectors222
were retrieved from the Climate Prediction Center (CPC) (Barnston and Livezey,223
1987) and the University Cooperation for Atmospheric Research (UCAR) (Hurrell224
et al, 2003). A detailed description of these indices can be found in Table 3.225
The considered time periods are as follows. A 31-year moving window run-226
ning from 1900-2010 is used to assess time-variations in 1) the similarity between227
AR-counts from NOAA-20C and ERA-20C and 2) the strength of their link to228
the NAO or NP indices. The “full” association including indices others than the229
aforementioned two is conducted for the 1950 - 2010 period except for AR-counts230
from ERA-Interim, in which case 1979 - 2013 is used. Finally, 1979/80 - 2009/10231
and 1950/51 - 2011/12 are considered for comparison with Lavers et al (2012) and232
Ramos et al (2015) respectively.233
3 Methods234
3.1 Atmospheric-River Detection and Tracking Algorithm235
In the present study, ARs are detected separately in 8 regions ranging from Mo-236
rocco to northern Norway and 5 regions ranging from southern California to the237
northern Gulf of Alaska respectively (see Fig. 1). Each detection region is defined238
as a “barrier” of grid-boxes approximately following the coastline. Due to distinct239
native horizontal resolutions, the exact coordinates of these barriers slightly differ240
from one dataset to another (the barriers shown in Fig. 1 refer to the ERA-20C241
dataset). Using the native resolution is preferable to interpolating to a common242
coarse grid, which would lead to a degradation of the higher-resolution datasets.243
For a given detection region formed by a barrier of bgrid-boxes, the following244
detection and tracking algorithm was applied every six hours (see also Figure 2).245
1. The grid-box of maximum IVT along bis retained. This grid-box is hereafter246
referred to as the “targeted grid-box” e.247
2. If the IVT value at eexceeds the predefined percentile threshold Pd(the detec-248
tion percentile) the AR-tracking algorithm is activated, otherwise it proceeds249
to the next point in time.250
20th-Century Atmospheric Rivers 7
3. Then, the direction (D) of the IVT-flow at eis calculated (see Equation 2) and251
discretized into the 8 cardinal directions: N, NE, E, SE, S, SW, W, NW. In252
the following example, we assume that Dis from the W.253
4. Out of the 8 possible neighbouring grid-boxes surrounding e, the algorithm254
considers the upstream grid-box sas well as the two grid-boxes neighbour-255
ing s(i. e., following the example, the 3 grid-boxes to the West, North-West256
and South-West of e). Among these 3 candidate grid-boxes the grid-box of257
maximum IVT is detected.258
5. If this maximum IVT value exceeds the predefined percentile threshold Pt(the259
tracking percentile which not necessarily equals Pd, see also Table 4), the grid-260
box is retained as the new targeted grid box e. In this case, the algorithm261
proceeds to 3). Otherwise, it is stopped at this point in time and proceeds to262
the next point in time.263
6. The algorithm continues until 5) is not met any more or until the detected264
IVT structure exceeds a length of lgrid-boxes or in case a grid-box is detected265
twice, which can occur if the algorithm completely orbits a low pressure system.266
Note that ldepends on the horizontal resolution of the dataset and equals 32,267
40, 70 and 107 grid-boxes for NCEP/NCAR, NOAA-20C, ERA-20C and ERA-268
Interim respectively. For the ideal case of a purely meridional AR with no zonal269
displacement, this roughly corresponds to a longitude of 11000 km.270
7. If the longitude of the detected IVT structure exceeds a threshold of 3000 km271
(spherical distance is considered), the detection region bis said to be affected272
by an AR at this point in time. If it is shorter than 3000 km, the structure is273
not considered an AR.274
Considering the reference period 1979-20092,Pdand Ptwere calculated sepa-275
rately for each grid box and month. Based on a comparison with the ARs detected276
in Neiman et al (2008) and Dettinger et al (2011), Lavers et al (2012) suggested the277
use of the 85th percentile for Pdwhich, however, was replaced by other plausible278
values in some studies [e. g. Warner et al (2015)]. Thus, a secondary goal of the279
present study is to explore how sensitive the results are to variations not only in280
Pdbut also in Pt. To this aim, our tracking algorithm was applied 6 times using 6281
distinct combinations of the two parameters (see Table 4). The corresponding six282
values will hereafter be referred to as the “percentile sample”. Its range describes283
the method-related sensitivity of the results. Additional sensitivity test were con-284
ducted 1) taking into account persistent and independent AR events only and/or285
2) intentionally turning-off our algorithm’s capability to track towards the N, NE,286
E and SE and/or 3) considering a length criterion of >2000 instead of >3000287
km. An event is considered “persistent” if a given target region is continuously288
affected by an AR for at least 18 hours and if it is separated from other events by289
more than 24 hours (Lavers et al, 2012).290
3.2 Reanalysis Comparison and Association with Circulation Indices291
For each target region and season (OND or JFM), and each of the 6 AR defini-292
tions mentioned above, the seasonal AR-counts from the two long-term reanalyses293
2common to all applied reanalysis datasets
8 S. Brands et al.
(NOAA-20C and ERA-20C) are compared in terms of similarity in their climato-294
logical mean as expressed by the bias:295
bias =¯y¯x
¯x×100 (5)
where ¯xand ¯yare the climatological mean values of the seasonal AR-counts in296
the two reanalyses. Here, NOAA-20C is assumed to be the reference reanalysis x.297
Similarity in terms of inter-annual variability is measured by correlating these298
counts with Spearman’s rank correlation coefficient (rs). Prior to calculating rs,299
the year-to-year AR-count time series are optionally de-trended using Poisson300
regression with a log link function (Lavers et al, 2012).301
To identify possible variations along the course of the entire study period (1900-302
2010), a 31-year window moving forward by one year from the start of the study303
period (1900-1930) till its end (1980-2010) is used and the above mentioned sim-304
ilarity measures, as well as the de-trending applied prior to calculating rs, are305
calculated separately for each sub-period. Apart from comparing the AR-counts306
from the two long-term reanalyses, rs is also used to associate these counts with307
the circulation indices listed in Table 3. Since the latter are continuous variables,308
they are optionally de-trended using ordinary least-squares regression instead of309
Poisson regression. The significance of rs is assessed with a two-tailed Student310
t-test conducted at the 5%-level, assuming temporal independence of the applied311
time series.312
4 Results313
4.1 AR-Detection and Tracking314
Figure 3 provides an illustrative example of the algorithm’s capability to detect315
and track AR structures. The figure shows an AR affecting southern Norway on316
11 January 1971 OO UTC, as retrieved from NCEP/NCAR, NOAA-20C and317
ERA-20C (panels a, b and c respectively). Colour shadings and vector lengths are318
proportional to the magnitude of the vertically integrated water vapour flux. The319
direction of the flow is indicated by the orientation of the vectors and the cyan line320
represents the AR-track found by the algorithm. The initial landfall of this AR was321
detected earlier and this particular point in time is chosen to show the algorithm’s322
capability to track towards the N, NE, E and SE (SE in this case, as described323
below) at any point along the AR track. This “eastward tracking” capability was324
not accounted for in the initial formulation of the Lavers et al (2012) algorithm,325
able to track towards the S, SW, W, NW only. Albeit this was corrected in the326
later versions of this algorithm (Lavers and Villarini, 2013, 2015), these do not327
do account for 180curves as those shown in Figure 3. Starting from a given328
detection barrier (e. g. 10W for the case of western Europe), the Lavers and329
Villarini (2013) algorithm moves towards the West and tracks the maximum IVT330
threshold at each longitude. For the structure being an AR in Lavers and Villarini331
(2013), the tracked IVT values must exceed the assumed percentile threshold along332
a longitudinal distance of 20. What is key for the understanding of our method333
is that the Lavers and Villarini (2013) algorithm only detects one grid-box per334
longitude. To perform a 180turn, however, a second IVT value exceeding the335
20th-Century Atmospheric Rivers 9
threshold must be located at the same longitude further to the South (see Figure336
3c) and this is not accounted for by Lavers and Villarini (2013), to the authors’337
knowledge. Telling the algorithm to move to the east, starting from the detection338
barrier, does not solve this problem either. Here, it will be shown that even though339
this limitation is of minor importance in Europe, it is detrimental to AR-detection340
in some regions along the west coast of North America (see below).341
In spite of distinct native horizontal resolutions and applied data assimila-342
tion strategies, the three reanalyses produce virtually identical results for the343
AR event shown in Figure 3. Since the direction of the flow is scanned prior344
to searching the grid-box of maximum IVT, the algorithm correctly moves up-345
stream after detecting the AR in southern Norway. The “curves” of the flow are346
captured well and so is the SE flow between the British Isles and the Iberian347
Peninsula. Finally, the algorithm stops in the central subtropical Atlantic be-348
cause the allowed maximum of tracked grid-boxes (l) is exceeded. As an ex-349
tension to this illustrative example, the Atmospheric River Archive available at350
http://www.meteo.unican.es/atmospheric-rivers documents all ARs detected351
in the 13 target regions displayed in Figure 1 during the period 1900-2010 (ERA-352
20C is compared to NOAA-20C) and 1979-2014 (only ERA-Interim is shown).353
To draw some more general conclusions on the relevance of the “eastward354
tracking” capability, Figure 4 displays the fraction of ARs that are detected if this355
capability is intentionally turned off (Fnoeast):356
Fnoeast =ARnoeast
ARall
×100 (6)
where ARnoeast is the seasonal AR-count retrieved from an algorithm not357
capable to track towards the N, NE, E and SE, and ARall is the respective count358
obtained from the fully capable algorithm as described above.359
Figure 4 illustrates that eastward tracking is more relevant during OND than360
during JFM and more so in North America than in Europe. In the Gulf of Alaska,361
up to 70% of the ARs are “lost” if eastward tracking is not considered, which is due362
to the fact that ARs approaching this region from southerly directions frequently363
have a slight eastward component near landfall and turn to westerly directions364
when further tracked upstream. For an illustrative example of this phenomenon,365
the interested reader is referred to the AR-detections in December 2014 (see reanal-366
ysis: ”ERA-Interim”, continent: ”western North America” and region: ”northern367
Gulf of Alaska” at http://www.meteo.unican.es/atmospheric-rivers).368
Finally, 90% of the AR-events documented in Dettinger et al (2011) (see their369
table 1) coincide with the ERA-Interim based AR-detections provided by the At-370
mospheric River Archive if the target day documented in Dettinger et al (2011)371
is “relaxed” by ±18 hours. The “missing” 10% can largely be explained by the372
comparatively long AR-length criterion applied here (>3000 km). If our algorithm373
is re-run with a shorter length criterion (>2000 km) the coincidence rate rises to374
97%. Interestingly, even though a longer length criterion is assumed, our archive375
