1. Introduction
Since its original definition in the 1990s (Newell et al. 1992; Newell and Zhu 1994; Zhu and Newell 1998), the concept of the atmospheric river (AR) has increasingly entered the public consciousness, driven largely by the severe storms that the “rivers in the sky” bring to the west coasts of the Americas and Europe. Now, in the mid-2020s, ARs grace the front pages of major news sites and even have their own rating system (categories 1–5, like hurricanes; see Ralph et al. 2019). They have been linked to the polar jet, tracked across ocean basins, sorted into moisture- and wind-dominant and dusty varieties, and even described hundreds of years before direct detection was possible using tree-ring and precipitation data (Gonzales et al. 2020; Voss et al. 2021; Zhou et al. 2021; Maclennan et al. 2022; Zhou et al. 2022; Borkotoky et al. 2023; Simon et al. 2024).
Globally, ARs are particularly impactful because, depending on the AR definition, they produce the majority of midlatitude extreme precipitation (see Arabzadeh et al. 2020). Changes in AR characteristics will thus contribute significantly to overall changes in the water cycle, from drought risk to dangerous flooding. Global warming–induced change is hardly unique to ARs—subdaily and convective precipitation extremes may increase even more rapidly than stratiform precipitation (Morrison et al. 2019; Martel et al. 2020; Chinita et al. 2021). But because of the sheer amount of extreme precipitation produced by ARs, changes in their characteristics will be crucial in determining changes in the overall extreme precipitation profile across the extratropics. In addition, certain recent studies have shown disproportionate historical increases in AR-type extreme precipitation (Lochbihler et al. 2017; Hatsuzuka et al. 2021).
To first order, the expected AR response to climate change is one of moistening. As dictated by the Clausius–Clapeyron equation, warmer air has a higher saturation vapor pressure, allowing for higher integrated water vapor (IWV) values within ARs (which, in turn, may lead to higher integrated vapor transport (IVT), vertically integrated moisture flux convergence (VIMFC), and precipitation; see Allen and Ingram 2002; Held and Soden 2006; Morrison et al. 2019; Martel et al. 2020; Chinita et al. 2021). Since relative humidity is projected to remain somewhat constant under global warming (Douville et al. 2022)—despite some evidence of historical decreases (e.g., Willett et al. 2020)—this suggests that the extent of AR conditions will increase. Modeling studies bear this theory out. The frequency of AR conditions is expected to increase by up to 50+% globally and 100+% regionally by the end of the twenty-first century under high-end climate scenarios (Lavers et al. 2013; Gao et al. 2015, 2016; Hagos et al. 2016; Espinoza et al. 2018; Massoud et al. 2019; O’Brien et al. 2022; Shields et al. 2023). ARs will be larger, moister, and produce more extreme precipitation (Dettinger 2011; Ramos et al. 2016; McClenny et al. 2020; Payne et al. 2020; Wang et al. 2023), even over regions like the northeastern United States where ARs receive less publicity (Jong et al. 2024). In fact, AR IWV is expected to increase more quickly than the ∼7% K−1 Clausius–Clapeyron (C-C) rate relative to SST and near-surface temperatures (Warner et al. 2015; Gao et al. 2015; McClenny et al. 2020), although this may be an effect of differential upper-atmospheric warming (Payne et al. 2020).
Interestingly, wind—the other contributor to IVT, the “dynamic component” to IWV’s “thermodynamic component”—is often projected to decrease in ARs as temperature rises (Dettinger 2011; Gao et al. 2015; Ramos et al. 2016; McClenny et al. 2020; Payne et al. 2020; Wang et al. 2023). Though this effect has been hypothesized to emerge from the slowing of prevailing westerlies in association with a weakened meridional temperature gradient due to Arctic amplification (Dettinger 2011; see also Francis and Vavrus 2012, 2015), evidence for historical (Barnes 2013; Riboldi et al. 2020) and future (Barnes and Polvani 2015; Dai and Song 2020) phase speed and wind decreases is contested, and a simple relationship between the degree of warming and midlatitude wind speed is uncertain. Changes in AR conditions are further complicated by the fact that storm tracks—and, by association, AR tracks (Zhang et al. 2019)—are expected to shift poleward due to a combination of factors, including baroclinicity changes, Hadley cell expansion, ozone depletion, and radiative forcing (Shaw et al. 2016; see also Yin 2005; Tamarin and Kaspi 2017).
A range of studies have examined historical trends in atmospheric river characteristics. On the global scale, Ma et al. (2020) described a poleward shift of the Southern Hemisphere ARs in MERRA-2 and NCEP–NCAR data during 1979/80–2018, driven by a corresponding shift of the westerly jet. Li and Ding (2024) observed even larger poleward shifts globally for DJF ARs in the 1979–2022 ERA5 data. Guan et al. (2023) found global increases in 1980–2020 AR frequency—measured in events per year—for all intensity categories in MERRA-2 and ERA-5 data. Shearer et al. (2020) reported an increase in MERRA-2 AR area and a potential poleward shift between 1983 and 2016. Relatedly, Algarra et al. (2020) identified an increase in anomalous moisture uptake associated with ARs in ERA-Interim between 1980 and 2017. Regionally, studies have looked at trends in AR characteristics (Gonzales et al. 2019; Sharma and Déry 2019; Gonzales et al. 2020; Liang et al. 2023) or, more commonly, AR-related extreme precipitation (e.g., Wille et al. 2021; Ramseyer et al. 2022; Henny et al. 2023; Jong et al. 2024). However, most of these studies use a single AR detection algorithm, raising the question of whether results are robust to detection sensitivity.
In 2016, the AR Tracking Method Intercomparison Project (ARTMIP) was formed: a grassroots international collaboration that pioneered the approach of using an ensemble of distinct AR detection tools (ARDTs)—each tailored to a specific purpose—to gauge detection sensitivity and distinguish robust results from those particular to a given methodology (Shields et al. 2018; Rutz et al. 2019). Subsequently, ARTMIP tier 2 expanded the available data and conducted additional sensitivity studies (Collow et al. 2022). This ambitious project has documented large discrepancies between ARDTs both historically and in future climate projections. But while ARTMIP did report slight increases in AR frequency over the 2000–19 time frame, it primarily focuses on diagnosing inter-ARDT spread, as opposed to ARDT-specific trends in historical reanalysis data.
As we move into the mid-2020s, we now have 45 years of medium–high-resolution reanalysis data since the 1979–80 start of the satellite era, as well as the ERA5 back extension to 1940 and JRA-55 data beginning in late 1957. These data are not infallible, even within the satellite era. Because the number of observations incorporated into reanalysis datasets changes practically by the month, these datasets may be heterogeneous despite being produced by internally self-consistent models, raising the possibility that, for example, IWV changes may be underestimated within various subsets of the analysis period (due to a documented failure to keep pace with the expected moistening rate; see Bosilovich et al. 2017; Xue et al. 2019; Collow et al. 2022). Indeed, reanalysis datasets are subject to various classes of errors which might obscure the true climate signal (Thorne and Vose 2010). Nevertheless, they are already being used to examine trends in a wide range of atmospheric variables including but not limited to AR characteristics (e.g., Algarra et al. 2020; Gonzales et al. 2019; Gonzales et al. 2020; Ma et al. 2020; Shearer et al. 2020; Guan et al. 2023), and they provide skillful reproductions of the global temperature trends that underpin atmospheric moistening (Simmons et al. 2017, 2021).
This study therefore aims to provide a more comprehensive snapshot of changes in atmospheric river characteristics, evaluated across a wider range of ARDTs and across multiple reanalysis datasets. To do this, we will first run three “variants” of our own relatively permissive AR detection applied to MERRA-2, JRA-55, and ERA5 data to maximize record length and assess sensitivity to slight changes in the nature of the IVT threshold. Then, results will be retested using the seven global AR datasets provided in the ARTMIP tier 2 reanalysis catalog. This will capture emerging robust trends while also transparently identifying the points of uncertainty that may complicate such results.
2. Data and methodology
a. Reanalysis data
ARs are detected using model-derived vertically IVT as in the ARTMIP tier 2 reanalysis catalog (see Collow et al. 2022). For the novel variants, these fields are taken 6-hourly from three reanalyses: NASA’s MERRA-2 (Gelaro et al. 2017), the fifth major global reanalysis produced by the European Center for Medium-Range Weather Forecasts (ERA5; Hersbach et al. 2020), and the Japan Meteorological Agency’s JRA-55 (Kobayashi et al. 2015). Subsequent synoptic analysis uses IWV and 850-hPa wind vectors from all three reanalyses and VIMFC from ERA5 and MERRA-2. MERRA-2 IVT, IWV, and VIMFC fields are hourly averages; all others are instantaneous. For computational efficiency, all ERA5 fields are linearly interpolated to the 0.5° latitude × 0.625° longitude MERRA-2 grid prior to variant AR detection. For more details, see section SM1 in the online supplemental material.
b. AR detection methodology
-
At least 2000 km in length, where length is the longest pointwise distance within the AR.
-
A length-to-width ratio of at least 2.5, using mean width, that is, area divided by length.
-
A high latitude (poleward of 20° N/S) to low-latitude area ratio of at least 2.5.
If a region of extreme IVT meets the length requirement but fails the other measures, the IVT threshold is increased by 50 kg m−1 s−1, and each subregion is retested iteratively until either 1) the IVT region meets the geometric requirements, 2) the length falls below 2000 km, or 3) the global IVT threshold maximum reaches 600 m−1 s−1 (in which case the object is discarded). Because this method of detecting AR “cores” will attenuate the AR, a refilling step is included. This is done using the SciPy function “binary_dilation,” set to expand the AR region by one unit in the zonal and meridional directions. This function was applied 3 + 2 × (N − 1) times, where N is the number of threshold increases. Extreme IVT poleward of 45° latitude was refilled entirely. Then, ARs are relabeled individually by detecting connected components, and components comprising fewer than 100 grid points are eliminated.
This provides a crude but relatively effective way to 1) separate ARs from the tropical moisture pool while 2) identifying the tropical “roots” of ARs without artificially cutting the field off at a set latitude and 3) not sacrificing the high-latitude AR components that are typically less intense, despite using a relatively permissive absolute IVT threshold. Any future iterations of this algorithm will be focused on 1) adding tracking functionality and 2) eliminating the rare instances of disconnected ARs resulting from the refilling procedure. In this paper, the intent is primarily to supplement and extend AR data beyond the ARTMIP time frame. For more details about the advantages and disadvantages of this algorithm, see supplemental material section SM2 and Figs. S1–S3.