contains more events than the Dettinger et al (2011) archive.376
10 S. Brands et al.
4.2 Temporal Variations in Reanalysis Similarity during 1900-2010377
Figure 5 displays the year-to-year AR-count sequence obtained from NOAA-20C378
(blue) and ERA-20C (red) respectively; results are for the OND-season in this case.379
As above, the lines and shadings refer to the mean and range of the 6 seasonal380
AR-count values per year obtained from the 6 considered percentile combinations381
listed in Table 4. For the sake of completeness, AR-counts from NOAA-20C extend382
to 2012. Panels a to h refer to the results for Europe, panels i to h to the results for383
western North America. Figure 6 shows the respective results for the JFM-season.384
Note that the mean value of the 1900-2010 AR counts is displayed in the header of385
each panel for each of the two datasets (first number = NOAA-20C mean, second386
number = ERA-20C mean).387
Results for the 31-year “running” bias in the AR-counts (ERA-20C minus388
NOAA-20C w.r.t to NOAA-20C, see Equation 5) are shown in Fig. 7. On the x-389
axis of each panel, the centre-year of a specific sub-period is displayed (e. g. “1920”390
refers to the time period 1905-1935). We will hereafter refer to this centre year391
instead of mentioning the entire sub-period. On the y-axis, the bias is displayed as392
the percentage deviation from the mean of the reference reanalysis for that sub-393
period, which is NOAA-20C. Again, the lines and shadings refer to the mean and394
range of the 6 bias values obtained from the percentile sample. To measure the395
stationarity of the bias, the standard deviation (std) of the 81 percentile-sample396
mean values for a given season and target region is displayed in the header of each397
panel (first number = OND std, second number = JFM std).398
A visual inspection of the year-to-year time-series relevant for Europe (see pan-399
els a to h in Figures 5 and 6) reveals that up to at least the 1970s (1930s in northern400
Norway and 1940s in northern Iberia) NOAA-20C produces systematically more401
ARs than ERA-20C whereas the opposite is the case from approximately the 1980s402
onward. This translates into a change in the sign of the bias from negative values403
down to approximately 40% at the start of the 20th-century to positive values up404
to approximately +25% in the recent past (see panels a to h in Figure 7). As indi-405
cated by the standard deviation in the header of each panel, the non-stationarity406
of the bias is more pronounced in OND than in JFM, with the largest values407
obtained for the British Isles.408
Contrary to what was found for Europe, ERA-20C produces up to twice as409
many ARs as NOAA-20C in western North America (exception: southern Cali-410
fornia, see panels i to m in 7). Such a large bias might be explained by the fact411
that the 56-member ensemble of NOAA-20C, during the “data-sparse” start of412
the 20th-century, suffers such a large spread that the percentile thresholds listed413
in Table 4 are exceeded by the ensemble-mean values far less often than during414
the later (“data-rich”) period, leading to a reduction in AR detections for this415
reanalysis [see also Champion et al (2015)]. ERA-20C is a deterministic reanalysis416
and is therefore not affected by this issue. Nevertheless, due to the general lack417
of data, it cannot be expected to provide realistic AR-counts at the start of the418
century either. By approximately 1920s (with the exception of the northern Gulf419
of Alaska), the bias for western North America decreases to a magnitude compa-420
rable to the that found for Europe. As for Europe, temporal variations in the bias421
are more pronounced during the OND- than during the JFM-season, particularly422
over the southern and northern Gulf of Alaska.423
20th-Century Atmospheric Rivers 11
Figure 8 displays the results of the running correlation analyses. On the y-axis,424
the rank correlation coefficient (rs), as well as the critical values for a two-tailed425
t-test applied at a test-level of 5% are shown (see dashed lines). Regarding the426
European regions, rs is systematically lower and its range (reflecting the method427
related uncertainty) systematically larger during the OND than during the JFM428
season. Values generally decrease as one moves backward in time. With rs ex-429
ceeding +0.6 in nearly any case, the AR counts’ inter-annual variability is roughly430
similar in both datasets even at the very beginning of the 20th century. From 1955431
onwards, rs is greater than- or close to +0.8, indicating a close similarity during432
the last 7 decades of comparison. However, OND values in Norway —for unknown433
reasons— are smaller during the recent past than during the mid-20th-century434
(see panels g and h in Figure 8).435
In contrast to the result for Europe, rs values along the west coast of North436
America are insignificant or even negative at the start of the century (note the437
distinct scale of the y-axes). Another distinction is that the rs values in OND438
are much closer to those obtained for JFM and actually are larger during the439
first decades of the 20th century. Following the running rs forward in time, a440
value of approximately +0.5 is at the latest reached around 1935 and a value of441
approximately +0,8 is so around 1965.442
The method-related sensitivity of the results is small in comparison to the mean443
value (compare shadings with lines in Figures 5 to 7), which is generally also the444
case for the forthcoming results. Reducing the AR-length criterion to >2000 km445
slightly improves the reanalysis agreement without, however, bringing the huge446
differences found over North America at the start of the century to an acceptable447
level (not shown).448
4.3 Temporal Variations in the Link to the NAO and Aleutian Low during449
1900-2010450
Since the inter-annual variability of the AR-counts in Europe was found to be451
similar in the two long-term reanalyses even at the start of the 20-century, we452
proceed to assess their association with the seasonal-mean NAO (the station-based453
index is used here). To this, a running rank correlation analysis is applied in the454
aforementioned configuration, i. e. a 31-year moving window is used. The same is455
done for the AR-counts along the west coast of North America, having in mind456
that insignificant rs were obtained at the start of the 20th-century when comparing457
the two reanalysis there. The results for the OND and JFM seasons are displayed458
in Figures 9 and 10 respectively. Blue lines and shadings are for AR-counts from459
NOAA-20C and red ones are for AR-counts from ERA-20C. Also shown are the460
critical values for a significant rs at a test-level of 5% (see dashed lines).461
Similar to the well-known correlation dipole for seasonal precipitation totals462
(Hurrell, 1995), AR-counts in southern Europe are inversely related to the NAO463
whereas in northern Europe a positive relationship is found, which is in agreement464
with the Lavers and Villarini (2013) results (see panels a to h in Figures 9 and 10).465
These relationships are generally weaker and less stationary (i.e. variable in time)466
during OND than during JFM. In the two southernmost and the two northernmost467
regions, rs in JFM is significant for any of the 81 considered sub-periods indicating468
a temporally robust link to the NAO during this season. In northern Iberia and469
12 S. Brands et al.
western France, however, rs is significant from 1940 until the end of the 1970s only.470
Similarly, over the British Isles, rs is insignificant from approximately 1915 to 1921471
and —for NOAA-20C— also from 1960 to 1970, indicating that the NAO-link in472
the three central regions of the European Atlantic seaboard is subject to non-473
negligible variations along the course of the century. During the OND season, rs is474
generally insignificant except for Morocco and southern Iberia from approximately475
1915 to 1930 and from 1975 onwards, and for southern Norway from approximately476
1940 to 1970. Albeit somewhat larger during OND than during JFM, dataset-477
induced differences are generally small for Europe.478
As expected from the results of the reanalysis comparison, dataset-induced479
differences concerning the link between AR-counts in western North America and480
the Aleutian low can be larger than 0.5 correlation points at the start of the cen-481
tury (see panels i to m in Figures 9 and 10). In the two southernmost regions,482
rs is insignificant or prone to large dataset-differences along the entire study pe-483
riod, except during the JFM-season where significantly negative rs are obtained484
from 1945 to 1960 (in North California-Oregon-Washington only) and from 1990485
onwards (in both regions).486
In the 3 remaining regions, rs for JFM is significantly negative from approx-487
imately 1950 onwards, except for the northern Gulf of Alaska where insignifi-488
cant values are obtained in the very recent past (from 1990 onwards). During489
OND, dataset-induced differences in the results are relatively large until at least490
1955. Thereafter, these differences diminish, revealing sig. negative rs in British491
Columbia and the southern Gulf of Alaska, which, however, decrease when ap-492
proaching the present, eventually becoming insignificant from 1970 / 1980 on-493
wards. This decrease is most pronounced in British Columbia. The OND values494
for the northern Gulf of Alaska are constantly sig. negative from approximately495
1955 onwards.496
As can be seen from Figures S01 and S02 in the supplementary online ma-497
terial, similar results are obtained when the AR-count and index time series are498
de-trended (separately in each 31-year period of the running analyses) prior to499
calculating rs.500
4.4 Relationship to the Large-Scale Atmospheric Circulation during 1950-2010501
Figure 11 shows the rs between the seasonal AR counts in the eight considered Eu-502
ropean target regions and the seasonal-mean large-scale circulation indices relevant503
there. Unlike the running analyses conducted above, rs in this section is calculated504
once for the period 1950-20103, or 1979-2013 in case ARs from ERA-Interim are505
considered. As above, the bars and errorbars in a given panel refer to the mean506
and range of the percentile sample (see Table 4). The critical values obtained from507
a two-sided t-test conducted at a test-level of 5% are indicated by dashed lines.508
Along the rows, results for ARs retrieved from NCEP/NCAR, NOAA-20C, ERA-509
20C and ERA-Interim are displayed from the top to the bottom. The OND-, JFM-510
and ONDJFM results are provided in columns 1-3.511
The three re-analyses covering the 1950-2010 period produce very similar re-512
sults (see rows 1-3 in Figure 11). During both OND and JFM (see columns 1+2),513
3note that the indices provided by the Climate Prediction Center are available from 1950
onwards only
20th-Century Atmospheric Rivers 13
relationships to the NAO are strongly negative in the southern European regions,514
weaker in the central regions and strongly positive in the northern regions, thereby515
depicting the well-known correlation dipole found for precipitation in earlier stud-516
ies (Hurrell, 1995; Qian et al, 2000). Since rHNAO over the period 1950-2010 is517
significant for almost any region irrespective of the considered season and dataset518
and since the magnitude of rs is close to 0.8 in some cases, the NAO, and particu-519
larly the NAO based on SLP, is the most important circulation pattern influencing520
extended winter AR counts in Europe if the results are seen as a whole. Excep-521
tions from this general finding are mainly found during the OND season in which522
case the AR-counts over western Iberia, northern Iberia and western France are523
more strongly linked to the EA than to the NAO (rsEA lies in between +0.5 and524
+0.7) and those over the British Isles are more strongly linked to the SCAND525
(rsSC AN D ≈ −0.4). Links to the NAO and SCAND are more pronounced during526
the JFM than during the OND season whereas the opposite is found for the links to527
the EA. During JFM, rsSC AN D is between 0.4 and0.65 in the three northern-528
most regions and +0.4 in the two southernmost ones. During OND, rsSCAN D
529
is significant in the three northernmost regions only. Links to the EA/WR are530
significant in the latter three regions during OND and in northern Norway during531
JFM. Links to the POL are generally insignificant except during JFM in northern532
Iberia and western France. When considering the entire winter half-year (see col-533
umn 3 in Figure 11), the strength of the teleconnections generally lies in between534
the values obtained for OND and JFM.535
A series of additional sensitivity tests were conducted for the ONDJFM sea-536
son and the respective results are displayed in Figure 12. The first column refers537
to solely considering persistent ARs, the second to “turning off” our algorithm’s538
capability to track towards the N, NE, E and SE, and the third to using a length539
criterion of >2000 instead of 3000 km (over the sphere). From these additional540
experiments, it becomes obvious that the inclusion of the persistence criterion541
weakens the link between the ONDJFM-AR counts and the NAO indices partic-542
ularly over the British Isles (compare first column in Figure 12 with last column543
in Figure 11). This effect is most appreciable in case the experimental set-up con-544
sidered in Lavers et al (2012)4is used in combination with a length criterion of545
>3000 km, in which case rsCP C NAO is consistently insignificant (see Table 5a).546
For the experimental set-up used in Ramos et al (2015)5, the detrimental effect547
of the persistence criterion leads to insignificant rsCP C N AO for two out of six548
percentile combinations irrespective of the applied length criterion (see Table 5).549
Finally, neither disabling the algorithm’s capability to track towards the N, NE, E550
and SE nor applying the alternative length criterion does notably alter the results551
in this region of the world (compare columns 2 and 3 in Figure 12 with the last552
column in Figure 11).553
The respective results for the west coast of North America and the circulation554
indices relevant there are shown in Figure 13 and Figure 14 respectively. Instanta-555
neous AR counts along the Gulf of Alaska are positively correlated with the PNA556
and negatively correlated with the NP (see Figure 13). Yet significant in both557
seasons, these links are more pronounced during JFM than during OND (compare558
4i. e. considering persistent ARs during ONDJFM 1979/80 - 2009/10 derived from ERA-
Interim
5i. e. considering persistent ARs during ONDJFM 1950/51 - 2011/12 derived from
NCEP/NCAR
14 S. Brands et al.
first and second column). During JFM, AR-counts in SouthCal, NorthCal-OR-WA559
and British Columbia are also significantly associated with the NP index, with the560
exception of the AR-counts in SouthCal and NorthCal-OR-WA obtained from561
ERA-20C, in which case results are on the limit to significance (see panel h). It is562
during the JFM-season only when ARs over the two aforementioned regions are563
significantly related to the WP. Teleconnections involving the AR-counts in the564
Gulf of Alaska are systematically weaker during 1979-2013 than during 1950-2010565
(compare last row to rows 1-3 in Figure 13). This finding is not dataset-dependent566
(see the “late” results of the running correlation analyses in Figures 9 and 10) and567
might be explained by the systematic strengthening of the wintertime Aleutian568
low after the Pacific Climate Shift in 1976/77 (Deser et al, 2004).569
Unlike in Europe, the persistence criterion’s effect on rs is not systematic along570
the west coast of North America, i. e. can lead to a slight increase or decrease in rs571
(compare first column in Figure 14 with last column in Figure 13). If the algorithms572
capability to track towards the N, NE, E and SE is disabled, teleconnections573
with the PNA an NP become insignificant in the northern and southern Gulf574
of Alaska (compare second column in Figure 14 with last column in Figure 13).575
Thus, the inclusion of this capability is key to properly capture the inter-annual576
variability of the AR-counts in these regions. As was the case for Europe, applying577
the alternative length criterion does not notably alter the results (compare last578
column in Figure 14 with last column in Figure 13).579
5 Summary, Discussion and Concluding Remarks580
On the basis of a new algorithm operating on the magnitude and direction of IVT,581
time series of year-to-year AR occurrence counts were calculated for 13 target re-582
gions along the west coasts of Europe (including North Africa) and North America.583
This was done separately for the OND and JFM-seasons using 6-hourly instanta-584
neous data from 4 distinct reanalyses, two of which extend back to the early 20th585
century (1900). In principle, no AR-persistence criterion was considered.586
A “running” comparison of the seasonal AR counts from the two long-term587
reanalyses over the period 1900-2010 revealed:588
1. Biases which are especially pronounced in, but not limited to, the early 20th-589
century. With up to >100%, the biases during this early period are more590
severe in western North America than in Europe.591
2. Along the west coast of Europe, the two reanalyses produce a similar inter-592
annual variability even at the start of the 20th-century (rank correlation 593
+0.6). This is in sharp contrast to the near-to-zero correlation found along the594
west coast of North America during the same period. In this region, rs steadily595
increases until approximately 1945-75 and thereafter remains constant at a level596
+0.8.597
Encouraged by finding 2), the stationarity of AR-NAO link was traced back598
to the early 20th-century using a 31-year running correlation analysis over the599
period 1900-2010. Albeit rs for individual target regions can vary along the time-600
axis, particularly during the OND-season, the dipole found on continental-scale601
(i.e. looking at the conjunction of target regions) is generally found in each sub-602
period, indicating that it is a robust feature along the course of the entire 20th603
20th-Century Atmospheric Rivers 15
century. Applying the same method for AR-counts along the west coast of North604
America and the strength of the Aleutian Low (as represented by the North Pacific605
Index) revealed larger variations in time which —during the early 20th century—606
are attributable to dataset uncertainties including uncertainties in the NP index607
itself (Trenberth and Hurrell, 1994). From the 1940-1970 climate period onwards,608
however, these uncertainties are small and the detected non-stationarities in the609
above link —which are most pronounced over British Columbia and the southern610
Gulf of Alaska— are likely to reflect real processes. A detailed assessment of the611
causes for this is recommended for the future.612
For the reliable period 1950-2010, the search for atmospheric drivers of sea-613
sonal AR-occurrence counts was extended to circulation patterns others than the614
NAO and NP. For western Europe, the NAO was found to be the most important615
atmospheric driver of AR activity if the results are seen as a whole. In particular,616
the OND and ONDFJM AR-counts over the British Isles and western Iberia are617
significantly linked to the NAO if no persistence criterion is applied. Remarkably,618
if the Lavers et al (2012) persistence criterion is applied, rs values in the two619
aforementioned regions drop to insignificance (or near insignificance) for the ex-620
perimental set-ups that most closely mirrow those applied in Lavers et al (2012)621
and Ramos et al (2015). However, despite conceptual similarities, the tracking al-622
gorithm applied here is not identical to that used in the above mentioned studies.623
Therefore, it cannot be ultimately demonstrated that the persistence criterion is624
the responsible for, e. g., the insignificant AR-NAO link found in Ramos et al625
(2015). During the OND-season, AR-counts along the Atlantic coast of Iberia and626
France were found to be more strongly linked to the East Atlantic pattern than627
to the North Atlantic Oscillation. As formerly pointed out in Lavers et al (2012),628
AR-counts over the British Isles were found to be significantly associated with the629
Scandinavian index. Here, it was shown that this index is a significant driver of630
the AR-activity in Norway.631
Apart from the aforementioned links to the Aleutian low, the PNA was found632
to significantly alter the AR-counts in British Columbia and the Gulf of Alaska633
during the JFM-season, which is in agreement with the Guan and Waliser (2016).634
During the OND season these links are generally weaker, leading to insignificant635
results over British Columbia. It is during JFM only when AR-counts along the636
U.S. west coast are significantly related to the West Pacific pattern.637
The above mentioned uncertainties in the “quasi-observed” climatological mean638
AR-counts should be taken into account when evaluating the bias of e. g. the639
CMIP5 Earth System Models (Taylor et al, 2012) against either of the two long-640
term reanalyses, particularly during the early 20th century. This type of uncer-641
tainty is also expected to hinder the association of specific AR events from either of642
the two reanalyses with hydrological extreme events documented by other sources.643
Finally, the close agreement on the seasonal AR-counts’ inter-annual variability644
back to 1900-31 for Europe and 1920-51 for western North America permits to645
assess their variability (and predictability) with longer time series, as was shown646
here for their link to the NAO and NP indices. A logical future step is to relate647
AR-activity to sea-surface temperature variations on multiple time-scales (Zhang648
et al, 1997; Trenberth et al, 1998; Delworth and Mann, 2000; Broennimann, 2007).649
Acknowledgements The authors would like to thank Jorge Eiras-Barca, Daniel Garaboa,650
Dr. Gonzalo Miguez-Macho, Dr. David Lavers and two anonymous referees for their construc-651
16 S. Brands et al.
tive criticism and helpful advice. They acknowledge the use of the climate indices provided652
by UCAR https://climatedataguide.ucar.edu/climate-data/ and the Climate Prediction653
Center http://www.cpc.ncep.noaa.gov/, the ECMWF ERA-20C and ERA-Interim reanaly-654
ses http://apps.ecmwf.int/datasets, the NOAA-CIRES 20th Century Reanalysis version 2655
http://www.esrl.noaa.gov/psd/data/gridded/data.20thC_ReanV2.html and the NCEP/NCAR656
reanalysis 1 http://www.esrl.noaa.gov/psd/data/gridded/data.ncep.reanalysis.html. SB657
would like to thank the TRAGSA Group and the CSIC JAE-PREDOC programme for financial658
support.659
References660
Bao J, Michelson S, Neiman P, Ralph F, Wilczak J (2006) Interpretation661
of enhanced integrated water vapor bands associated with extratropical cy-662
clones: Their formation and connection to tropical moisture. Mon Weather Rev663
134(4):1063–1080, DOI {10.1175/MWR3123.1}664
Barnston A, Livezey R (1987) Classification, seasonality and persistence of low-665
frequency atmospheric circulation patterns. Mon Weather Rev 115(6):1083–666
1126, DOI {10.1175/1520-0493(1987)115h1083:CSAPOLi2.0.CO;2}667
Brands S, Gutierrez JM, Herrera S, Cofino AS (2012) On the use of reanalysis data668
for downscaling. J Clim 25(7):2517–2526, DOI {10.1175/JCLI-D-11- 00251.1}669
Broennimann S (2007) Impact of El Nino Southern Oscillation on European cli-670
mate. Rev Geophys 45(2), DOI {10.1029/2006RG000199}671
Champion AJ, Allan RP, Lavers DA (2015) Atmospheric rivers do ot explain UK672
summer extreme rainfall. J Geophys Res Atmos DOI {10.1002/2014JD022863}673
Compo GP, Whitaker JS, Sardeshmukh PD, Matsui N, Allan RJ, Yin X, Glea-674
son BE Jr, Vose RS, Rutledge G, Bessemoulin P, Broennimann S, Brunet M,675
Crouthamel RI, Grant AN, Groisman PY, Jones PD, Kruk MC, Kruger AC,676
Marshall GJ, Maugeri M, Mok HY, Nordli O, Ross TF, Trigo RM, Wang XL,677
Woodruff SD, Worley SJ (2011) The Twentieth Century Reanalysis Project. Q678
J R Meteorol Soc 137(654, A):1–28, DOI {10.1002/qj.776}679
Dacre HF, Clark PA, Martinez-Alvarado O, Stringer MA, Lavers DA (2015) How680
do atmospheric rivers form? Bull Amer Meteorol Soc 96(8):1243–1255, DOI681
{10.1175/BAMS-D-14-00031.1}682
Dee DP, Uppala SM, Simmons AJ, Berrisford P, Poli P, Kobayashi S, Andrae U,683
Balmaseda MA, Balsamo G, Bauer P, Bechtold P, Beljaars ACM, van de Berg L,684
Bidlot J, Bormann N, Delsol C, Dragani R, Fuentes M, Geer AJ, Haimberger L,685
Healy SB, Hersbach H, Holm EV, Isaksen L, Kallberg P, Koehler M, Matricardi686
M, McNally AP, Monge-Sanz BM, Morcrette JJ, Park BK, Peubey C, de Ros-687
nay P, Tavolato C, Thepaut JN, Vitart F (2011) The ERA-Interim reanalysis:688
configuration and performance of the data assimilation system. Q J R Meteorol689
Soc 137(656, Part a):553–597, DOI {10.1002/qj.828}690
Delworth T, Mann M (2000) Observed and simulated multidecadal variabil-691
ity in the Northern Hemisphere. Clim Dyn 16(9):661–676, DOI {10.1007/692
s003820000075}693
Deser C, Phillips A, Hurrell J (2004) Pacific interdecadal climate variability: Link-694
ages between the tropics and the North Pacific during boreal winter since 1900.695
J Clim 17(16):3109–3124, DOI {10.1175/1520-0442(2004)017h3109:PICVLBi2.696
0.CO;2}697
20th-Century Atmospheric Rivers 17
Dettinger MD, Ralph FM, Das T, Neiman PJ, Cayan DR (2011) Atmospheric698
Rivers, Floods and the Water Resources of California. Water 3(2):445–478, DOI699
{10.3390/w3020445}700
Eiras-Barca J, Brands S, Miguez-Macho G (2016) Seasonal variations in North701
Atlantic atmospheric river activity and associations with anomalous precip-702
itation over the Iberian Atlantic Margin. J Geophys Res:in press, DOI703
{10.1002/2015JD023379}704
Ferguson CR, Villarini G (2012) Detecting inhomogeneities in the Twentieth Cen-705
tury Reanalysis over the central United States. J Geophys Res Atmos 117, DOI706
{10.1029/2011JD016988}707
Garaboa D, Eiras-Barca J, Huhn F, nuzuri V PM (2015) Langrangian coherent708
structures along atmospheric rivers. Chaos709
Gimeno L, Nieto R, V´azquez M, Lavers D (2014) Atmospheric rivers: a mini-710
review. Front Earth Sci, doi: 10.3389/feart.2014.00002711
Gimeno L, Stohl A, Trigo RM, Dominguez F, Yoshimura K, Yu L, Drumond712
A, Maria Duran-Quesada A, Nieto R (2012) Oceanic and terrestrial sources of713
continental precipitation. Rev Geophys 50, DOI {10.1029/2012RG000389}714
Guan B, Waliser DE, Molotch NP, Fetzer EJ, Neiman PJ (2012) Does the715
Madden-Julian Oscillation influence wintertime atmospheric rivers and snow-716
pack in the Sierra Nevada? Mon Weather Rev 140(2):325–342, DOI {10.1175/717
MWR-D-11-00087.1}718
Guan B, Waliser DE (2016) Detection of atmospheric rivers: Evaluation and ap-719
plication of an algorithm for global studies. J Geophys Res 120:in press, DOI720
{10.1002/2015JD024257}721
Higgins R, Schemm J, Shi W, Leetmaa A (2000) Extreme precipitation events722
in the western United States related to tropical forcing. J Clim 13(4):793–820,723
DOI {10.1175/1520-0442(2000)013h0793:EPEITWi2.0.CO;2}724
Hurrell J (1995) Decadal trends in the North-Atlantic Oscillation - Regional tem-725
peratures and precipitation. Science 269(5224):676–679, DOI {10.1126/science.726
269.5224.676}727
Hurrell J, Kushnir Y, Ottersen G, Visbeck M (2003) The North Atlantic Oscilla-728
tion: Climate significance and environmental Impact, Geophysical Monograph729
Series, vol 134. AGU, Washington, D. C.730
Jiang T, Deng Y (2011) Downstream modulation of North Pacific atmospheric731
river activity by East Asian cold surges. Geophys Res Lett 38, DOI {10.1029/732
2011GL049462}733
Kalnay E, Kanamitsu M, Kistler R, Collins W, Deaven D, Gandin L, Iredell M,734
Saha S, White G, Woollen J, Zhu Y, Chelliah M, Ebisuzaki W, Higgins W,735
Janowiak J, Mo K, Ropelewski C, Wang J, Leetmaa A, Reynolds R, Jenne736
R, Joseph D (1996) The NCEP/NCAR 40-year reanalysis project. Bull Amer737
Meteorol Soc 77(3):437–471738
Kim HM, Alexander MA (2015) ENSO’s modulation of water vapor transport739
over the Pacific-North American region. J Clim 28(9):3846–3856, DOI {10.1175/740
JCLI-D-14-00725.1}741
Kim J, Waliser DE, Neiman PJ, Guan B, Ryoo JM, Wick GA (2013) Effects of742
atmospheric river landfalls on the cold season precipitation in California. Clim743
Dyn 40(1-2):465–474, DOI {10.1007/s00382-012-1322-3}744
Knippertz P, Wernli H (2010) A Lagrangian climatology of tropical moisture ex-745