The three variants are distinguished by their IVT conditions as follows:
-
Semifixed: Fixed-reference climatology 90th percentile of 6-hourly IVT magnitude, bounded by 50 kg m−1 s−1 on the low end and 300 kg m−1 s−1 on the high end (see Fig. S1).
-
Fixed: 300 kg m−1 s−1; not quite as permissive as the 250 kg m−1 s−1 thresholds (Rutz et al. 2014; Gershunov et al. 2017) but much more lenient than the stricter absolute or anomaly thresholds found in various studies (e.g., Mahoney et al. 2016; Lora et al. 2017; Reid et al. 2020; Ramseyer et al. 2022).
-
Fixed poleward: Fixed-threshold ARs that have an average poleward IVT component of at least 50 kg m−1 s−1 [see Guan and Waliser (2015)].
Figure S4 shows the annual-mean AR frequency resulting from each of these nine “perturbations.” These novel variant datasets have oceanic maxima of up to 30+% AR coverage in the North Pacific, North Atlantic, and Southern Ocean, consistent with the more lenient ARTMIP tier 2 reanalysis datasets Lora_v2, ARCONNECT_v2, and ClimateNet_DL (see section 2c; Collow et al. 2022). The fixed and fixed poleward thresholds return low overland AR frequency, while the more flexible semifixed thresholds result in a relatively constant 6%–9% AR frequency over land. Over the Southern Ocean, the fixed poleward AR frequency is markedly decreased, suggesting a higher proportion of zonally oriented ARs (see Garreaud et al. 2024). Overall, differences are larger between IVT threshold conditions than between reanalyses.
c. ARTMIP tier 2 intercomparison
To increase spread within the 1980–2019 period, we use the seven global AR datasets from the ARTMIP tier 2 reanalysis catalogue (https://doi.org/10.26024/rawv-yx53; Shields et al. 2018; Rutz et al. 2019; Collow et al. 2022). In contrast to the variants, which are fundamentally based on the same algorithm, the ARTMIP datasets use drastically different detection methods (see Table 1). Reid500 (Reid et al. 2020) uses a 500 kg m−1 s−1 IVT threshold more akin to those used in East Asia to filter out the intense moist monsoon flow (e.g., Liang et al. 2023). Mundhenk_v3 (Mundhenk et al. 2016) uses a 250 kg m−1 s−1 anomalous IVT threshold with the seasonal climatology removed; Lora_v2 (Lora et al. 2017; Skinner et al. 2020) adjusts IVT thresholds based on the seasonal IWV climatology; GuanWaliser_v2 (hereafter GuanWaliser; Guan and Waliser 2015) has a fully percentile-based threshold; TempestLR (McClenny et al. 2020) uses the Laplacian of IVT; ARCONNECT_v2 (Shearer et al. 2020) distinguishes between the AR body and core, each with its own threshold; and ClimateNet_DL (Prabhat et al. 2021) uses deep learning trained on expert labeling rather than a fully automated search. As a result, ARTMIP provides a more robust sample for assessing trend uncertainty within its common 1980–2019 period, while the variants provide the ability to study longer-term variation and sensitivity to different IVT conditions. For computational efficiency, all ERA5 ARTMIP data are converted to the 0.5° latitude × 0.625° longitude MERRA-2 grid using the “nearest” interpolation scheme. Note that the Reid500 output has a continuity error at the date line (listed as an erratum in the tier 2 reanalysis catalogues). Additionally, as none of the variant or ARTMIP datasets in this study use a purely time-varying detection threshold, the results may still not encompass the full diversity of behaviors associated with ARDTs implemented to date.
The ARDTs used in this study, comprising 16 datasets in total. IVT and IWV threshold units are kilograms per meter per second and kilograms per meter square, respectively.
d. A stricter approach to statistical significance
When conducting a single significance test, a p value of 0.05 means that the probability of attaining the given result if the null hypothesis were true is less than 5%. But when running N simultaneous significance tests on an array of true null hypotheses, this 5% probability translates to 0.05 × N expected false positives—for a global grid, approximately the area of North America. This is the “multiple comparisons problem,” and it is what has led certain authors to point to atmospheric science results as overinterpreted (Ventura et al. 2004; Wilks 2006, 2016). Since atmospheric fields almost always have some spatial correlation, these spurious trends may be clustered together to give the appearance of a significant finding.
3. Trends in underlying atmospheric conditions
Changes in AR characteristics are inherently linked to changes in the global distribution of IVT and its constituents, IWV and advection [approximated here as in McClenny et al. (2020) by 850-hPa wind speed]. Figure 1 gives an overview of these variables in MERRA-2, JRA-55, and ERA5, calculated using 6-hourly data and subsequently averaged for each year. Over the 1980–2023 period, IVT increased on average by 5.5%, while IWV increased by 3.9% which translates to 4.5% K−1 using the ∼0.86 K of warming derived from the NASA GISS Surface Temperature Analysis (GISTEMP Team 2024; this warming is depicted accurately by reanalyses; see Simmons et al. 2017, 2021). As of the 2015 end date of Bosilovich et al. (2017), reanalyses severely underestimated the atmospheric moistening rate. Although recent increases have raised the trend significantly, it still fails to match the theoretical ∼7% C-C rate, likely due to the pre-1994 stalling of moistening which has been attributed to observational system changes (Allan et al. 2022; Douville et al. 2022) In contrast, 850-hPa wind speeds have not increased since the late 1990s in any reanalysis (note that 850-hPa wind, as a proxy, cannot fully characterize the dynamical contribution to IVT).
Global mean (a) IVT magnitude, (b) IWV, and (c) 850-hPa wind speed. Linear regression trend information for the 1980–2023 period is given in the legend, with Δ signifying the cumulative change and p signifying the p value, or the probability of obtaining a result at least this extreme if the null hypothesis were true.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0234.1
The 6-hourly IVT results differ significantly from those obtained by first averaging the zonal and meridional IVT components monthly or annually and then computing their magnitude (see Fig. S5; Collow et al. 2022; Fig. 1b). In that case, the time series take on a more obvious mean-shift characteristic without an underlying trend, with a sudden jump in the late 1990s. Although this property is not entirely shared by IVT in its more conventional definition, this late-1990s changepoint is evident here as well and is roughly co-occurring with an increase in MERRA-2 (McCarty et al. 2016; Gelaro et al. 2017; Collow et al. 2022), a decrease in ERA5 (Hersbach et al. 2020; Bell et al. 2021) assimilated wind observations, and an increase in JRA-55 assimilated microwave observations (Kobayashi et al. 2015). While attribution is beyond the scope of this study, it is important to keep in mind that observed changes could reflect numerous factors: changes in observational data assimilation, interdecadal variability, or an acceleration of the climate change signal.
4. Trends in AR frequency, area, and mean latitude
Figure 2 shows time series of mean (Fig. 2a) global and F2 (Fig. 2b) land AR area, expressed as a percentage of (Fig. 2a) the Earth’s surface area (ESA) and (Fig. 2b) the land surface area (LSA). A wide range of mean values is represented, in agreement with previous results of ARTMIP (Collow et al. 2022). Highly restrictive criteria such as the 500 kg m−1 s−1 threshold used in Reid500 result in lower AR coverage, while more permissive criteria such as those employed by the variants, GuanWaliser, and ClimateNet_DL (here in its higher-coverage MERRA-2 version; see Collow et al. 2022) result in much higher AR coverage. Globally, the ARTMIP tier 2 reanalysis (hereafter ARTMIP tier 2 or simply ARTMIP) datasets span about twice the range of the variants; over land, the variant spread is enhanced due to the percentile IVT threshold applied to moisture-limited regions in the semifixed variants. The fixed poleward variants are more similar to the fixed variants, suggesting that purely zonally oriented ARs are uncommon in the global mean (despite higher frequency in the Southern Ocean; see Fig. S4). Note that total AR coverage conveys little about AR occurrence (i.e., No. of simultaneous ARs) or individual AR area; ARDTs can achieve similar coverage in different ways, as with Mundhenk_v3 and ARCONNECT_v2 which produce many small ARs and fewer, larger ARs, respectively (Fig. S6).
Total AR (a) global and (b) land area for the 16 datasets used in this study. The three variants (nine datasets total) are aggregated into a single mean value (thick black line) and range (gray shading); individual values are shown for semifixed (dotted), fixed (dashed), and fixed-poleward (solid) datasets in thin black lines. The 1980–2019/23 trend information is given in the legend, with ESA, LSA, and Δ signifying Earth’s total surface area, land surface area, and the cumulative percentage change, respectively, and p signifying the p value, or the probability of obtaining a result at least this extreme if the null hypothesis were true. Note that percentage change is relative to the mean value and does not correspond to the y-axis % ESA scale.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0234.1
In general, the 1940–79 period is presented for completeness but is not considered for trend analysis in this paper due to 1) the small and changing number of available variant AR datasets and absence of ARTMIP data prior to 1980, 2) apparent inhomogeneities in the time series (as in Fig. 2a and the synoptic variable means shown in section 5), and 3) preexisting concerns about the utility of reanalysis data for long-term trend monitoring (e.g., Thorne and Vose 2010). Over the remaining 1980–2023 period, AR coverage trends are strongly positive globally (nine of nine variant and six of seven ARTMIP datasets), consistent with documented increases in global AR frequency and area (Shearer et al. 2020; Guan et al. 2023). Land increases have lower significance of only seven of nine variant and one of seven ARTMIP datasets (though no decreases were found). The average change in global (land) AR area is 7.4% (7.5%). ClimateNet_DL, which uses deep learning trained on human expert input, is an outlier with no area change over the analysis period. On the other end of the spectrum, ARCONNECT_v2 shows an anomalous 20% increase in AR land area since 1980, potentially indicating an increase in its low average coastal persistence (see Shields et al. 2023).
As with area, a wide range of mean latitude values (in degrees poleward) is represented (Fig. 3). On the low end is Reid500, with its high IVT requirement likely favoring moist low latitudes, and on the high end, Lora_v2, which defines its IVT threshold as a function of IWV (minimum of 225 kg m−1 s−1) and so is relatively permissive at high latitudes and strict at low latitudes. The variants fall between the GuanWaliser and Lora_v2 ARDTs in the poleward half of the spread. Over land, these differences are exaggerated, as permissive methods extend further north while strict methods must shift even further equatorward to find sufficiently moist land areas. Again, trends are more robust globally than over land; globally, all nine variant and three of seven ARTMIP datasets meet the FDR-adjusted significance condition for increasing mean latitude, whereas over land, six of nine variant and 0 ARTMIP datasets do so. This time, ARCONNECT_v2 is the low-end outlier. Globally, the poleward shift lies between 0.3° and 0.8° latitude whether calculated over the entire AR domain each year or averaged first over individual ARs, but the effect is larger and more robust in the Southern Hemisphere (see Fig. S7). This is consistent with the expected poleward shift of jet streams and associated storm tracks in a warming climate (Chang et al. 2012; Tamarin-Brodsky and Kaspi 2017; Chemke 2022; Zhang et al. 2024) as well as previous observational studies which suggested poleward shifts in ARs and moisture transport globally (Shearer et al. 2020; Li and Ding 2024) and over the Southern Hemisphere (Sousa et al. 2018; Ma et al. 2020) in the late twentieth and early twenty-first centuries.
AR (a) global and (b) land mean latitude (in degrees poleward) for the 16 datasets used in this study, averaged over the total AR area for each year. The three variants (nine datasets total) are aggregated into a single mean value (thick black line) and range (gray shading); individual values are shown for semifixed (dotted), fixed (dashed), and fixed-poleward (solid) datasets in thin black lines. The 1980–2019/23 trend information is given in the legend, with ESA, LSA, and Δ signifying Earth’s total surface area, land surface area, and the cumulative percentage change, respectively, and p signifying the p value, or the probability of obtaining a result at least this extreme if the null hypothesis were true.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0234.1
Regionally, AR frequency has increased robustly in the variants over the North American east coast, the circumpolar band of the Southern Ocean, high-latitude regions of the North Pacific, and several other regions throughout the Northern Hemisphere (Fig. 4). This is somewhat consistent with trends derived from the ARTMIP tier 2 sample, although significance is lower in the ARTMIP sample possibly due to 1) greater ARDT diversity and 2) the shorter time frame during the recent period which dominates observed areal expansion. Some regions of increasing AR frequency appear in ARTMIP but not the variants (e.g., tropical regions not well defined in the variants) or in the variants but not ARTMIP (e.g., the northern Atlantic). The regions that do exhibit good consensus agree with documented historical (Ma et al. 2020; Shearer et al. 2020; Ramseyer et al. 2022) and projected future (Espinoza et al. 2018; Massoud et al. 2019; Shields et al. 2023) AR frequency change. These regions also tend to occur where the AR response to radiative forcing is expected to be distinguishable from internal variability earliest (Tseng et al. 2022). Relatedly, the frequency of extreme IVT (EIVT; defined using the fixed and semifixed 300 kg m−1 s−1 thresholds) exhibits much more widespread significant changes, with 18% of Earth’s surface having experienced increases in two-thirds of available datasets as compared to 5% for variant ARs (see Fig. S8 for replication with fixed 400 and 500 kg m−1 s−1 thresholds).
(left) The number of datasets with nonzero frequency during at least 20 years for (a) fixed and semifixed 300 kg m−1 s−1 EIVT, (c) AR variants, and (e) ARTMIP tier 2 global AR datasets. (right) The percentage of these viable datasets at each grid point with FDR-controlled significant positive (red shading) and negative (blue shading) annual frequency trends. Gray stippling indicates two-thirds consensus using an uncontrolled p < 0.05 significance criterion. In the case of co-occurring positive and negative trends, the trend sign with greater consensus is displayed, except in the case of equal values which are indicated by pink shading. Trend consensus is not displayed for regions where only one dataset satisfies the completeness criterion.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0234.1
5. Trends in AR intensity: IVT, IWV, |V850|, and VIMFC
Example distributions of AR synoptic characteristics are given for the variants in Fig. 5 (for ARTMIP tier 2 synoptic means, see Figs. S9–S11). For these datasets, IVT peaks are roughly collocated with AR frequency maxima, while IWV is largest at low latitudes and 850-hPa wind speed is largest at high latitudes, the latter especially over the Southern Ocean. These opposing wind and moisture distributions combine to create the typical ∼30°–60°N/S AR IVT maxima and imply a latitude gradient of AR nature globally: more moisture based at low latitudes and more wind based at high latitudes [consistent with Zhou et al. (2022)]. VIMFC associated with ARs is elevated over mountainous and coastal regions where frictional or orographic convergence mechanisms create forcing for precipitation. Fixed-threshold datasets also have a secondary meridional VIMFC gradient indicating that AR moisture is more likely to converge once it reaches high latitudes. This field resembles the correlation between IVT and extreme precipitation (Gimeno-Sotelo and Gimeno 2023); regions with large mean VIMFC are more vulnerable to extreme precipitation during AR conditions.
AR variant mean 1980–2023 (a)–(c) IVT, (d)–(f) IWV, (g)–(i) 850-hPa wind speed, and (j)–(l) VIMFC, separated into (left) semifixed, (middle) fixed, and (right) fixed-poleward AR detection datasets. In each case, the mean is taken over the three (two, for VIMFC) corresponding AR datasets from MERRA-2, JRA-55, and ERA5 (MERRA-2 and ERA5 for VIMFC). Data are shown only for regions where at least 50% of years have AR activity.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0234.1
a. Mean AR characteristics
As with area and latitude, the ARTMIP tier 2 datasets yield a much wider range of IVT, IWV, |V850|, and especially VIMFC mean values than the variants (Figs. 6 and 7). The area ordering of ARDTs (Fig. 2) can essentially be reversed to infer the distribution of moisture transport intensity (Fig. 6c; see Shields et al. 2023), although differences begin to emerge when dealing with IWV, |V850|, and VIMFC. One exception is Lora_v2, which is the median ARTMIP dataset for area but has the lowest mean IVT values and high wind speeds, likely indicative of its higher mean latitude (see Fig. 3). Reid500 has the most extreme IVT, IWV, and wind speed, with ClimateNet_DL reliably at the bottom at or below the level of the most permissive of the variants. Over land (Fig. 7), the GuanWaliser algorithm has the mildest average conditions, on par with those seen from the semifixed variants since both algorithms allow for low IVT values. Interestingly, Lora_v2’s high land |V850| diverges sharply from that of the semifixed variants despite their own high mean latitude; this is likely due to the fact that Lora_v2 has a much higher minimum IVT (225 kg m−1 s−1) than either the semifixed variant or GuanWaliser algorithms; as such, it attenuates AR frequency over both southern (due to threshold IWV dependence) and land (due to higher minimum threshold) regions relative to certain other ARDTs (see Fig. 3; Collow et al. 2022).
Annual AR global (a) IWV, (b) 850-hPa wind speed, (c) IVT, and (d) VIMFC anomalies (relative to the post-1980 mean) for the variant mean (thick black line) and range (gray shading) and ARTMIP tier 2 (colored lines) AR datasets. The 1980–2019/23 trend information is given in the legend, with Δ signifying the cumulative percentage change relative to the mean and p signifying the p value, or the probability of obtaining a result at least this extreme if the null hypothesis were true. Post-1980 means are shown in the right-hand bar for (colors) ARTMIP and variant (thick black) mean, (thin dark gray) fixed poleward, (thin medium gray) fixed, and (thin light gray) semifixed datasets. For the variants, the number of FDR-controlled significant positive and negative trends is given in the legend. Note that VIMFC is not included for JRA-55-based datasets.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0234.1
Annual AR land (a) IWV, (b) 850-hPa wind speed, (c) IVT, and (d) VIMFC anomalies (relative to the post-1980 mean) for the variant mean (thick black line) and range (gray shading) and ARTMIP tier 2 (colored lines) AR datasets. The 1980–2019/23 trend information is given in the legend, with Δ signifying the cumulative percentage change relative to the mean and p signifying the p value, or the probability of obtaining a result at least this extreme if the null hypothesis were true. Post-1980 means are shown in the right-hand bar for (colors) ARTMIP and variant (thick black) mean, (thin dark gray) fixed poleward, (thin medium gray) fixed, and (thin light gray) semifixed datasets. For the variants, the number of FDR-controlled significant positive and negative trends is given in the legend. Note that VIMFC is not included for JRA-55-based datasets.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0234.1
Despite these large discrepancies between ARDT mean characteristics, trends and even interannual variability are often consistent between datasets. Globally, AR IWV has increased in all but one dataset (Fig. 6a) and AR IVT has increased in six of nine variant and four of seven ARTMIP datasets (Fig. 6c). Similar positive IVT and IWV trends are found over land, though with a larger portion of increases occurring post-2000 and slightly lower consensus. AR |V850| has mixed trends globally (Fig. 6b) and more significant decreases (six of nine variant, two of seven ARTMIP) over land (Fig. 7b). VIMFC has decreased more robustly, with (globally) three of six variant and two of six ARTMIP and (over land) six of six variant and four of six ARTMIP datasets showing significant decreases (note that JRA-55 VIMFC is not included). In general, interannual variability is larger over land than globally (note the different y-axis scales) and for IWV, VIMFC, and |V850| as compared to IVT. Among the outliers is, again, ClimateNet_DL, with the largest IVT increase (∼3%) of any dataset and large opposing |V850| trends, as well as ARCONNECT_v2, which has the largest increases in IWV (nearly 5% over land) and area (∼14% globally, 20% over land). This IWV increase may be linked to ARCONNECT_v2’s anomalous AR area increase combined with its lack of the usual poleward latitude shift which would temper the moistening signal.
A summary of these results is given in Fig. 8. Interquartile ranges (IQRs) of 1980–2019/23 changes are as follows: approximately 6%–9% for area, 0.5%–1% for IVT, 1.5%–2.5% for IWV, 0% to 0.5% for |V850|, and −3% to −5% for VIMFC globally, with larger IWV increases and |V_850| and VIMFC decreases over land. These values mirror model-based expectations of opposing thermodynamical and dynamical effects on AR evolution (e.g., Dettinger 2011; Gao et al. 2015; Ramos et al. 2016; McClenny et al. 2020; Payne et al. 2020; Wang et al. 2023). The small mean IVT increase is also consistent with the idea that ARs defined using time-invariant IVT threshold conditions (i.e., not evolving between model periods) will have relatively constant mean IVT in a warming climate due to the fixed lower boundary condition (see Shields et al. 2023).