ports to the Northern Hemispheric extratropics. J Clim 23(4):987–1003, DOI746
18 S. Brands et al.
{10.1175/2009JCLI3333.1}747
Knippertz P, Wernli H, Glaeser G (2013) A global climatology of tropical moisture748
exports. J Clim 26(10):3031–3045, DOI {10.1175/JCLI-D-12-00401.1}749
Lavers DA, Villarini G (2013) The nexus between atmospheric rivers and ex-750
treme precipitation across Europe. Geophys Res Lett 40(12):3259–3264, DOI751
{10.1002/grl.50636}752
Lavers DA, Villarini G (2015) The contribution of atmospheric rivers to precip-753
itation in Europe and the United States. Journal of Hydrology 522:382–390,754
DOI {10.1016/j.jhydrol.2014.12.010}755
Lavers DA, Allan RP, Wood EF, Villarini G, Brayshaw DJ, Wade AJ (2011)756
Winter floods in Britain are connected to atmospheric rivers. Geophys Res Lett757
38, DOI {10.1029/2011GL049783}758
Lavers DA, Villarini G, Allan RP, Wood EF, Wade AJ (2012) The detection of759
atmospheric rivers in atmospheric reanalyses and their links to British winter760
floods and the large-scale climatic circulation. J Geophys Res Atmos 117, DOI761
{10.1029/2012JD018027}762
Leung LR, Qian Y (2009) Atmospheric rivers induced heavy precipitation and763
flooding in the western US simulated by the WRF regional climate model. Geo-764
phys Res Lett 36, DOI {10.1029/2008GL036445}765
Neiman P, Persson P, Ralph F, Jorgensen D, White A, Kingsmill D (2004)766
Modification of fronts and precipitation by coastal blocking during an intense767
landfalling winter storm in southern California: Observations during CALJET.768
Mon Weather Rev 132(1):242–273, DOI {10.1175/1520-0493(2004)132h0242:769
MOFAPBi2.0.CO;2}770
Neiman PJ, Ralph FM, Wick GA, Lundquist JD, Dettinger MD (2008) Meteoro-771
logical characteristics and overland precipitation impacts of atmospheric rivers772
affecting the West Coast of North America based on eight years of SSM/I satel-773
lite observations. J Hydrometeorol 9(1):22–47, DOI {10.1175/2007JHM855.1}774
Newell R, Newell N, Zhu Y, Scott C (1992) Tropospheric rivers - A pilot-study.775
Geophys Res Lett 19(24):2401–2404, DOI {10.1029/92GL02916}776
Newman M, Kiladis GN, Weickmann KM, Ralph FM, Sardeshmukh PD (2012)777
Relative contributions of synoptic and low-frequency eddies to time-mean at-778
mospheric moisture transport, including the role of atmospheric rivers. J Clim779
(25):73417361, DOI {http://dx.doi.org/10.1175/JCLI-D-11- 00665.1}780
Payne AE, Magnusdottir G (2014) Dynamics of landfalling atmospheric rivers over781
the North Pacific in 30 Years of MERRA reanalysis. J Clim 27(18):7133–7150,782
DOI {10.1175/JCLI-D-14-00034.1}783
Poli P, Hersbach H, Tan D, Dee D, Th´epaut JN, Simmons A, Peubey C, Laloyaux784
P, Komori T, Berrisford P, Dragani R, Temolet Y, H´olm E, Bonavita M, Isaksen785
L, Fisher M (2013) The data assimilation system and initial performance eval-786
uation of the ECMWF pilot reanalysis of the 20th-century assimilating surface787
observations only (era-20c). Tech. rep., ERA Report Series 14788
Qian B, Corte-Real J, Xu H (2000) Is the North Atlantic Oscillation the most im-789
portant atmospheric pattern for precipitation in Europe? J Geophys Res Atmos790
105(D9):11,901–11,910, DOI {10.1029/2000JD900102}791
Ralph F, Neiman P, Wick G (2004) Satellite and CALJET aircraft observations792
of atmospheric rivers over the eastern north pacific ocean during the winter of793
1997/98. Mon Weather Rev 132(7):1721–1745, DOI {10.1175/1520-0493(2004)794
132h1721:SACAOOi2.0.CO;2}795
20th-Century Atmospheric Rivers 19
Ralph FM, Coleman T, Neiman PJ, Zamora RJ, Dettinger MD (2013) Observed796
impacts of duration and seasonality of atmospheric-river landfalls on soil mois-797
ture and runoff in coastal Northern California. J Hydrometeorol 14(2):443–459,798
DOI {10.1175/JHM-D-12-076.1}799
Ramos AM, Trigo RM, Liberato MLR, Tome R (2015) Daily precipitation extreme800
events in the Iberian Peninsula and its association with atmospheric Rivers. J801
Hydrometeorol 16(2):579–597, DOI {10.1175/JHM-D-14-0103.1}802
Sodemann H, Stohl A (2013) Moisture origin and meridional transport in atmo-803
spheric rivers and their association with multiple cyclones. Mon Weather Rev804
141(8):2850–2868, DOI {10.1175/MWR-D-12-00256.1}805
Sterl A (2004) On the (in)homogeneity of reanalysis products. J Clim 17(19):3866–806
3873, DOI {10.1175/1520-0442(2004)017h3866:OTIORPi2.0.CO;2}807
Sturaro G (2003) A closer look at the climatological discontinuities present in the808
NCEP/NCAR reanalysis temperature due to the introduction of satellite data.809
Clim Dyn 21(3-4):309–316, DOI {10.1007/s00382-003-0334-4}810
Taylor KE, Stouffer RJ, Meehl GA (2012) An overview of CMIP5 and the811
experiment design. Bull Amer Meteorol Soc 93(4):485–498, DOI {10.1175/812
BAMS-D-11-00094.1}813
Trenberth K, Hurrell J (1994) Decadal atmosphere-ocean variations in the Pacific.814
Clim Dyn 9(6):303–319, DOI {10.1007/BF00204745}815
Trigo RM, Pozo-V´azquez D, Osborn TJ, Castro-D´ıez Y, G´amiz-Fortis S, Esteban-816
Parra MJ (2004) North Atlantic oscillation influence on precipitation, river flow817
and water resources in the Iberian Peninsula. Int J Climatol 24(8):925–944,818
DOI {10.1002/joc.1048}819
Trenberth K, Branstator G, Karoly D, Kumar A, Lau N, Ropelewski C (1998)820
Progress during TOGA in understanding and modeling global teleconnections821
associated with tropical sea surface temperatures. J Geohphys Res Atmos822
103(C7):14,291–14,324, DOI {10.1029/97JC01444}823
Warner MD, Mass CF, Salath´e Jr EP (1998) Changes in winter atmospheric rivers824
along the North American west coast in CMIP5 climate models. J Hydromete-825
orol 16:118-128, DOI {http://dx.doi.org/10.1175/JHM-D-14- 0080.1}826
Zhang Y, Wallace J, Battisti D (1997) ENSO-like interdecadal variability: 1900-827
93. J Clim 10(5):1004–1020, DOI {10.1175/1520-0442(1997)010h1004:ELIVi2.0.828
CO;2}829
Zhu Y, Newell R (1994) Atmospheric rivers and bombs. Geophys Res Lett830
21(18):1999–2002, DOI {10.1029/94GL01710}831
Zhu Y, Newell R (1998) A proposed algorithm for moisture fluxes from atmospheric832
rivers. Mon Weather Rev 126(3):725–735, DOI {10.1175/1520-0493(1998)833
126h0725:APAFMFi2.0.CO;2}834
20 S. Brands et al.
Table 1 Considered reanalysis datasets, 6h-instantaneous values are applied in any case.
Listed are the acronyms used throughout the study, the full names, horizontal resolutions (lat.
×lon.), reference publications and the number of runs conducted for each reanalysis. The
ensemble-mean data from NOAA-20C are used in the present study.
Acronym Full Name Resolution Reference Nr. runs
NCEP/NCAR NCEP/NCAR Reanalysis 1 2.5×2.5Kalnay et al (1996) 1 run
NOAA-20C NOAA CIRES 20th-Century Reanalysis v2 2×2Compo et al (2011) 56-member ensemble
ERA-20C ECMWF ERA-20C Reanalysis 1.125×1.125Poli et al (2013) 1 run
ERA-Interim ECMWF ERA-Interim Reanalysis 0.75×0.75Dee et al (2011) 1 run
Table 2 List of multipliers used for multiplication with qu and qv at a given pressure level p.
Horizontal bars indicate that the data at the corresponding pressure level were not available
from the data provider. See text for more details.
p NCE P /NC AR NOAA 20C
300 5000 2500
350 - 5000
400 10000 5000
450 - 5000
500 10000 5000
550 - 5000
600 10000 5000
650 - 5000
700 12500 5000
750 - 5000
800 - 5000
850 11250 5000
900 - 5000
925 7500 -
950 - 5000
1000 3750 2500
Table 3 Considered large-scale atmopsheric circulation indices
Used Acronym Description Provider Citation
H-NAO J. Hurrell’s NAO index based on PCA and SLP fields UCAR Hurrell et al (2003)
station-NAO J. Hurrell’s NAO index based on SLP station values
CPC-NAO NAO index based on rotated PCA and geopotential height fields CPC Barnston and Livezey (1987)
EA East Atlantic Pattern index
SCAND Scandinavian Pattern index
EA/WR East Atlantic / Western Russia index
PNA Pacific-North American pattern index
WP West Pacific Index
NP North Pacific / Aleutian Low index UCAR Trenberth and Hurrell (1994)
20th-Century Atmospheric Rivers 21
Table 4 The 6 percentile combinations used for AR detection and tracking. Pdis the percentile
threshold used for detection at the region of AR-arrival and Ptis the percentile threshold used
along the track of the AR.