The 1980–2019/23 percentage changes in AR area, IVT, IWV, |V850|, and VIMFC, averaged (or summed, for area) over the total AR area for each year. For each variable, global values are shown on the left and land values on the right. Box-and-whiskers denote the quartiles and extremes of trends in each variable. Statistically significant trends are also plotted as (ARTMIP) black and (variant) white circles, with the variants counted as three ARDTs (semifixed, fixed, and fixed poleward). Note that JRA-55 VIMFC data are not included. Variant markers are displayed if trends are significant for two of three (or two of two, for VIMFC) reanalysis datasets.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0234.1
Spatially, the FDR control procedure—which constrains the Type 1 error rate to 5% (see section 2d)—yields adjusted significance thresholds as low as 1 × 10−7, especially for fields with trends evenly distributed about zero. As a result, FDR-controlled trend consensus shows only two features on a gridpoint basis: increases in IWV and decreases in |V850| (Fig. 9). The IWV increases are present in both the variant and the ARTMIP samples, but much more robust among the variants, occurring over the Gulf Stream along the North American east coast and over southern Brazil, the south-central Pacific, and much of the western North Pacific (WNP) (Figs. 13c,d). Some ARTMIP datasets show increases in tropical AR IWV, which is mostly filtered out in the variants by the 20°N/S attenuation condition. Geospatial |V850| decreases are also significant mostly in the variants and occur over the WNP and parts of the North American Arctic (Fig. 13f).
The percentage of viable (left) ARTMIP tier 2 global and (right) novel variant AR datasets at each grid point with FDR-controlled significant positive (red shading) and negative (blue shading) annual-mean trends in AR conditions. Gray stippling indicates two-thirds consensus using a simple p < 0.05 significance criterion. In the case of co-occurring positive and negative trends, the trend sign with greater consensus is displayed, except in the case of equal values which are indicated by pink shading. Regions with fewer than 20 years of data are excluded for each dataset as are regions with only one dataset satisfying this completeness criterion. Note that VIMFC data for JRA-55-based datasets are not included.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0234.1
It is not surprising that consensus is lower for ARTMIP than for the variants, as ARTMIP features much greater ARDT diversity and has a shorter analysis period in the recent phase of rapid change. Still, it is noteworthy that fields which show robust global mean increases can be locally undetectable over the entire domain under FDR control. Using a more standard (uncontrolled) p < 0.05 threshold (gray stippling) yields a wide range of significant geospatial results consistent with observed global- and land-mean trends as well as previous AR trends studies (see also Fig. S12; Algarra et al. 2020; Gonzales et al. 2020; Ramseyer et al. 2022; Liang et al. 2023).
Another way to calculate global mean change is to compute trend magnitudes first at each grid point and then average spatially. The results obtained from this calculation are similar to the aggregate area synoptic trends, with IVT increasing by 0.8%–1.3% and IWV by 1.7%–3.6% on average (Fig. 10). However, they reveal significant regional heterogeneity, with local rates of change often being much larger. AR IWV has increased by 8+% over parts of Europe, the Arctic, and the tropics, and 850-hPa wind speed has decreased by similar magnitudes over the subtropical North Pacific, Europe, and the adjacent North Atlantic. IVT changes are more modest but locally still far exceed the <1% global mean, and VIMFC changes are spatially noisy but lean negative on average. One downside of such spatial averaging procedures is that they may overemphasize trends in moisture-limited regions which have known high detection sensitivity (see Lora et al. 2020) and fewer available ARDTs (see Fig. 4); note, for instance, that high AR frequency areas such as the midlatitude oceans typically see smaller rates of change than boundary, high- or low-latitude areas. Finally, it must be emphasized that only some of these changes represent statistically significant trends, as discussed in the previous figure.
Change in annual-mean AR characteristics for (left) 1980–2019 ARTMIP tier 2 ARDTs and (right) 1980–2023 variant ARDTs in reanalysis datasets, displayed as a percentage change relative to the mean for IVT, IWV, and |V850| and as an absolute magnitude for VIMFC. Grid points with statistically significant trends in at least two-thirds of available datasets are denoted by (FDR controlled) large yellow and (p < 0.05) small white stippling. Panel titles list the IQR of global mean trend values for the ARDTs. Regions with fewer than 20 years of data are excluded for each ARDT as are regions with only one ARDT satisfying this completeness criterion.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0234.1
b. AR mean versus maximum
Calculating average AR characteristics as in section 5a by aggregating all ARs into a single area mask erases system characteristics such as those relevant to forecasters or the public concerned about a specific event. Therefore, a desirable alternative approach is to calculate values first for each AR and then average over all ARs. We do this here by labeling the ARs in each dataset while removing features consisting of fewer than 100 grid points (20 for the 1.25° resolution JRA-55 GuanWaliser dataset) to filter out disconnected clumps of pixels that may significantly bias per-AR values (note that this eliminates many fleck-like features from ClimateNet_DL as well as a lesser but still significant number of small ARs from Reid500, TempestLR, and Mundhenk_v3; judging which features to optimally filter is beyond the scope of this section).
This method yields two interesting results (Fig. 11; see also Fig. S6). First, the total area increase comes equally from AR occurrence (No. of simultaneous ARs at each time step) and individual AR area; ARs are not only becoming more numerous but also larger. Judging by the variants alone, the increase in individual AR area would far outweigh that in occurrence, but the opposite seems to be true of the ARTMIP tier 2 datasets which are often more stringent than the variants. This distinction shows that AR geometry and spatial distribution trends are sensitive to detection methodology and highlights the utility of the ARTMIP/ensemble approach.
Violin plots of 1980–2019/23 percentage changes in AR occurrence (i.e., the No. of simultaneous ARs at each 6-hourly time step), area, IVT, IWV, |V850|, and VIMFC, calculated for each AR individually (or at each time step, for occurrence) before averaging over all ARs each year (or over all time steps that year, for occurrence). For synoptic variables, AR-mean values are shown on the left and AR-maximum values on the right. Box-and-whiskers denote the quartiles and extremes of trends in each variable. Statistically significant trends are also plotted as (ARTMIP) black and (variant) white circles, with the variants counted as three ARDTs (semifixed, fixed, and fixed poleward). Note that JRA-55 VIMFC data are not included. Variant markers are displayed if trends are significant for two of three (or two of two, for VIMFC) reanalysis datasets.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0234.1
Second, while trends in individual AR-mean synoptic characteristics are very similar to those calculated using an aggregate area mask, AR-maximum trends diverge significantly from those values. Since 1980, the most intense IVT in the average AR has increased by nearly 3%—approximately 3–6× the rate of mean IVT increase, likely due to the fixed lower boundary condition. AR-maximum IWV has increased by over 3% (1.5–2× the AR-mean increase), and despite decreases in aggregate-area |V850|, AR-maximum |V850| has not decreased. Most striking is the AR-maximum VIMFC trend, which exhibits low consensus but leans positive (+1%–3.5% IQR) in contrast to large decreases in mean values. For ARs, changes in mean characteristics evidently do not fully capture changes in the most intense core conditions which are often associated with maximum societal impacts. Instead, the worst AR conditions are intensifying more rapidly (see also Fig. S13).
c. Fixed-frequency analysis
That AR-maximum values are independent of the boundary condition touches on a property of ARs which has heretofore been overlooked by our calculations and the ARDT methodologies represented in this study: the confounding effect of area increases. McClenny et al. (2020) found that with AR areal expansion, increased sampling of grid points farther from AR cores obscured features with the highest IVT and precipitation rates, artificially decreasing or even reversing trends. This same principle likely affects our results. As an example, suppose that we are interested in ARs in Portland, Maine, which has seen ERA5 fixed AR frequency increase from ∼200 time steps (50 days) yr−1 in 1980 to ∼232 time steps (58 days) yr−1 in 2023 sorting for time steps with 50+% AR coverage within 1.5° of the city. It is perfectly correct to calculate IVT means over all AR points on these time steps and thereby quantify how the average AR has changed (+2.5% IVT). But this leaves another important question unanswered, namely, what would those 200 AR time steps look like in 2023, given that they represent the most intense ARs of the year? Similarly, how have the worst 1, or 5, or 10 AR events changed? In fact, mean AR IVT on the 100 worst AR time steps each year has increased by 7% near Portland in this ARDT, from 578 to 618 kg m−1 s−1.
One way to generalize this question to a geospatial grid is to calculate means over the most intense fixed-area subset of AR grid points (irrespective of AR boundaries) with respect to IVT, IWV, |V850|, and VIMFC each year. To minimize sensitivity, we test three area thresholds of 25% (Fig. S14), 50% (Figs. 12 and 13), and 60% (Fig. S15) of the mean annual-total AR area for each ARDT, with similar results observed in each case. In the literature, AR detection thresholds that evolve over time or between analysis periods allow researchers to characterize changes in ARs as defined relative to a changing background climate state. In the absence of such a true time-varying ARDT threshold, we use this fixed-frequency approach to address the question posed above, that is, how is a given (most intense) subset of AR points changing on a global scale? Note that this method does introduce large seasonal and geographic biases: moist settings will be preferentially selected for IWV, windy settings for |V850|, and high-convergence (likely mountainous) settings for VIMFC, even though the selected spatiotemporal region will differ from year to year.
Annual AR global (a) IWV, (b) 850-hPa wind speed, (c) IVT, and (d) VIMFC anomalies (relative to the post-1980 mean) for the variant mean (thick black line) and range (gray shading) and ARTMIP tier 2 (colored lines) AR datasets, averaged over the top 50% area of the most extreme conditions (A50) each year. The 1980–2019/23 trend information is given in the legend, with Δ signifying the cumulative percentage change and p signifying the p value, or the probability of obtaining a result at least this extreme if the null hypothesis were true. Mean values are shown in the right-hand bar for (colors) ARTMIP, (thick black) variant mean, (thin dark gray) fixed poleward, (thin medium gray) fixed, and (thin light gray) semifixed variant datasets. For the variants, the number of FDR-controlled significant positive and negative trends is given in the legend. Note that VIMFC is not included for JRA-55-based datasets.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0234.1
Annual AR land (a) IWV, (b) 850-hPa wind speed, (c) IVT, and (d) VIMFC anomalies (relative to the post-1980 mean) for the variant mean (thick black line) and range (gray shading) and ARTMIP tier 2 (colored lines) AR datasets, averaged over the top 50% area of most extreme conditions (A50) each year. The 1980–2019/23 trend information is given in the legend, with Δ signifying the cumulative percentage change and p signifying the p value, or the probability of obtaining a result at least this extreme if the null hypothesis were true. Mean values are shown in the right-hand bar for (colors) ARTMIP, (thick black) variant mean, (thin dark gray) fixed poleward, (thin medium gray) fixed, and (thin light gray) semifixed variant datasets. For the variants, the number of FDR-controlled significant positive and negative trends is given in the legend. Note that VIMFC is not included for JRA-55-based datasets.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0234.1
We find that the 1980–2019/23 intensification of AR conditions is amplified within this subset as the statistical effect of area increases compounds any preexisting mean value increases (Figs. 12 and 13; see also Fig. S16). IVT has increased by ∼3%–4% globally and over land; IWV has increased by ∼4%–6%; and VIMFC has increased by 6%–9% (Fig. S17), all with nearly unanimous consensus significance. Even |V850|, which has increased by no other metric, has modest consensus increases globally (but not over land). Over land, trend magnitude is similar but significance is lower due to the recurring pattern of a more stepwise or delayed post-2000 increase. As with some other fields, variant trends lie on the higher end of the distribution in part because their longer time frame captures more of the recent increases in intensity. Taken together, these results show persistent and comparatively large intensification of moisture-based variables associated with ARs.
These figures, however, provide little insight into the geography of fixed-frequency changes. To address this while highlighting extreme conditions, we take synoptic variables for the single worst AR time step at each location annually, i.e., approximating a once-per-year extreme. The resulting slope fields (Fig. 14) average to rates of change that are broadly consistent with (if somewhat smaller than) those of the global-mean fixed-frequency methodology. Furthermore, the IVT and IWV trend distributions are congruent with those of AR mean values (Fig. 10). In some regions, the worst yearly AR IVT has increased by more than 15%. IWV changes are more evenly distributed but still exceed +8% over Europe and certain polar regions. The |V850| has changed less robustly and mostly by a smaller magnitude, while VIMFC shows large (but nonsignificant) positive tendencies over eastern North America and East Asia, among other eastern continental and tropical regions (see also Fig. S18). We hypothesize that the larger cumulative change found in annual maximum relative to time-mean AR moisture variables—especially IVT—results not just from the selection of the annual maximum specifically but also from the subsetting which fixes the frequency of events both in space and in time.
Change in annual maximum AR characteristics for (left) 1980–2019 ARTMIP tier 2 ARDTs and (right) 1980–2023 variant ARDTs in reanalysis datasets, displayed as a percentage change relative to the mean for IVT, IWV, and |V850| and as an absolute magnitude for VIMFC. Grid points with statistically significant trends in at least two-thirds of available datasets are denoted by (FDR controlled) large yellow and (p < 0.05) small white stippling. Panel titles list the IQR of global mean trend values for the ARDTs. Regions with fewer than 20 years of data are excluded for each ARDT as are regions with only one ARDT satisfying this completeness criterion.
Citation: Journal of Climate 38, 6; 10.1175/JCLI-D-24-0234.1
There are many caveats to such methods. Annual-maximum values are disproportionately likely to be contaminated by tropical cyclones (TCs) in active TC regions (see Fig. S19), and increasing AR frequency would translate to intensification of fixed-frequency subsets even if the PDF of AR conditions were unchanged. Additionally, as in all other geospatial results, we see widespread FDR-controlled consensus mostly for IWV and mostly for the variants, despite the considerable similarity of the linear regression slope fields obtained from the variant and ARTMIP samples. Sensitivity to outliers is a topic which should be addressed especially with annual maximum values in future work. Nevertheless, these tendencies warrant continued monitoring, as increases in extreme AR IVT, IWV, and VIMFC may have important implications for severe AR-associated storm impacts during the coming decades.
6. Discussion and conclusions
The aim of this study was to describe changes that have taken place in ARs between the ∼1980 start of the satellite era and the near present. To do so, we used output from eight AR detection tools (ARDTs): one novel algorithm separated into three variants with different IVT conditions and seven global AR datasets from the ARTMIP tier 2 catalog. Trends in area, occurrence, pointwise frequency, mean latitude, integrated vapor transport (IVT), integrated water vapor (IWV), 850-hPa wind speed (|V850|), and vertically integrated moisture flux convergence (VIMFC) were assessed using MERRA-2, ERA5, and JRA-55 reanalysis data. These trends were computed using multiple averaging methodologies and compared across ARDTs to establish robustness, with the new ARDT variants facilitating more up-to-date trend calculations as well as some degree of sensitivity analysis.
A summary of results is given in Table 2, with 1980–2019/23 trends sorted into three confidence levels: high confidence (agreement among 75+% of available datasets, with at most 10% of datasets in opposition), medium confidence (agreement among 50+% of available datasets with at most 10% of datasets in opposition), and low confidence (agreement among 25+% of available datasets with no opposing trends). Despite minor differences in the analysis period (1980–2019 for ARTMIP and 1980–2023 for the variants) and major differences in detection methodologies, AR aggregate area (i.e., total coverage each year) and mean IWV have increased globally with high confidence. IVT, land IWV, and global mean latitude have increased with medium confidence. In general, land trends are less robust (perhaps due to higher variability) than global trends, with no consensus poleward shift and only low-confidence area increases. The |V850| and VIMFC behave differently according to subsetting method—both shifting more positive when evaluated over fixed-frequency subsets of the most intense AR points—and so are given split cells to represent this sensitivity. Further analysis indicates that the poleward latitude shift is most robust in the Southern Hemisphere, and within-AR calculations show AR maxima intensifying more rapidly than AR means. Over the 1940–79 presatellite period which is not considered for trend analysis, the ERA-5 and JRA-55 variant datasets depict smaller, further equatorward ARs with lower mean winds and IVT but higher mean VIMFC and moisture content.
Summary of trend confidence for variables calculated across (variant) variant only and (“all”) both variant and ARTMIP tier 2 AR datasets. Positive (+) and negative (−) trends are separated into high confidence (+++/−−−; agreement among 75+% of ARDTs with at most 10% of datasets in opposition), medium confidence (++/−−; agreement among 50+% of ARDTs with at most 10% of datasets in opposition), and low confidence (+/−; agreement among 25+% of ARDTs with no opposing trends) categories. “No trend” is signified by a circle. For variables with significant sensitivity to the calculation method, two values are displayed corresponding to (left) whole-area mean and (right) fixed top 50% area (A50) results. For whole-sample consensus, the variants are counted as three ARDTs.
These results are consistent with both observed and expected changes in AR characteristics. Over the last two decades, numerous studies have documented how ARs will become larger and more frequent, moister and rainier, and more primarily hazardous to human society in the future (Dettinger 2011; Lavers et al. 2013; Gao et al. 2015, 2016; Hagos et al. 2016; Ramos et al. 2016; Espinoza et al. 2018; Massoud et al. 2019; McClenny et al. 2020; Payne et al. 2020; Rhoades et al. 2020; O’Brien et al. 2022; Wang et al. 2023). Recently, studies have begun to detect these changes in the past. Already, global and/or regional increases in AR frequency, area, moisture uptake, poleward latitude, and IVT and decreases in winds have been observed (Gershunov et al. 2017; Gonzales et al. 2019; Algarra et al. 2020; Shearer et al. 2020; Ramseyer et al. 2022; Guan et al. 2023; Liang et al. 2023; Li and Ding 2024).
In characterizing changes in ARs, one must consider the dichotomy between area and intensity increases. The time-invariant thresholds used to define ARs in historical data—typically IVT thresholds—mean that as ARs expand, measures of intensity will change more slowly because of the presence of a fixed lower boundary condition. On the other hand, taking means over a fixed-area most-intense subset of AR points results in larger observed intensity increases (3%–4% IVT, 4%–6% IWV, 6%–10% VIMFC) because, by definition, more AR area means a larger area of extreme conditions even if the statistical distribution of AR conditions has not itself changed (though, as we can see from the whole area and within-AR means, it has changed). This effect is especially pronounced for |V850| and VIMFC—two variables which are not increasing in the global mean and so decrease in the AR mean but increase over a fixed-frequency domain corresponding to the most extreme conditions. This raises a question of priorities: when do we care more about the average AR, and when do we care more about the most intense conditions? One advantage of ARTMIP—and a core aspect of its philosophy—is that scientists might choose datasets optimized to analyze the aspect most pertinent to their specific research question.
While global-mean trends are both significant and moderately consistent between datasets, geospatial synoptic trends are often not detectable (IWV being the main exception) using the FDR control methodology, which sometimes results in significance thresholds 3–5 orders of magnitude lower than uncontrolled testing. AR frequency trends are more reliably detectable, with increases over the Southern Ocean and the eastern coastal regions of North America consistent with findings that forced changes will emerge relatively early in these locations (see Tseng et al. 2022). Frequency and intensity trends sometimes coincide, as over the North Pacific and the eastern coastal North American regions where AR conditions have become simultaneously more frequent and moister—a compound intensification of AR-related weather. When shifting to a more typical p < 0.05 significance condition, increases in the Southern Ocean and the eastern North American AR IVT, decreases in Northern Hemisphere AR |V850|, and some increases in AR |V850| over the western portion of the Southern Hemisphere are evident, and agreement between the variants and ARTMIP is improved (though geospatial trend consensus remains significantly lower for ARTMIP than for the more self-similar variants). Thus, one by-product of this work was to demonstrate large differences in the interpretability of geospatial change based on FDR controlled, standard uncontrolled, (untested) mean slope, and global-mean linear regression methods. Trends that were spatially consensus-significant under FDR control tended to correspond to those that we classified as “high confidence” in the global-mean sense of Table 2.