Number PdPt
1 85 75
2 85 80
3 85 85
4 90 75
5 90 80
6 90 85
Table 5 Rank correlation coefficient (rounded to the next integer ×100) measuring the link
between AR-counts and Climate Prediction Centre’s NAO index during the ONDJFM-season,
with and without considering the Lavers et al (2012) persistence criterion. Results are for the
experimental set-ups most closely refelecting Lavers et al (2012) (setup 1) and Ramos et al
(2015) (setup 2); see text for more details. Significant results (α= 0.05, two-sided t-test) are
printed in bold. a) results for an AR length criterion of >3000 km, b) results for >2000 km.
a) 3000 km Setup 1 Setup 2
Without persistence criterion 48,49,46,45,45,40 -37,-38,-37,-39,-40,-41
With persistence criterion 17, 15, 14, 24, 19, 26 -26,-35, -21, -32, -23, -34
b) 2000 km
Without persistence criterion 52,54,55,53,53,51 -33,-33,-32,33,-35,-36
With persistence criterion 44,40,37, 34, 42,48 -18, -31, -20, -31, -25, -32
22 S. Brands et al.
0
400
800
1200
meters a.s.l.
1
8
7
6
5
4
3
2
(b) W North America(a) W Europe and N Africa
(b)(a)
1
2
3
4
5
1 Morocco
2 S Iberia
3 W Iberia
4 N Iberia
5 W France
6 British Isles
7 S Norway
8 N Norway
1 California
2 N California-OR-WA
3 British Columbia
4 S Gulf of Alaska
5 N Gulf of Alaska
Fig. 1 Target regions used for AR-detection and tracking for the case of ERA-20C. Also shown
is the corresponding orography. The detection “barriers” used for the 3 remaining reanalyses
are in the direct vicinity of those shown here.
20th-Century Atmospheric Rivers 23
Step A
Does IVT exceed the detec-
tion percentile (Pd) at any of
the grid-boxes defining the
target region?
Move to the next
point in time and
proceed to step A
Yes
Yes
No
No
Select as new targeted
grid-box e the neighbor
exceeding Pt with largest
IVT.
Move to the targeted grid-box
(e) where IVT is largest.
Step C
Is the detected AR-track > 3000km?
Step D
The target region is said to be affec-
ted by an AR at this point in time
Step B
Calculate the direction of IVT at e. All 8 cardinal directions are
accounted for. E. g. if IVT comes from the West, then consider
the 3 neighboring grid-boxes located to the West, Southwest
and Northwest of e. All of the following three questions must be
affirmed to proceed with a “yes”:
1. Does any of the 3 neighboring grid-boxes exceed
the tracking percentile (Pt)?
2. Is the AR-track shorter than the allowed maximum?
3. Is e targeted for the first time at the this point in time?
Yes
No
Fig. 2 Schematic overview of the proposed AR detection and tracking algorithm
24 S. Brands et al.
c) ERA-20C
IVT
kg . m-1 . s-1
200
300
400
500
600
700
b) NOAA-20C a) NCEP/NCAR
Fig. 3 Illustrative example for an AR affecting southern Norway on 11 January 1971 OO UTC
for a) NCEP/NCAR, b) NOAA-20C and c) ERA-20C. Colour shadings and vector lengths are
proportional to the strength of the vertically integrated water vapour flux. The direction of
the flow is indicated by the orientation of the vectors. The cyan line represents the AR-track
found by the algorithm.
20th-Century Atmospheric Rivers 25
20 40 60 80 100
Morocco
S Iberia
W Iberia
N Iberia
W France
British Isles
S Norway
N Norway
SouthCal
NorthCal−OR−WA
British Columbia
S Gulf of Alaska
N Gulf of Alaska
NCEP/NCAR
Relative frequency (%)
20 40 60 80 100
NOAA−20C
Relative frequency (%)
20 40 60 80 100
Morocco
S Iberia
W Iberia
N Iberia
W France
British Isles
S Norway
N Norway
SouthCal
NorthCal−OR−WA
British Columbia
S Gulf of Alaska
N Gulf of Alaska
ERA−20C
Relative frequency (%)
20 40 60 80 100
ERA−Interim
Relative frequency (%)
a) b)
d)c)
OND JFM Range
Fig. 4 Fraction of ARs that are detected if the capability to track towards the north, north-
east, east or south-east is intentionally disabled (see Equation 6) for a) NCEP/NCAR, b)
NOAA-20C, c) ERA-20C and d) ERA-Interim. Results are for the October-to-December
(OND) and January-to-March (JFM) seasons, considering the time period 1979-2010. Squares
/ circles and errorbars refer to the mean and range of the 6 results obtained from the 6 con-
sidered percentile-threshold combinations listed in Table 4, i. e. refer to the method-related
uncertainty of the results.
26 S. Brands et al.
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900
AR countAR countAR countAR countAR countAR countAR count
1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
a) Mo ro cc o (23. 7 / 2 1.3 )
h) N Norw ay ( 2 8.7 / 30.1 )g ) S No r way (29.9 / 27 .1)
f) Br it ish Isles (4 2 .3 / 4 0 .1 ) e) W Fran ce ( 3 8. 5 / 3 2.6 )
d) N Ib er ia (3 0 .7 / 2 9 .2 )c) W Ib e ria (35 .5 / 31 .5)
b) S Ib e ria (27 .4 / 27 .6)
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
m ) N Gu l f of Ala ska ( 2 6.7 / 3 3 .7 )
l) S Gu lf o f Al ask a (36.0 / 40. 5) k) Br it ish Colum bia ( 41 .8 / 44. 8)
j) No rthCa l- OR-WA (34.3 / 37 .6)i) Sou t h Cal (19 .5 / 13.7 )
2) Nort h Am er ica n West Coa st
1) Europ e an At l ant ic Seaboard
year
year
year year
NOAA–20C ERA20C
Fig. 5 Year-to-year sequence of seasonal AR-occurrence counts during the OND season for
NOAA-20C (blue) and ERA-20C (red). The lines and shadings refer to the mean and range of
the percentile sample (see Table 4), i. e. refer to the method-related uncertainty of the results.
Displayed are 1900-2012 time series for NOAA-20C and 1900-2010 time series for ERA-20C.
20th-Century Atmospheric Rivers 27
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
0
50
100
150
a) Morocco (35.9 / 33.9)
h) N Norway (29.3 / 31.3)g) S Norway (28.7 / 27.7)
year
year
year
year
f) British Isles (41.9 / 40.5) e) W France (41.5 / 36.3)
d) N Iberia (35.7 / 34.9)c) W Iberia (47.6 / 42.2)
b) S Iberia (38.7 / 40.3)
m) N Gulf of Alaska (24.9 / 30.0)
l) S Gulf of Alaska (32.9 / 35.3) k) British Columbia (36.0 / 39.6)
j) NorthCal-OR-WA (35.7 / 39.7)i) SouthCal (23.0 / 15.6)
2) Nort h Am er ica n West Coa st
1) Europ e an At l ant ic Seaboard
NOAA–20C
ERA20C
AR countAR countAR countAR countAR countAR countAR count
Fig. 6 As Figure reff.f04, but for the JFM-season.
28 S. Brands et al.
a) Morocco (8.2 / 5.3)
h) N Norway (9.8 / 8.2)g) S Norway (13.7 / 9.7)
f) British Isles (15.5 / 11.5) e) W France (9.8 / 8.6)
d) N Iberia (9.4 / 6.8)c) W Iberia (10.7 / 5.1)
b) S Iberia (9.1 / 5.1)
m) N Gulf of Alaska (28.9 / 13.9)
l) S Gulf of Alaska (18.3 / 10.0) k) British Columbia (9.7 / 10.0)
j) NorthCal-OR-WA (9.2 / 7.4)
i) SouthCal (7.6 / 6.9)
bias (%)bias (%)bias (%)bias (%)bias (%)bias (%)bias (%)
2) North American West Coast
1) European Atlantic Seaboard
1920 1930 1940 1950 1960 1970 1980 1990
40
20
0
20
1920 1930 1940 1950 1960 1970 1980 1990
40
20
0
20
1920 1930 1940 1950 1960 1970 1980 1990
40
20
0
20
1920 1930 1940 1950 1960 1970 1980 1990
40
20
0
20
1920 1930 1940 1950 1960 1970 1980 1990
40
20
0
20
1920 1930 1940 1950 1960 1970 1980 1990
40
20
0
20
1920 1930 1940 1950 1960 1970 1980 1990
40
20
0
20
1920 1930 1940 1950 1960 1970 1980 1990
40
20
0
20
1920 1930 1940 1950 1960 1970 1980 1990
50
0
50
100
1920 1930 1940 1950 1960 1970 1980 1990
50
0
50
100
1920 1930 1940 1950 1960 1970 1980 1990
50
0
50
100
1920 1930 1940 1950 1960 1970 1980 1990
50
0
50
100
1920 1930 1940 1950 1960 1970 1980 1990
50
0
50
100 year
year
OND JFM
Fig. 7 Relative difference in the climatological mean AR-occurrence counts (NOAA-20C mi-
nus ERA-20C with respect to NOAA-20, in %, see Equation 4) along the course of the 20th
century, obtained by applying a 31-year running window starting in 1900-1931 and ending in
1900-2010. Along the x-axis of each panel, the centre year of each sub-period is displayed.
Lines and shadings refer to the mean and range of the percentile sample (see Table 4), i. e.
refer to the method-related uncertainty of the results; blue = OND season, red = JFM season.
To measure the stationarity of the the bias, the standard deviation (std) of the 81 mean bias
values (as depicted by the lines) is displayed in the header of each panel. The first number
refers to std for OND, the second to std for JFM. Note the distinct scale of the y-axes for
Europe/North Africa and western North America.
20th-Century Atmospheric Rivers 29
a) Morocco (0.09 / 0.03)
h) N Norway (0.06 / 0.09)g) S Norway (0.05 / 0.05)
f) British Isles (0.06 / 0.02) e) W France (0.05 / 0.03)
d) N Iberia (0.05 / 0.04)c) W Iberia (0.09 / 0.05)
b) S Iberia (0.04 / 0.04)
m) N Gulf of Alaska (0.23 / 0.30)
l) S Gulf of Alaska (0.19 / 0.22) k) British Columbia (0.15 / 0.16)
j) NorthCal-OR-WA (0.16 / 0.16)i) SouthCal (0.11 / 0.20)
rsrsrsrsrsrsrs
2) North American West Coast
1) European Atlantic Seaboard
1920 1930 1940 1950 1960 1970 1980 1990
0.6
0.7
0.8
0.9
1
1920 1930 1940 1950 1960 1970 1980 1990
0.6
0.7
0.8
0.9
1
1920 1930 1940 1950 1960 1970 1980 1990
0.6
0.7
0.8
0.9
1
1920 1930 1940 1950 1960 1970 1980 1990
0.6
0.7
0.8
0.9
1
1920 1930 1940 1950 1960 1970 1980 1990
0.6
0.7
0.8
0.9
1
1920 1930 1940 1950 1960 1970 1980 1990
0.6
0.7
0.8
0.9
1
1920 1930 1940 1950 1960 1970 1980 1990
0.6
0.7
0.8
0.9
1
1920 1930 1940 1950 1960 1970 1980 1990
0.6
0.7
0.8
0.9
1
1920 1930 1940 1950 1960 1970 1980 1990
0
0.2
0.4
0.6
0.8
1
1920 1930 1940 1950 1960 1970 1980 1990
0
0.2
0.4
0.6
0.8
1
1920 1930 1940 1950 1960 1970 1980 1990
0
0.2
0.4
0.6
0.8
1
1920 1930 1940 1950 1960 1970 1980 1990
0
0.2
0.4
0.6
0.8
1
1920 1930 1940 1950 1960 1970 1980 1990
0
0.2
0.4
0.6
0.8
1
year
year
OND JFM
Fig. 8 As Figure 8 but for the rank correlation coefficient (rs) between the seasonal AR-
occurrence counts from NOAA-20C and ERA-20C. Dashed horizontal lines mark the critical
values below / above which rs is significant at a test-level of 5%. Note the distinct scale of the
y-axes for Europe/North Africa and western North America.
30 S. Brands et al.
a) Morocco (0.12 / 0.10)
h) N Norway (0.11 / 0.09)g) S Norway (0.12 / 0.12)
f) British Isles (0.11 / 0.08) e) W France (0.11 / 0.11)
d) N Iberia (0.12 / 0.08)c) W Iberia (0.12 / 0.11)
b) S Iberia (0.10 / 0.15)
m) N Gulf of Alaska (0.11 / 0.09)
l) S Gulf of Alaska (0.11 / 0.12) k) British Columbia (0.18 / 0.17)
j) NorthCal-OR-WA (0.07 / 0.13)i) SouthCal (0.13 / 0.13)
1) AR link to NAO
2) AR link to Aleutian Low
1920 1930 1940 1950 1960 1970 1980 1990
-0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
year
year
NOAA–20C
ERA-20C
rs rs rs rs rs rs rs
Fig. 9 As Figure 8 but for the rank correlation coefficient (rs) between the OND AR-
occurrence counts in NOAA-20C (blue) or ERA-20C (read) and the station-based NAO or
North Pacific index. Dashed horizontal lines mark the critical values below / above which rs
is significant at a test-level of 5%.
20th-Century Atmospheric Rivers 31
a) Morocco (0.06 / 0.06)
h) N Norway (0.07 / 0.07)g) S Norway (0.08 / 0.07)
f) British Isles (0.11 / 0.09) e) W France (0.16 / 0.18)
d) N Iberia (0.15 / 0.16)c) W Iberia (0.09 / 0.11)
b) S Iberia (0.06 / 0.09)
m) N Gulf of Alaska (0.12 / 0.15)
l) S Gulf of Alaska (0.15 / 0.16) k) British Columbia (0.15 / 0.17)
j) NorthCal-OR-WA (0.17 / 0.16)i) SouthCal (0.13 / 0.13)
1) AR link to NAO
2) AR link to Aleutian Low
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
1920 1930 1940 1950 1960 1970 1980 1990
0.5
0
0.5
year
year
NOAA–20C ERA-20C
rs rs rs rs rs rs rs
Fig. 10 As Figure 9, but for the JFM-season.
32 S. Brands et al.
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
H−NAO CPC−NAO EA SCAND EA/WR POL
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
rs
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
rs
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
rs
H−NAO CPC−NAO EA SCAND EA/WR POL
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
rs
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
H−NAO CPC−NAO EA SCAND EA/WR POL
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
OND
(a) (b) (c)
(d) (e) (f)
(g) (h) (i)
(j) (k) (l)
NCEP/NCAR
(1950-2010)
NOAA-20C
(1950-2010)
ERA-20C
(1950-2010)
ERA-Interim
(1979-2013)
JFM ONDJFM
Fig. 11 Rank correlation coefficient (rs) between seasonal AR-occurrence counts and
seasonal-mean atmospheric circulation indices for the 8 considered target regions in Eu-
rope/North Africa ordered from the South to the North, i.e. the first bar or each group of bars
refers to Morocco and the last to northern Norway respectively. Results are for NCEP/NCAR,
NOAA-20C, ERA-20C and ERA-Interim (each row corresponds to a dataset) and for OND,
JFM, ONDJFM and ONDFJM considering the persistence criterion (each column corresponds
to a season-definition). Bars and errorbars refer to the mean and range of the 6 results obtained
from the 6 considered percentile-threshold combinations, i. e. refer to the method-related un-
certainty of the results. Dashed horizontal lines mark the critical values below / above which r
is significant at a test-level of 5%. Results are for 1950-2010 except for ERA-Interim in which
case they are for 1979-2013.
20th-Century Atmospheric Rivers 33
rsrsrs
H−NAO CPC−NAO EA SCAND EA/WR POL
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
rs
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
NCEP/NCAR
(1950-2010)
NOAA-20C
(1950-2010)
ERA-20C
(1950-2010)
ERA-Interim
(1979-2013)
ONDJFM (persistence) ONDJFM (no eastward) ONDJFM (2000 km)
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
H−NAO CPC−NAO EA SCAND EA/WR POL
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
H−NAO CPC−NAO EA SCAND EA/WR POL
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
(a) (b) (c)
(d) (e) (f)
(g) (h) (i)
(j) (k) (l)
Fig. 12 As Figure 11, but for AR-counts (first column) calculated upon persistent events only,
(second column) obtained without eastward tracking capability and (third column) obtained
with a length criterion of >2000 km.
34 S. Brands et al.
OND
NCEP/NCAR
(1950-2010)
NOAA-20C
(1950-2010)
ERA-20C
(1950-2010)
ERA-Interim
(1979-2013)
JFM ONDJFM
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
rs
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
rs
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
rs
PNA NP WP
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
rs
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
PNA NP WP
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
PNA NP WP
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
(a)
(d)
(g)
(j)
(b)
(e)
(h)
(k)
(c)
(f)
(i)
(l)
Fig. 13 As Figure 11, but for rs between seasonal AR-occurrence counts in the 5 regions
along the west coast of North America and the circulation coefficients relevant there, PNA =
Pacific-North American, NP = North Pacific, WP = West Pacific. The first bar or each group
of bars refers to southern California, the last one to the northern Gulf of Alaska.
20th-Century Atmospheric Rivers 35
NCEP/NCAR
(1950-2010)
NOAA-20C
(1950-2010)
ERA-20C
(1950-2010)
ERA-Interim
(1979-2013)
ONDJFM (no eastward)ONDJFM (persistence) ONDJFM (2000 km)
rs rs
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
rs
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
rs
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
PNA NP WP
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
PNA NP WP
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
PNA NP WP
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
(a)
(d)
(g)
(j)
(b)
(e)
(h)
(k)
(c)
(f)
(i)
(l)
Fig. 14 As Figure 13, but for AR-counts (first column) calculated upon persistent events only,
(second column) obtained without eastward tracking capability and (third column) obtained
with a length criterion of >2000 km.
... We use a method similar to Brands et al. (2017) who identified ARs as lines following the maximum IVT values on 6-hourly fields starting from a specific region (Outten, 2019). The first step consists in computing the IVT percentiles for each grid point and calendar month only considering the period 1979-2009 as in Brands et al. (2017). ...
... We use a method similar to Brands et al. (2017) who identified ARs as lines following the maximum IVT values on 6-hourly fields starting from a specific region (Outten, 2019). The first step consists in computing the IVT percentiles for each grid point and calendar month only considering the period 1979-2009 as in Brands et al. (2017). The second step is to create a line along the region where we want to identify landfalling ARs, here the Norwegian coast (see red line in Fig. 3). ...
... The second step is to create a line along the region where we want to identify landfalling ARs, here the Norwegian coast (see red line in Fig. 3). Contrary to Brands et al. (2017) who divided Norway in two regions (lines 6 and 7 in their Fig. 1), we suppose that only one AR reaches Norway at every time step. ...
Article
Full-text available
Extreme precipitation events in Norway in all seasons are often linked to atmospheric rivers (AR). We show that during the period 1979–2018 78.5% of the daily extreme precipitation events in Southwestern Norway are linked to ARs, this percentage decreasing to 59% in the more northern coastal regions and ∼40% in the inland regions. The association of extreme precipitation with AR occurs most often in fall for the coastal areas and in summer inland. All Norwegian regions experience stronger winds and 1–2 °C increase of the temperature at 850 hPa during AR events compared to the climatology, the extreme precipitation largely contributing to the wet climatology (only considering rainy days) in Norway but also in Denmark and Sweden when the rest of Europe is dry. A cyclone is found nearby the AR landfall point in 70% of the cases. When the cyclone is located over the British Isles, as it is typically the case when ARs reach Southeastern Norway, it is associated with cyclonic Rossby wave breaking whereas when the ARs reach more northern regions, anticyclonic wave breaking occurs over Northern Europe. Cyclone-centered composites show that the mean sea level pressure is not significantly different between the eight Norwegian regions, that baroclinic interaction can still take place although the cyclone is close to its decay phase and that the maximum precipitation occurs ahead of the AR. Lagrangian air parcel tracking shows that moisture uptake mainly occurs over the North Atlantic for the coastal regions with an additional source over Europe for the more eastern and inland regions.
... Curry et al., 2019;Gao et al., 2015;Hagos et al., 2016;Huang et al., 2020;Mundhenk et al., 2018;Payne and Magnusdottir, 2015;Payne et al., 2020;Shields and Kiehl, 2016;Radić et al., 2015;Rivera and Dominguez, 2016;Swain et al., 2015;Warner et al., 2015) and Europe (e.g. Gao et al., 2016;Lavers et al., 2013;Brands et al., 2017;Whan et al., 2020). In particular, the performance of the state-of-the-art high-resolution GCMs for simulating ARs over East Asia and the sensitivity of such performance to the changes in model resolution has not yet been thoroughly investigated. ...
... The study of Lavers et al. (2020) has shown that the use of IVT between 925 and 700-hPa is sufficient to diagnose the moisture transport by ARs in models Reconstructed time series of the total solar irradiance and solar spectral irradiance Matthes et al. (2017) suffering such an output availability issue. Noted that this study uses a fixed (absolute) threshold of 500 kg m -1 s -1 to isolate AR features following the studies of Leung and Qian (2009), Mahoney et al. (2016), Whan and Zwiers (2016), Brands et al. (2017) and Ralph et al. (2019). Quantile-based varying (relative) thresholds are not used in this study as this method will rescale the thresholds respectively for the reanalysis data and different model simulations. ...
Article
Full-text available
This study evaluates the performance of the Met Office Hadley Centre Global Environment Model 3–Global Coupled version 3.1 (HadGEM3-GC3.1) in simulating the climatology of atmospheric rivers (ARs) over East Asia during the period 1951–2014. Compared to a high-resolution climate reanalysis dataset and three different precipitation observation datasets, better performances in simulating the characteristics of ARs, including the frequency of occurrences, intensity of moisture transport, meridional heat transport and contributions to precipitation, are demonstrated by the simulation at the N512 (0.35◦ × 0.23◦, ~25-km) horizontal resolution compared to those at the N96 (1.875◦ × 1.25◦, ~135-km) and N216 (0.83◦ × 0.56◦, ~65-km) resolutions. Also, the N512 experiment is more capable of representing the significant increase in AR frequency for the years with preceding El Niño events. However, the N512 experiment still presents an underestimation of AR frequency in different seasons. In contrast to the N96 and N216 simulations, it overestimates the AR-associated extreme precipitation amount over most of the study region, which is partly explained by the more precipitation favorable thermodynamic conditions along the simulated AR axes. Overall, a sufficiently high horizontal resolution is crucial for the GCM to realistically represent the observed characteristics of ARs over East Asia and to robustly project their potential changes in a warming climate, although the systematic errors still affect the realism of the simulated ARs.
... An AR is considered to be a landfalling AR if a region of enhanced IVT intersects a grid cell occupied by the Antarctic Peninsula. The object axis is then calculated following 71 . Axis calculation can be described as follows: ...