Previous studies have tied increases in hydrological extremes to the ∼7% K−1 Clausius–Clapeyron scaling of water vapor capacity with air temperature (Allen and Ingram 2002; Held and Soden 2006; Morrison et al. 2019; Martel et al. 2020; Chinita et al. 2021). For precipitation, the theory is relatively direct as extreme precipitation occurs when air is saturated with intense moisture convergence (see Allen and Ingram 2002; Held and Soden 2006). For ARs, the relationship is less straightforward, as ARs are not guaranteed to precipitate over their entire spatial extent. Indeed, they often only do so over a small fraction of their area; thus, the correlation between IVT and precipitation may be low outside of coastal or mountainous regions (e.g., Rutz et al. 2014; Gimeno-Sotelo and Gimeno 2023). In the ARDTs used in this study, mean AR IWV has increased by (IQR) 1.8%–2.8% K−1 globally between 1980 and 2019/23, after dividing by the corresponding global-mean temperature increases during that time (GISTEMP Team 2024; see also Simmons et al. 2017, 2021). However, AR IWV has increased by 4.6%–6.7% K−1 globally over most-intense fixed-frequency subsets, and local cumulative mean and maximum AR IWV increase can surpass 8%. Thus, our results indicate that while global-mean AR IWV scales well below both the ∼7% K−1 theoretical and ∼4%–6% K−1 reanalysis (Fig. 4) atmospheric moistening rates, this may be due to the time-invariant IVT thresholds used to define ARs, consistent with previously published ARTMIP literature. Future work might check this result by examining historical trends in ARs defined using actual time-varying thresholds, rather than proxy spatiotemporal subsets.
Despite the use of a wide range of ARDTs, some uncertainty remains, partly due to lower trend confidence over land but also due to the heterogeneity of the underlying reanalysis datasets. Because the number and diversity of assimilated observations increase by orders of magnitude throughout the analysis period (Kobayashi et al. 2015; McCarty et al. 2016; Gelaro et al. 2017; Hersbach et al. 2020), it is possible that biases may be introduced at various levels despite the core models remaining the same. For example, low-level moisture in MERRA-2 inputs is significantly affected by SSM/I data assimilation changes (Bosilovich et al. 2017). In fact, observational assimilation inhomogeneity is just one of several classes of error which might diminish the utility of reanalysis datasets in long-term trend analysis (Thorne and Vose 2010). While diagnosing these issues lies beyond this scope of this paper, further research may help clarify the respective roles of interdecadal variability, dataset uncertainty, and underlying climate change in the trends we have found, particularly relating to the ∼2015 increase in AR land area and intensity (correlated with sudden increases in extratropical extreme IVT area) and the effect of the pre-1994 stalling of reanalysis IWV trends (see Allan et al. 2022; Douville et al. 2022). This latter is particularly important as AR extent, IVT, and IWV appear to track with changes in global-mean IWV, meaning that the results obtained here may underestimate the true increases in AR scope and intensity and the temperature scaling of AR IWV since the start of the satellite era.
Reanalysis data remain our most expansive view of Earth’s atmosphere and the primary means by which atmospheric rivers are identified and analyzed. Rather than disregarding all trends as unreliable, we have endeavored to analyze them in a rigorous fashion so that future work might expand on their causes or caveats. Going forward, it will be increasingly possible to identify and attribute changes in the characteristics of different extreme weather types, from drought to extreme precipitation to convection to tropical cyclones. Atmospheric rivers, and, more broadly, extreme moisture transport, play a crucial role in connecting and regulating many of these events. As such, it is important to document to the best of our ability, the changing nature of atmospheric rivers in a warming world.
Acknowledgments.
This work was made possible by funding from NASA’s NPP Program and Modeling, Analysis, and Prediction Program. The authors thank NASA for providing the computational resources and MERRA-2 data that were used throughout this work, as well as Dr. Allison Collow for her feedback on the initial draft and three anonymous reviewers for constructive feedback which greatly strengthened the paper. The authors would also like to acknowledge the Atmospheric River Tracking Method Intercomparison Project (ARTMIP) and all its contributors, who are responsible for producing much of the data that was used in this paper. ARTMIP is a grassroots community effort and includes a collection of international researchers from various universities, laboratories, and agencies. Cochairs and committee members include Jonathan Rutz, Christine Shields, L. Ruby Leung, F. Martin Ralph, Michael Wehner, Ashley Payne, Travis O’Brien, Allison Collow, Juan Lora, and Paul Ullrich. Details on catalogs and developers can be found on the ARTMIP website. ARTMIP has received support from the U.S. Department of Energy Office of Science, Biological, and Environmental Research (BER) as part of the Regional and Global Climate Modeling Program and the Center for Western Weather and Water Extremes (CW3E) at Scripps Institute for Oceanography at the University of California, San Diego.
Data availability statement.
All data used are freely available to the public: MERRA-2 (https://disc.gsfc.nasa.gov/datasets?project=MERRA-2); ERA5 (https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-single-levels?tab=overview, https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-pressure-levels?tab=overview); JRA-55 (https://climatedataguide.ucar.edu/climate-data/jra-55); and ARTMIP tier 2 data for reanalysis datasets (https://www.earthsystemgrid.org/dataset/ucar.cgd.artmip.html). Scripts used in the production of this paper are available at https://github.com/lexihenny/ar_trends_paper. Novel AR detection data are available at https://www.dropbox.com/scl/fo/cugtzb8naj2jnus5t04qu/AMi8HJmvWt52akJS4Lj9USs?rlkey=znpm8jlsx25zgvmqekl1o4o69&st=cf0eo3e3&dl=0.
REFERENCES
Algarra, I., R. Nieto, A. M. Ramos, J. Eiras-Barca, R. M. Trigo, and L. Gimeno, 2020: Significant increase of global anomalous moisture uptake feeding landfalling Atmospheric Rivers. Nat. Commun., 11, 5082, https://doi.org/10.1038/s41467-020-18876-w.
Allan, R. P., K. M. Willett, V. O. John, and T. Trent, 2022: Global changes in water vapor 1979–2020. J. Geophys. Res. Atmos., 127, e2022JD036728, https://doi.org/10.1029/2022JD036728.
Allen, M. R., and W. J. Ingram, 2002: Constraints on future changes in climate and the hydrologic cycle. Nature, 419, 228–232, https://doi.org/10.1038/nature01092.
Arabzadeh, A., M. R. Ehsani, B. Guan, S. Heflin, and A. Behrangi, 2020: Global intercomparison of atmospheric rivers precipitation in remote sensing and reanalysis products. J. Geophys. Res. Atmos., 125, e2020JD033021, https://doi.org/10.1029/2020JD033021.
Banacos, P. C., and D. M. Schultz, 2005: The use of moisture flux convergence in forecasting convective initiation: Historical and operational perspectives. Wea. Forecasting, 20, 351–366, https://doi.org/10.1175/WAF858.1.
Barnes, E. A., 2013: Revisiting the evidence linking Arctic amplification to extreme weather in midlatitudes. Geophys. Res. Lett., 40, 4734–4739, https://doi.org/10.1002/grl.50880.
Barnes, E. A., and L. M. Polvani, 2015: CMIP5 projections of Arctic amplification, of the North American/North Atlantic circulation, and of their relationship. J. Climate, 28, 5254–5271, https://doi.org/10.1175/JCLI-D-14-00589.1.
Bell, B., and Coauthors, 2021: The ERA5 global reanalysis: Preliminary extension to 1950. Quart. J. Roy. Meteor. Soc., 147, 4186–4227, https://doi.org/10.1002/qj.4174.
Benjamini, Y., and Y. Hochberg, 1995: Controlling the false discovery rate: A practical and powerful approach to multiple testing. J. Roy. Stat. Soc., 57, 289–300, https://doi.org/10.1111/j.2517-6161.1995.tb02031.x.
Borkotoky, S. S., A. P. Williams, and S. Steinschneider, 2023: Six hundred years of reconstructed Atmospheric River activity along the US West Coast. J. Geophys. Res. Atmos., 128, e2022JD038321, https://doi.org/10.1029/2022JD038321.
Bosilovich, M. G., F. R. Robertson, L. Takacs, A. Molod, and D. Mocko, 2017: Atmospheric water balance and variability in the MERRA-2 reanalysis. J. Climate, 30, 1177–1196, https://doi.org/10.1175/JCLI-D-16-0338.1.
Chang, E. K. M., Y. Guo, and X. Xia, 2012: CMIP5 multimodel ensemble projection of storm track change under global warming. J. Geophys. Res., 117, D23118, https://doi.org/10.1029/2012JD018578.
Chemke, R., 2022: The future poleward shift of Southern Hemisphere summer mid-latitude storm tracks stems from ocean coupling. Nat. Commun., 13, 1730, https://doi.org/10.1038/s41467-022-29392-4.
Chinita, M. J., M. Richardson, J. Teixeira, and P. M. A. Miranda, 2021: Global mean frequency increases of daily and sub-daily heavy precipitation in ERA5. Environ. Res. Lett., 16, 074035, https://doi.org/10.1088/1748-9326/ac0caa.
Collow, A. B. M., and Coauthors, 2022: An overview of ARTMIP’s Tier 2 reanalysis intercomparison: Uncertainty in the detection of atmospheric rivers and their associated precipitation. J. Geophys. Res. Atmos., 127, e2021JD036155, https://doi.org/10.1029/2021JD036155.
Dai, A., and M. Song, 2020: Little influence of Arctic amplification on mid-latitude climate. Nat. Climate Change, 10, 231–237, https://doi.org/10.1038/s41558-020-0694-3.
Dettinger, M., 2011: Climate change, atmospheric rivers, and floods in California—A multimodel analysis of storm frequency and magnitude changes. J. Amer. Water Resour. Assoc., 47, 514–523, https://doi.org/10.1111/j.1752-1688.2011.00546.x.
Douville, H., S. Qasmi, A. Ribes, and O. Bock, 2022: Global warming at near-constant tropospheric relative humidity is supported by observations. Commun. Earth Environ., 3, 237, https://doi.org/10.1038/s43247-022-00561-z.
Espinoza, V., D. E. Waliser, B. Guan, D. A. Lavers, and F. M. Ralph, 2018: Global analysis of climate change projection effects on atmospheric rivers. Geophys. Res. Lett., 45, 4299–4308, https://doi.org/10.1029/2017GL076968.