Article
Full-text available
Northern sections of the Larsen Ice Shelf, eastern Antarctic Peninsula (AP) have experienced dramatic break-up and collapse since the early 1990s due to strong summertime surface melt, linked to strengthened circumpolar westerly winds. Here we show that extreme summertime surface melt and record-high temperature events over the eastern AP and Larsen C Ice Shelf are triggered by deep convection in the central tropical Pacific (CPAC), which produces an elongated cyclonic anomaly across the South Pacific coupled with a strong high pressure anomaly over Drake Passage. Together these atmospheric circulation anomalies transport very warm and moist air to the southwest AP, often in the form of “atmospheric rivers”, producing strong foehn warming and surface melt on the eastern AP and Larsen C Ice Shelf. Therefore, variability in CPAC convection, in addition to the circumpolar westerlies, is a key driver of AP surface mass balance and the occurrence of extreme high temperatures. Landfalling atmospheric rivers on the Antarctic Peninsula, which lead to strong surface melting that can cause ice shelf collapse, have been linked to localized deep convection in the central tropical Pacific northeast of Fiji.
... The importance of IVT in extreme precipitation events and floods has been analyzed in detail for the west coast of the USA [36,39,40]. Similar conclusions have also been reached for Europe [41][42][43][44]. ...
Article
Full-text available
In this work, the high-resolution spatial and temporal variability of precipitable water (PW) over Poland is presented. PW is one of the key parameters of the atmosphere taken into account in thermodynamic and radiation models. The daily PW values from years 2001–2010, calculated with the use of the WRF model, were compared with PW from soundings. The WRF modeled PW is in close agreement with measurements for the whole column of the troposphere and for individual levels: below 1.5 km, 1.5–3 km, 3–6 km and 6–10 km. The best agreement is observed in the lower part of the troposphere, especially for winter months. At the levels of 1.5 km to 10 km, the WRF model overestimates the PW values throughout the year, whereas up to 1.5 km PW is underestimated. The study shows an increasing trend of PW annual values between 1983 and 2010, but the trend is statistically insignificant. A significant positive trend with a high Sen’s slope is observed for the summer season up to 3 km in the troposphere, along with a significant negative tendency for spring. The trends in PW over Poland and Central Europe identified in this study contribute to the ongoing discussion on the observed climate changes.
... The algorithm first detects regions of high moisture transport using six-hourly fields of vertically integrated water vapor transport (IVT) between 100 and 1000-hPa. To do this, we use an IVT threshold of 500 kg m −1 s −1 following the studies of Leung and Qian (2009), Whan and Zwiers (2016), Mahoney et al. (2016), Brands et al. (2017) and Ralph et al. (2019). This is to obtain AR plumes with relatively intense IVT, which would be closely associated with heavy precipitation (Mahoney et al. 2016). ...
Article
Full-text available
Atmospheric rivers (ARs) play an important role in the climate of East Asia due to their close linkage to precipitation extremes. In this study, long-term trends in ARs over East Asia for the period 1951–2015 are investigated using long-term records of historical climate, including the ERA5 climate reanalysis and the APHRODITE precipitation dataset. These datasets are produced at a relatively high spatial resolution of 0.25° × 0.25°, which allows for evaluation of the long-term trends in the fine-scale characteristics of ARs. The results indicate a significant decreasing trend in ARs and the associated precipitation over the north of East Asia. These dynamical changes dominate the decreasing trend of total summer precipitation amounts in parts of northern China. The decreasing trend in ARs over the north is principally related to the intensification and southward displacement of the southwesterly monsoon flow in boreal summer. In contrast, increasing AR activity and the associated precipitation and heavy rain events over the south of East Asia are observed. These changes are associated with a warmer and more humid environment along AR axes, as well as the southward shift of ARs driven by the dynamical responses of the large-scale environments in the context of climate warming. These responses include the intensification of the upper-level westerly jet accompanied with the strengthening of the South Asian Anticyclone during summer season. Moreover, in contrast to the general decreasing trends in boreal summer, AR activity during boreal winter-spring exhibits significant increasing trends, implying a potential weakening of the seasonality of ARs. This study shows that ARs are important synoptic mechanisms within observed precipitation trends over East Asia, such that understanding their response to a warming climate is a prerequisite to characterizing the nature of future precipitation changes in this region.
... For ARs striking the western coast of North America, they have been either differentiated based on patterns of synopticscale atmospheric circulation or phases of climate modes (e.g., Payne and Magnusdottir 2014;Guirguis et al. 2019;Kim et al. 2019), or classified into several types based on their characteristics [e.g., integrated water vapor transport (IVT) field, shape of ARs, origins, and inland penetration] and then linked to the variations in large-scale atmospheric circulation and sea surface temperature (SST) (e.g., Rutz et al. 2015;Ryoo et al. 2015;Zhang and Villarini 2018;Zhou and Kim 2019;Tan et al. 2020). For ARs in the North Atlantic (NA), their preferred landfall locations at Europe have been shown to be modulated by various modes of climate variability [e.g., the North Atlantic Oscillation (NAO), East Atlantic pattern (EA), and Scandinavian pattern (SCAND)] (Lavers et al. 2012;Lavers and Villarini 2013b;Eiras-Barca et al. 2016;Brands et al. 2017;Pasquier et al. 2019). ...
Article
Full-text available
Using reanalysis and high-resolution ensemble simulations, we characterize cold season (December−March) North Atlantic (NA) atmospheric river (AR) tracks by grouping them into four distinct clusters; then for each cluster, we link the year-to-year variations in track count to large-scale climate variability and examine the climatological effects of the cluster on extreme precipitation and winds. The four clusters share similar prevailing AR track orientation, but differ in AR genesis locations and dominate over different regions. Cluster 1, with the longest average track of the four clusters, originates near the U.S. East Coast during La Niña and positive North Atlantic Oscillation (NAO) years and produces extreme precipitation and winds primarily over the eastern coast of North America. Cluster 2, which is weak in intensity and short-lived, forms north of 30°N of the open ocean during positive NAO years and contributes to more than 25% of the precipitation and wind extremes along the coasts of Northwestern Europe. Cluster 3, with the strongest intensity and longest duration among the four clusters, is generated surrounding the Gulf of Mexico during El Niño and negative NAO years and produces respectively more than 50% and 40% of the extreme precipitation and wind events over the eastern U.S. Cluster 4, the smallest and weakest among the four clusters, is favored under negative NAO conditions and generates roughly 25% of the extreme precipitation and winds along the coast of the Iberian Peninsula. The similarities and discrepancies between reanalysis and model simulations and among different member simulations are also discussed.
... One of the relative methods used in both latitude-dependent and -independent thresholds is the percentile-based method. In latitude-dependent thresholds, the percentile applies at each detection point (e.g., Brands et al., 2017), while in latitude-independent thresholds, a climatological value related to the specific area is used, similar to what was presented by Lavers et al. (2012) and Lavers and Villarini (2013). They proposed the 85th percentile of IVT as the minimum requirement threshold to determine ARs. ...
Article
Most water vapor in various subtropical regions is transported by atmospheric rivers (ARs). It is crucial to study this phenomenon, especially in regions more vulnerable to climate change due to population growth and aridity. The study's primary purpose was to detect ARs in the Middle East. According to expectations, ARs were classified based on strength and temporal-spatial distribution and their changes were examined to characterize the risks of the phenomenon. The results showed that ARs had significantly increased over the past 50 years (1971-2020). The most positive increase was related to weak ARs (Cat 1) in the last two decades and exceptional events (Cat 5) in the last decade. The results also indicated the role of terrain in limiting the path of the AR axis and multiple sources of moisture in their strongest form. The occurrence of hazardous events (Cat 4, 5) was estimated at up to 22% of all ARs in the Middle East, with the highest frequency in the last three decades. It was also found that events with a maximum IVT of >1000 kg m–1 s–1 occurred ten times during the last half of the last century. Also, an AR with a maximum IVT of 1102 kg m–1 s–1 occurred once in the region. The temporal distribution of these extreme events showed that they could occur at any time, depending on the conditions of AR formation. However, extreme events mainly occurred in March.
... Therefore, they have influenced human society substantially by causing extreme rainfalls and consequent floods after penetrating inland. In the last decade, many studies have examined the role of ARs in the hydro-climatologic field including rainfall, snow accumulation, and droughts over United States (Dettinger, 2013;Lamjiri et al., 2017;Lavers and Villarini, 2013a;Nayak et al., 2016;Zechiel and Chiao, 2021), South America (Viale et al., 2018), Europe (Brands et al., 2017;Ionita et al., 2020;Lavers and Villarini, 2013b;Ridder et al., 2018), and Africa Lorente-Plazas et al., 2020;Ramos et al., 2018). For example, Lamjiri et al. (2017) reported that ARs are associated with a major fraction (more than 60%) of extreme storms, with precipitationtotal return periods longer than 2 years, over the U.S. west coast. ...
Article
Atmospheric rivers (ARs) have been recognized as the main mechanism of water supply to many coastal regions across the globe, and large fractions of extreme precipitation events have been attributed to them in coastal landforms. Though more than 70% boundary of South Korea is coastal, the research literature on ARs and their impacts on precipitation is scarce, if not absent. In this study, we aim to explore the AR activity over South Korea and estimate their seasonal and spatial impact on the precipitation and its extremes over the region. To this end, we use gridded reanalysis data for identifying ARs and a recently developed high-resolution precipitation dataset for South Korea. We find that the frequency of ARs varies significantly across the seasons, while appearing as an important source of annual precipitation into South Korea. Further, we conclude that more than half of precipitation extremes may be attributed to ARs, and most of the top-ten extreme precipitation events witnessed by South Korea happen concurrently with ARs, especially in winter and spring seasons. Lastly, our analysis indicates that the recent changes in inter-annual variations of winter precipitation may be attributed to the strengthening of AR contributions, which suggests that more research on ARs and their impacts in a warmer world is critical for the region. Overall, our findings demonstrate that ARs are one of the main mechanisms of oceanic moisture inflow into South Korea, which will likely have important implications to water management in the country, through targeted forecasting of ARs and by utilizing the information in climate mechanism-based water planning and management.
... The Madden-Julian Oscillation is also found to impact ARs along the US West Coast Guan et al. 2012). In the NA, AR counts over various European countries have been linked to the North Atlantic Oscillation (NAO), Scandinavian pattern (SCP), and East Atlantic (EA) pattern (Lavers et al. 2012;Brands et al. 2017). The NAO favors AR landfall in the southern Europe during its negative phase (Eiras- Barca et al. 2016), whereas a positive NAO is conducive to AR landfall at the northern Europe (Lavers and Villarini 2013b). ...
Article
Full-text available
The sea surface temperature (SST)-forced and internal variability in cold-season (December–March) atmospheric river (AR) occurrence frequency during 1951–2010 over the North Atlantic (NA) basin are examined using a 30-member ensemble of high-resolution atmospheric simulations. The first leading mode of the forced variability features a north–south wobbling pattern modulated by an out-of-phase combination of El Niño-Southern Oscillation (ENSO) and the North Atlantic Oscillation (NAO). Co-existing El Niño and negative NAO act to shift ARs equatorward, whereas concurrent La Niña and positive NAO tend to displace ARs poleward. The second leading mode is characterized by a meridional concentration and dispersion of AR occurrence at a basin scale and can be linked to the Scandinavian pattern and the SST difference between the central and easternmost tropical Pacific. The third leading mode is dominated by an oscillation of AR occurrence north and south of 40°N in the eastern NA basin, and modulated by an in-phase combination of ENSO and the NAO. Its time series exhibits a significant upward trend, which can be linked to the SST warming in the Indo-western Pacific since the 1970s. The internal variability in cold-season NA AR occurrence frequency is then quantified by means of the signal-to-noise ratio. The calculations show that the internal variability is relatively weak over the Great Antilles and central-to-eastern US but extremely strong over Northwestern Europe, which can be attributed to the strong SST control associated with ENSO and the chaotic variations of the NAO, respectively.