Francis, J. A., and S. J. Vavrus, 2012: Evidence linking Arctic amplification to extreme weather in mid-latitudes. Geophys. Res. Lett., 39, L06801, https://doi.org/10.1029/2012GL051000.
Francis, J. A., and S. J. Vavrus, 2015: Evidence for a wavier jet stream in response to rapid Arctic warming. Environ. Res. Lett., 10, 014005, https://doi.org/10.1088/1748-9326/10/1/014005.
Gao, Y., J. Lu, L. R. Leung, Q. Yang, S. Hagos, and Y. Qian, 2015: Dynamical and thermodynamical modulations on future changes of landfalling atmospheric rivers over western North America. Geophys. Res. Lett., 42, 7179–7186, https://doi.org/10.1002/2015GL065435.
Gao, Y., J. Lu, and L. R. Leung, 2016: Uncertainties in projecting future changes in atmospheric rivers and their impacts on heavy precipitation over Europe. J. Climate, 29, 6711–6726, https://doi.org/10.1175/JCLI-D-16-0088.1.
Garreaud, R. D., M. Jacques-Coper, J. C. Marín, and D. A. Narváez, 2024: Atmospheric rivers in South-Central Chile: Zonal and tilted events. Atmosphere, 15, 406, https://doi.org/10.3390/atmos15040406.
Gelaro, R., and Coauthors, 2017: The Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2). J. Climate, 30, 5419–5454, https://doi.org/10.1175/JCLI-D-16-0758.1.
Gershunov, A., T. Shulgina, F. M. Ralph, D. A. Lavers, and J. J. Rutz, 2017: Assessing the climate-scale variability of atmospheric rivers affecting western North America. Geophys. Res. Lett., 44, 7900–7908, https://doi.org/10.1002/2017GL074175.
Gimeno-Sotelo, L., and L. Gimeno, 2023: Where does the link between atmospheric moisture transport and extreme precipitation matter? Wea. Climate Extremes, 39, 100536, https://doi.org/10.1016/j.wace.2022.100536.
GISTEMP Team, 2024: GISS Surface Temperature Analysis (GISTEMP), version 4. NASA Goddard Institute for Space Studies, accessed 1 March 2024, https://data.giss.nasa.gov/gistemp/.
Gonzales, K. R., D. L. Swain, K. M. Nardi, E. A. Barnes, and N. S. Diffenbaugh, 2019: Recent warming of landfalling atmospheric rivers along the west coast of the United States. J. Geophys. Res. Atmos., 124, 6810–6826, https://doi.org/10.1029/2018JD029860.
Gonzales, K. R., D. L. Swain, E. A. Barnes, and N. S. Diffenbaugh, 2020: Moisture- versus wind-dominated flavors of atmospheric rivers. Geophys. Res. Lett., 47, e2020GL090042, https://doi.org/10.1029/2020GL090042.
Guan, B., and D. E. Waliser, 2015: Detection of atmospheric rivers: Evaluation and application of an algorithm for global studies. J. Geophys. Res. Atmos., 120, 12 514–12 535, https://doi.org/10.1002/2015JD024257.
Guan, B., D. E. Waliser, and F. M. Ralph, 2023: Global application of the atmospheric river scale. J. Geophys. Res. Atmos., 128, e2022JD037180, https://doi.org/10.1029/2022JD037180.
Hagos, S. M., L. R. Leung, J.-H. Yoon, J. Lu, and Y. Gao, 2016: A projection of changes in landfalling atmospheric river frequency and extreme precipitation over western North America from the Large Ensemble CESM simulations. Geophys. Res. Lett., 43, 1357–1363, https://doi.org/10.1002/2015GL067392.
Hatsuzuka, D., T. Sato, and Y. Higuchi, 2021: Sharp rises in large-scale, long-duration precipitation extremes with higher temperatures over Japan. npj Climate Atmos. Sci., 4, 29, https://doi.org/10.1038/s41612-021-00184-9.
Held, I. M., and B. J. Soden, 2006: Robust responses of the hydrological cycle to global warming. J. Climate, 19, 5686–5699, https://doi.org/10.1175/JCLI3990.1.
Henny, L., C. D. Thorncroft, and L. F. Bosart, 2023: Changes in seasonal large-scale extreme precipitation in the mid-Atlantic and northeast United States, 1979–2019. J. Climate, 36, 1017–1042, https://doi.org/10.1175/JCLI-D-22-0088.1.
Hersbach, H., and Coauthors, 2020: The ERA5 global reanalysis. Quart. J. Roy. Meteor. Soc., 146, 1999–2049, https://doi.org/10.1002/qj.3803.
Jong, B.-T., H. Murakami, T. L. Delworth, and W. F. Cooke, 2024: Contributions of tropical cyclones and atmospheric rivers to extreme precipitation trends over the northeast US. Earth’s Future, 12, e2023EF004370, https://doi.org/10.1029/2023EF004370.
Kobayashi, S., and Coauthors, 2015: The JRA-55 reanalysis: General specifications and basic characteristics. J. Meteor. Soc. Japan, 93, 5–48, https://doi.org/10.2151/jmsj.2015-001.
Lavers, D. A., R. P. Allan, G. Villarini, B. Lloyd-Hughes, D. J. Brayshaw, and A. J. Wade, 2013: Future changes in atmospheric rivers and their implications for winter flooding in Britain. Environ. Res. Lett., 8, 034010, https://doi.org/10.1088/1748-9326/8/3/034010.
Li, Z., and Q. Ding, 2024: A global poleward shift of atmospheric rivers. Sci. Adv., 10, eadq0604, https://doi.org/10.1126/sciadv.adq0604.
Liang, J., Y. Yong, and M. K. Hawcroft, 2023: Long-term trends in atmospheric rivers over East Asia. Climate Dyn., 60, 643–666, https://doi.org/10.1007/s00382-022-06339-5.
Lochbihler, K., G. Lenderink, and A. P. Siebesma, 2017: The spatial extent of rainfall events and its relation to precipitation scaling. Geophys. Res. Lett., 44, 8629–8636, https://doi.org/10.1002/2017GL074857.
Lora, J. M., J. L. Mitchell, C. Risi, and A. E. Tripati, 2017: North Pacific atmospheric rivers and their influence on western North America at the Last Glacial Maximum. Geophys. Res. Lett., 44, 1051–1059, https://doi.org/10.1002/2016GL071541.
Lora, J. M., C. A. Shields, and J. J. Rutz, 2020: Consensus and disagreement in atmospheric river detection: ARTMIP global catalogues. Geophys. Res. Lett., 47, e2020GL089302, https://doi.org/10.1029/2020GL089302.
Ma, W., G. Chen, and B. Guan, 2020: Poleward shift of atmospheric rivers in the Southern Hemisphere in recent decades. Geophys. Res. Lett., 47, e2020GL089934, https://doi.org/10.1029/2020GL089934.
Maclennan, M. L., J. T. M. Lenaerts, C. Shields, and J. D. Wille, 2022: Contribution of atmospheric rivers to Antarctic precipitation. Geophys. Res. Lett., 49, e2022GL100585, https://doi.org/10.1029/2022GL100585.
Mahoney, K., and Coauthors, 2016: Understanding the role of atmospheric rivers in heavy precipitation in the southeast United States. Mon. Wea. Rev., 144, 1617–1632, https://doi.org/10.1175/MWR-D-15-0279.1.
Martel, J.-L., A. Mailhot, and F. Brissette, 2020: Global and regional projected changes in 100-yr subdaily, daily, and multiday precipitation extremes estimated from three large ensembles of climate simulations. J. Climate, 33, 1089–1103, https://doi.org/10.1175/JCLI-D-18-0764.1.
Massoud, E. C., V. Espinoza, B. Guan, and D. E. Waliser, 2019: Global climate model ensemble approaches for future projections of atmospheric rivers. Earth’s Future, 7, 1136–1151, https://doi.org/10.1029/2019EF001249.
McCarty, W., L. Coy, R. Gelaro, A. Huang, D. Merkova, E. B. Smith, M. Sienkiewicz, and K. Wargan, 2016: MERRA-2 input observations: Summary and assessment. NASA/TM-2016-104606/Vol. 46, 64 pp., https://ntrs.nasa.gov/archive/nasa/casi.ntrs.nasa.gov/20160014544.pdf.
McClenny, E. E., P. A. Ullrich, and R. Grotjahn, 2020: Sensitivity of atmospheric river vapor transport and precipitation to uniform sea surface temperature increases. J. Geophys. Res. Atmos., 125, e2020JD033421, https://doi.org/10.1029/2020JD033421.
Morrison, A., G. Villarini, W. Zhang, and E. Scoccimarro, 2019: Projected changes in extreme precipitation at sub-daily and daily time scales. Global Planet. Change, 182, 103004, https://doi.org/10.1016/j.gloplacha.2019.103004.
Mundhenk, B. D., E. A. Barnes, and E. D. Maloney, 2016: All-season climatology and variability of atmospheric river frequencies over the North Pacific. J. Climate, 29, 4885–4903, https://doi.org/10.1175/JCLI-D-15-0655.1.
Neiman, P. J., F. M. Ralph, G. A. Wick, J. D. Lundquist, and M. D. Dettinger, 2008: Meteorological characteristics and overland precipitation impacts of atmospheric rivers affecting the West Coast of North America based on eight years of SSM/I satellite observations. J. Hydrometeor., 9, 22–47, https://doi.org/10.1175/2007JHM855.1.
Newell, R. E., and Y. Zhu, 1994: Tropospheric rivers: A one-year record and a possible application to ice core data. Geophys. Res. Lett., 21, 113–116, https://doi.org/10.1029/93GL03113.
Newell, R. E., N. E. Newell, Y. Zhu, and C. Scott, 1992: Tropospheric rivers?—A pilot study. Geophys. Res. Lett., 19, 2401–2404, https://doi.org/10.1029/92GL02916.
O’Brien, T. A., and Coauthors, 2022: Increases in future AR count and size: Overview of the ARTMIP Tier 2 CMIP5/6 experiment. J. Geophys. Res. Atmos., 127, e2021JD036013, https://doi.org/10.1029/2021JD036013.
Payne, A. E., and Coauthors, 2020: Responses and impacts of atmospheric rivers to climate change. Nat. Rev. Earth Environ., 1, 143–157, https://doi.org/10.1038/s43017-020-0030-5.
Prabhat, and Coauthors, 2021: ClimateNet: An expert-labeled open dataset and deep learning architecture for enabling high-precision analyses of extreme weather. Geosci. Model Dev., 14, 107–124, https://doi.org/10.5194/gmd-14-107-2021.
Ralph, F. M., J. J. Rutz, J. M. Cordeira, M. Dettinger, M. Anderson, D. Reynolds, L. J. Schick, and C. Smallcomb, 2019: A scale to characterize the strength and impacts of atmospheric rivers. Bull. Amer. Meteor. Soc., 100, 269–289, https://doi.org/10.1175/BAMS-D-18-0023.1.
Ramos, A. M., R. Tomé, R. M. Trigo, M. L. R. Liberato, and J. G. Pinto, 2016: Projected changes in atmospheric rivers affecting Europe in CMIP5 models. Geophys. Res. Lett., 43, 9315–9323, https://doi.org/10.1002/2016GL070634.
Ramseyer, C. A., and Coauthors, 2022: Identifying eastern US atmospheric river types and evaluating historical trends. J. Geophys. Res. Atmos., 127, e2021JD036198, https://doi.org/10.1029/2021JD036198.
Reid, K. J., A. D. King, T. P. Lane, and E. Short, 2020: The sensitivity of atmospheric river identification to integrated water vapor transport threshold, resolution, and regridding method. J. Geophys. Res. Atmos., 125, e2020JD032897, https://doi.org/10.1029/2020JD032897.
Rhoades, A. M., and Coauthors, 2020: The shifting scales of western U.S. landfalling atmospheric rivers under climate change. Geophys. Res. Lett., 47, e2020GL089096, https://doi.org/10.1029/2020GL089096.
Rutz, J. J., W. J. Steenburgh, and F. M. Ralph, 2014: Climatological characteristics of atmospheric rivers and their inland penetration over the western United States. Mon. Wea. Rev., 142, 905–921, https://doi.org/10.1175/MWR-D-13-00168.1.
Rutz, J. J., and Coauthors, 2019: The Atmospheric River Tracking Method Intercomparison Project (ARTMIP): Quantifying uncertainties in atmospheric river climatology. J. Geophys. Res. Atmos., 124, 13 777–13 802, https://doi.org/10.1029/2019JD030936.
Riboldi, J., F. Lott, F. D’Andrea, and G. Rivière, 2020: On the linkage between Rossby wave phase speed, atmospheric blocking, and Arctic amplification. Geophys. Res. Lett., 47, e2020GL087796, https://doi.org/10.1029/2020GL087796.
Sharma, A. R., and S. J. Déry, 2019: Variability and trends of landfalling atmospheric rivers along the Pacific Coast of northwestern North America. Int. J. Climatol., 40, 544–558, https://doi.org/10.1002/joc.6227.
Shaw, T. A., and Coauthors, 2016: Storm track processes and the opposing influences of climate change. Nat. Geosci., 9, 656–664, https://doi.org/10.1038/ngeo2783.
Shearer, E. J., P. Nguyen, S. L. Sellars, B. Analui, B. Kawzenuk, K.-L. Hsu, and S. Sorooshian, 2020: Examination of global midlatitude atmospheric river lifecycles using an object-oriented methodology. J. Geophys. Res. Atmos., 125, e2020JD033425, https://doi.org/10.1029/2020JD033425.
Shields, C. A., and Coauthors, 2018: Atmospheric River Tracking Method Intercomparison Project (ARTMIP): Project goals and experimental design. Geosci. Model Dev., 11, 2455–2474, https://doi.org/10.5194/gmd-11-2455-2018.
Shields, C. A., and Coauthors, 2023: Future atmospheric rivers and impacts on precipitation: Overview of the ARTMIP Tier 2 High-Resolution Global Warming Experiment. Geophys. Res. Lett., 50, e2022GL102091, https://doi.org/10.1029/2022GL102091.
Simmons, A. J., P. Berrisford, D. P. Dee, H. Hersbach, S. Hirahara, and J.-N. Thépaut, 2017: A reassessment of temperature variations and trends from global reanalyses and monthly surface climatological datasets. Quart. J. Roy. Meteor. Soc., 143, 101–119, https://doi.org/10.1002/qj.2949.
Simmons, A. J., and Coauthors, 2021: Low frequency variability and trends in surface air temperature and humidity from ERA5 and other datasets. ECMWF Tech. Memo. 881, 99 pp., https://www.ecmwf.int/en/elibrary/81213-low-frequency-variability-and-trends-surface-air-temperature-and-humidity-era5.
Simon, S., J. Turner, M. Thamban, J. D. Wille, and P. Deb, 2024: Spatiotemporal variability of extreme precipitation events and associated atmospheric processes over Dronning Maud Land, East Antarctica. J. Geophys. Res. Atmos., 129, e2023JD038993, https://doi.org/10.1029/2023JD038993.
Skinner, C. B., J. M. Lora, A. E. Payne, and C. J. Poulsen, 2020: Atmospheric river changes shaped mid-latitude hydroclimate since the mid-Holocene. Earth Planet. Sci. Lett., 541, 116293, https://doi.org/10.1016/j.epsl.2020.116293.
Sousa, P. M., R. C. Blamey, C. J. C. Reason, A. M. Ramos, and R. M. Trigo, 2018: The ‘Day Zero’ Cape Town drought and the poleward migration of moisture corridors. Environ. Res. Lett., 13, 124025, https://doi.org/10.1088/1748-9326/aaebc7.
Tamarin, T., and Y. Kaspi, 2017: The poleward shift of storm tracks under global warming: A Lagrangian perspective. Geophys. Res. Lett., 44, 10 666–10 674, https://doi.org/10.1002/2017GL073633.
Tamarin-Brodsky, T., and Y. Kaspi, 2017: Enhanced poleward propagation of storms under climate change. Nat. Geosci., 10, 908–913, https://doi.org/10.1038/s41561-017-0001-8.
Thorne, P. W., and R. S. Vose, 2010: Reanalyses suitable for characterizing long-term trends. Bull. Amer. Meteor. Soc., 91, 353–362, https://doi.org/10.1175/2009BAMS2858.1.
Tseng, K.-C., and Coauthors, 2022: When will humanity notice its influence on atmospheric rivers? J. Geophys. Res. Atmos., 127, e2021JD036044, https://doi.org/10.1029/2021JD036044.
Ventura, V., C. J. Paciorek, and J. S. Risbey, 2004: Controlling the proportion of falsely rejected hypotheses when conducting multiple tests with climatological data. J. Climate, 17, 4343–4356, https://doi.org/10.1175/3199.1.
Voss, K. K., A. T. Evan, and F. M. Ralph, 2021: Evaluating the meteorological conditions associated with dusty atmospheric rivers. J. Geophys. Res. Atmos., 126, e2021JD035403, https://doi.org/10.1029/2021JD035403.
Wang, S., and Coauthors, 2023: Extreme atmospheric rivers in a warming climate. Nat. Commun., 14, 3219, https://doi.org/10.1038/s41467-023-38980-x.
Warner, M. D., C. F. Mass, and E. P. Salathe Jr., 2015: Changes in winter atmospheric rivers along the North American West Coast in CMIP5 climate models. J. Hydrometeor., 16, 118–128, https://doi.org/10.1175/JHM-D-14-0080.1.
Wilks, D. S., 2006: On “field significance” and the false discovery rate. J. Appl. Meteor. Climatol., 45, 1181–1189, https://doi.org/10.1175/JAM2404.1.
Wilks, D. S., 2016: “The stippling shows statistically significant grid points”: How research results are routinely overstated and overinterpreted, and what to do about it. Bull. Amer. Meteor. Soc., 97, 2263–2273, https://doi.org/10.1175/BAMS-D-15-00267.1.
Wille, J. D., and Coauthors, 2021: Antarctic atmospheric river climatology and precipitation impacts. J. Geophys. Res. Atmos., 126, e2020JD033788, https://doi.org/10.1029/2020JD033788.
Willett, K. M., R. J. H. Dunn, J. J. Kennedy, and D. I. Berry, 2020: Development of the HadISDH.marine humidity climate monitoring dataset. Earth Syst. Sci. Data, 12, 2853–2880, https://doi.org/10.5194/essd-12-2853-2020.
Xue, Y., J. Li, W. P. Menzel, E. Borbas, S.-P. Ho, Z. Li, and J. Li, 2019: Characteristics of satellite sampling errors in total precipitable water from SSMIS, HIRS, and COSMIC observations. J. Geophys. Res. Atmos., 124, 6966–6981, https://doi.org/10.1029/2018JD030045.
Yin, J. H., 2005: A consistent poleward shift of the storm tracks in simulations of 21st century climate. Geophys. Res. Lett., 32, L18701, https://doi.org/10.1029/2005GL023684.
Zhang, L., Y. Zhao, T. F. Cheng, and M. Lu, 2024: Future changes in global atmospheric rivers projected by CMIP6 models. J. Geophys. Res. Atmos., 129, e2023JD039359, https://doi.org/10.1029/2023JD039359.
Zhang, Z., F. M. Ralph, and M. Zheng, 2019: The relationship between extratropical cyclone strength and atmospheric river intensity and position. Geophys. Res. Lett., 46, 1814–1823, https://doi.org/10.1029/2018GL079071.
Zhou, Y., T. A. O’Brien, P. A. Ullrich, W. D. Collins, C. M. Patricola, and A. M. Rhoades, 2021: Uncertainties in atmospheric river lifecycles by detection algorithms: Climatology and variability. J. Geophys. Res. Atmos., 126, e2020JD033711, https://doi.org/10.1029/2020JD033711.
Zhou, Y., T. A. O’Brien, W. D. Collins, C. A. Shields, B. Loring, and A. A. Elbashandy, 2022: Characteristics and variability of winter northern Pacific atmospheric river flavors. J. Geophys. Res. Atmos., 127, e2022JD037105, https://doi.org/10.1029/2022JD037105.
Zhu, Y., and R. Newell, 1998: A proposed algorithm for moisture fluxes from atmospheric rivers. Mon. Wea. Rev., 126, 725–735, https://doi.org/10.1175/1520-0493(1998)126<0725:APAFMF>2.0.CO;2.