Article
Full-text available
Plain Language Summary ARs drive hydrological hazards over land related to extreme precipitation. As they make landfall, ARs typically bring heavy rains, strong winds, and low pressures to the coast. While these factors can cause storm surge and coastal flooding, little attention has been paid to possible coastal impacts of ARs. We establish relationships between ARs and HTFs on the US West Coast and identify the factors causing storm surge during ARs. HTFs occur at nearly the same time that ARs make landfall more often than expected from chance, suggesting that ARs contribute importantly to HTFs. Even so, few ARs lead to HTFs—favorable tides or mean sea‐level anomalies are usually needed on top of the storm surge from an AR to cause a HTF. Storm surge during an AR can be explained by the heavy rain, strong wind, and low pressure typically associated with the event; wind and pressure are the primary factors causing the surge, but rainfall can also have a secondary influence. Our results highlight that HTFs arise from the subtle interweaving of storm surge, tide, and mean sea‐level effects, thus providing important information to managers and modelers, and motivating future studies on relationships between ARs and coastal hazards globally.
Article
Full-text available
Atmospheric rivers (ARs) are narrow regions responsible for the majority of the poleward water vapour transport across the midlatitudes. They are characterized by high water vapour content and strong low level winds, and form a part of the broader warm conveyor belt of extratropical cyclones. Although the meridional water vapour transport within ARs is critical for water resources, ARs can also cause disastrous floods especially when encountering mountainous terrain. They were labelled as atmospheric rivers in the 1990s, and have since become a well-studied feature of the midlatitude climate. We briefly review the conceptual model, the methods used to identify them, their main climatological characteristics, their impacts, the predictive ability of numerical weather prediction models, their relationship with large-scale ocean-atmosphere dynamics, possible changes under future climates, and some future challenges.
Article
Full-text available
Extreme rainfall events continue to be one of the largest natural hazards in the UK. In winter, heavy precipitation and floods have been linked with intense moisture transport events associated with atmospheric rivers (ARs), yet no large-scale atmospheric precursors have been linked to summer flooding in the UK. This study investigates the link between ARs and extreme rainfall from two perspectives: (1) Given an extreme rainfall event, is there an associated AR? (2) Given an AR, is there an associated extreme rainfall event? We identify extreme rainfall events using the UK Met Office daily rain gauge data set and link these to ARs using two different horizontal resolution atmospheric data sets (ERA-Interim and Twentieth Century Reanalysis). The results show that less than 35% of winter ARs and less than 15% of summer ARs are associated with an extreme rainfall event. Consistent with previous studies, at least 50% of extreme winter rainfall events are associated with an AR. However, less than 20% of the identified summer extreme rainfall events are associated with an AR. The dependence of the water vapor transport intensity threshold used to define an AR on the years included in the study, and on the length of the season, is also examined. Including a longer period (1900–2012) compared to previous studies (1979–2005) reduces the water vapor transport intensity threshold used to define an AR.
Article
Full-text available
We show that filamentous Atmospheric Rivers (ARs) over the Northern Atlantic Ocean are closely linked to attracting Lagrangian Coherent Structures (LCSs) in the large scale wind field. The detected LCSs represent lines of attraction in the evolving flow with a significant impact on all passive tracers. Using Finite-Time Lyapunov Exponents, we extract LCSs from a two-dimensional flow derived from water vapor flux of atmospheric reanalysis data and compare them to the threedimensional LCS obtained from the wind flow. We correlate the typical filamentous water vapor patterns of ARs with LCSs and find that LCSs bound the filaments on the back side. Passive advective transport of water vapor in the AR from tropical latitudes is potentially possible.
Article
Full-text available
The vertically integrated water vapor transport (IVT) over the Pacific-North American sector during three phases of ENSO in boreal winter (December-February) is investigated using IVT values calculated from the Climate Forecast System Reanalysis (CFSR) during 1979-2010. The shift of the location and sign of sea surface temperature (SST) anomalies in the tropical Pacific Ocean leads to different atmospheric responses and thereby changes the seasonal mean moisture transport into North America. During eastern Pacific El Niño (EPEN) events, large positive IVT anomalies extend northeastward from the subtropical Pacific into the northwestern United States following the anomalous cyclonic flow around a deeper Aleutian low, while a southward shift of the cyclonic circulation during central Pacific El Niño (CPEN) events induces the transport of moisture into the southwestern United States. In addition, moisture from the eastern tropical Pacific is transported from the deep tropical eastern Pacific into Mexico and the southwestern United States during CPEN. During La Niña (NINA), the seasonal mean IVT anomaly is opposite to that of two El Niño phases. Analyses of 6-hourly IVT anomalies indicate that there is strong moisture transport from the North Pacific into the northwestern and southwestern United States during EPEN and CPEN, respectively. The IVT is maximized on the southeastern side of a low located over the eastern North Pacific, where the low is weaker but located farther south and closer to shore during CPEN than during EPEN. Moisture enters the southwestern United States from the eastern tropical Pacific during NINA via anticyclonic circulation associated with a ridge over the southern United States.
Article
Full-text available
An automated atmospheric rivers (ARs) detection algorithm is used for the North Atlantic Ocean basin that allows the identification and a comprehensive characterization of the major AR events that affected the Iberian Peninsula over the 1948-2012 period. The extreme precipitation days in the Iberian Peninsula and their association (or not) with the occurrence of ARs is analyzed in detail. The extreme precipitation days are ranked by their magnitude and are obtained after considering 1) the area affected and 2) the precipitation intensity. Different rankings are presented for the entire Iberian Peninsula, for Portugal, and for the six largest Iberian river basins (Minho, Duero, Tagus, Guadiana, Guadalquivir, and Ebro) covering the 1950-2008 period. Results show that the association between ARs and extreme precipitation days in the western domains (Portugal, Minho, Tagus, and Duero) is noteworthy, while for the eastern and southern basins (Ebro, Guadiana, and Guadalquivir) the impact of ARs is reduced. In addition, the contribution from ARs toward the extreme precipitation ranking list is not homogenous, playing an overwhelming role for the most extreme precipitation days but decreasing significantly with the less extreme precipitation days. Moreover, and given the narrow nature of the ARs, the location of the ARs over each subdomain is closely related to the occurrence (or not) of extreme precipitation days.
Article
In the present study, a novel atmospheric river (AR) detection scheme based on column integrated water vapor and column integrated water vapor flux is presented and applied for the Iberian Atlantic Margin (IAM) and a wider area covering the North Atlantic Ocean for the period 1979–2013. The seasonal cycle in AR frequency over the North Atlantic exhibits a relatively small amplitude, being more pronounced towards the east and south of the oceanic basin, as it is increasingly related to the seasonal cycle in storm activity and the meridional displacement of the subtropical high. In the eastern North Atlantic, downwind of the North American continent, it shows a more complex behaviour. The interannual variability of AR frequency is weak across the entire North Atlantic and it does not present consistent long-term spatio-temporal patterns. For the southern IAM, AR occurrence is slightly enhanced by the negative phase of the North Atlantic Oscillation during the previous days. Up to 80% of the anomalous precipitation events (above the 95th percentile) in the IAM are associated with ARs; these values exceeding 90% in winter, and decreasing to 75% in spring when convection not related to ARs becomes a significant precipitation mechanism. Moisture advection within ARs is thus a very important contributor to anomalous precipitation. Likewise, the strength of the associated storm systems and the characteristics of the ARs themselves are also very relevant factors. The percentage of total ARs linked to anomalous precipitation is relatively low, only reaching 20% where topographic features are favourable.
Article
Atmospheric rivers (ARs) are narrow, elongated, synoptic jets of water vapor that play important roles in the global water cycle and regional weather/hydrology. A technique is developed for objective detection of ARs on the global domain based on characteristics of the integrated water vapor transport (IVT). AR detection involves thresholding 6-hourly fields of ERA-Interim IVT based on the 85th percentile specific to each season and grid cell and a fixed lower limit of 100kgm-1s-1 and checking for the geometry requirements of length >2000km, length/width ratio >2, and other considerations indicative of AR conditions. Output of the detection includes the AR shape, axis, landfall location, and basic statistics of each detected AR. The performance of the technique is evaluated by comparison to AR detection in the western North America, Britain, and East Antarctica with three independently conducted studies using different techniques, with over ~90% agreement in AR dates. Among the parameters tested, AR detection shows the largest sensitivity to the length criterion in terms of changes in the resulting statistical distribution of AR intensity and geometry. Global distributions of key AR characteristics are examined, and the results highlight the global footprints of ARs and their potential importance on global and regional scales. Also examined are seasonal dependence of AR frequency and precipitation and their modulation by four prominent modes of large-scale climate variability. The results are in broad consistency with previous studies that focused on landfalling ARs in the west coasts of North America and Europe. Key Points: Introduces an IVT-based AR detection algorithm suitable for global studies Evaluates performance of the algorithm relative to previous studies in three regions Investigates climatology/seasonality/variability of global AR characteristics using the algorithm
Article
Orthogonally rotated principle component analysis (RPCA) of Northern Hemisphere 1-month mean 700 mb heights is used to identify and describe the seasonality and persistence of the major modes of interannual variability. The analysis is detailed and comprehensive, in that 1) a high resolution, approximately equal-area 358-point grid is used for the virtually maximum possible 35-yr period of record, 2) a positive bias in the NMC data base in the early 1950s in the subtropics is largely eliminated for the first time, and 3) homogeneous, separate analyses of each month of the year are carried out, detailing the month-to-month changes in the dominant circulation patterns. The conclusion from all considerations is that the RPCA method provides a physically meaningful, as well as statistically stable product with the simplicity of teleconnection patterns but with pattern choice and depiction superior to those of the teleconnection method. -from Authors
Article
Atmospheric rivers (ARs) are narrow corridors within the warm conveyor belt of extratropical cyclones in which the majority of the poleward water vapour transport occurs. These filamentary synoptic features are responsible for extreme precipitation and flooding in Europe and the central and western United States, and also play an essential role for water resources in these areas. Using gridded precipitation products across Europe and the continental United States and the ERA-Interim reanalysis, we investigate the fraction of precipitation from 1979 to 2012 that is related to ARs in these regions. The results are region- and month-dependent, with the largest contribution generally occurring during the winter season and being on the order of 30–50%. This is particularly true for Western Europe, the U.S. West Coast, and the central and northeastern United States. Our results suggest that ARs are important agents for water supply in Europe and the United States. We have also examined whether there have been changes over time in the fractional contribution of ARs to seasonal rainfall using zero-inflated beta regression. We find that there has been a decrease in the average AR-contribution over the Mediterranean region and over the central United States.
Article
Most extreme precipitation events that occur along the North American west coast are associated with winter atmospheric river (AR) events. Global climate models have sufficient resolution to simulate synoptic features associated with AR events, such as high values of vertically integrated water vapor transport (IVT) approaching the coast. From phase 5 of the Coupled Model Intercomparison Project (CMIP5), 10 simulations are used to identify changes in ARs impacting the west coast of North America between historical (1970-99) and end-of-century (2070-99) runs, using representative concentration pathway (RCP) 8.5. The most extreme ARs are identified in both time periods by the 99th percentile of IVT days along a north-south transect offshore of the coast. Integrated water vapor (IWV) and IVT are predicted to increase, while lower-tropospheric winds change little. Winter mean precipitation along the west coast increases by 11%-18% [from 4% to 6% (degrees C)(-1)], while precipitation on extreme IVT days increases by 15%-39% [from 5% to 19% (degrees C)(-1)]. The frequency of IVT days above the historical 99th percentile threshold increases as much as 290% by the end of this century.