Fewer deep cyclones projected for the midlatitudes in a warming climate, but with more intense rainfall

Cyclones are one of the most important weather systems for rainfall, particularly in the middle and high latitudes where they are responsible for more than 70% of both total and extreme precipitation (Hawcroft et al 2012, Pfahl and Wernli 2012, Dowdy and Catto 2017). For this reason, several studies have examined future changes in extratropical cyclone frequency and intensity, with projections consistently showing a poleward shift of the southern hemisphere storm track and fewer cyclones in the midlatitudes (Collins et al 2013). Projections are less robust in the Northern Hemisphere, although studies indicate potential increases in intense cyclones and cyclone rain (Collins et al 2013, Ulbrich et al 2013, Zhang and Colle 2017, Raible et al 2018). However, these projected changes are based on cyclones identified either at the surface or low levels, with little attention to cyclones higher in the troposphere, despite the sometimes substantial impacts from upper level cyclones including cut-off lows (Dowdy et al 2014, Barbero et al 2019). Cyclones that are vertically well organised, where a low pressure system can be clearly identified over a range of levels between the surface and 500 hPa, cause heavier rainfall than shallow cyclones identified over a more limited range of levels (Pepler and Dowdy 2020). However, no previous study has examined these vertically organised cyclones, here referred to as 'deep cyclones', in climate models, representing a major knowledge gap in relation to the occurrence of extreme weather in our warming climate.

Regional case studies in Australia have identified stronger projected future declines when cyclones are identified at 500 hPa when compared to projected future changes in surface cyclones (Dowdy et al 2014, Pepler et al 2016b). This result raises questions as to whether similar patterns can be identified in other regions of the globe, and what that may imply for future changes in cyclone-related total and extreme rainfall. Here we apply a cyclone tracking method to both mean sea level pressure (MSLP) and 500 hPa geopotential height (GPH) data from the ERA5 reanalysis (Hersbach et al 2020) and ten global climate models from the fifth Coupled Model Intercomparison Project (CMIP5, Taylor et al 2012) set of simulations. This is used to examine the occurrence of deep cyclones that are vertically well organised, and shallow cyclones that are identified only at a single level, including the relative importance of deep and shallow cyclones for total and extreme rainfall. Finally, we assess future changes in the frequency of different types of cyclone and how these contribute to projected changes in total and extreme daily rainfall through different regions of the world.

Cyclones are tracked using the University of Melbourne cyclone tracking scheme (Murray and Simmonds 1991, Simmonds et al 1999), available at https://cyclonetracker.earthsci.unimelb.edu.au/, with track data available at Pepler (2021a). In this tracking method, gridded fields are first converted to two hemispheric grids using a polar projection with an approximate resolution of 1.5° at 30° S, with 2° of spatial smoothing applied to remove gridpoint-scale lows. Cyclones are identified as local maxima in the Laplacian of the gridded field, which are then linked with a nearby minimum in the gridded data. Only cyclones which exceed a minimum threshold Laplacian when averaged over a 2° radius from the cyclone centre are retained, with both open and closed depressions included to allow for the lower spatial resolution of CMIP5 models. Cyclones are then joined into tracks using a probability matching technique. Surface cyclones are not tracked where model orography is above 1000 m, noting that the orography varies between models of different resolutions.

Cyclones are tracked using the ERA5 reanalysis (Hersbach et al 2020) and ten CMIP5 models (supplementary table S1), which were selected based on previous comprehensive work identifying models with skill at simulating Australian climate including cyclones and major modes of climate variability, while also considering model independence and spanning a broad range of projected future rainfall and temperature changes (CSIRO and Bureau of Meteorology 2015). Only one member was used for each model, with r1i1p1 chosen in all cases except CCSM4, for which r6i1p1 had more data available on Australian computer servers used for this project. Surface cyclones were tracked using 6-hourly MSLP, with an initial minimum intensity threshold of 0.5 hPa (deg. lat.)⁻². Due to data availability, upper level cyclones were tracked once per day using the daily mean 500 hPa GPH and a minimum intensity threshold of 1 m (deg. lat.)⁻². ERA5 daily GPH was calculated as the average of the four 6-hourly observations at 0000, 0600, 1200 and 1800 UTC. Cyclones were tracked for all years 1979–2005 in historical simulations and for 2006–2100 using both RCP4.5 and RCP8.5. Change figures mostly show projections for RCP8.5 at the end of the century (2070–2099) to maximise the signal-to-noise ratio in results.

To ensure a fair comparison between simulations, raw cyclone data was post-processed so each model had the same global number of cyclones during the common 1979–2005 period. This was based on the number of surface cyclones in ERA5 during the same period using a threshold of 1 hPa (deg. lat.)⁻², with the corresponding intensity threshold for other models and 500 hPa cyclones shown in supplementary table S1. Weaker ERA5 thresholds were also tested (0.5 or 0.8 hPa (deg. lat.)⁻²), as well as additional criteria such as a minimum duration of 24 h or a minimum movement of 1000 km (as in Hoskins and Hodges 2005, Neu et al 2013), but these choices made little difference to the results.

Cyclones were identified as deep or shallow through a three-step approach. First, an individual cyclone was classified as 'deep' if there was a corresponding low at the other level on the same UTC day within a 500 km radius, to account for the lower temporal resolution of 500 hPa cyclones. Secondly, a cyclone track was classified as 'deep' if it was deep at any time during the track, to allow for the temporal development of cyclones which are most well organised subsequent to the period of heaviest rainfall (Booth et al 2018, Pepler and Dowdy 2020). The remaining tracks are classified as 'shallow'. While vertically deep cyclones are typically stronger than shallow cyclones (Pepler and Dowdy 2020), the distinction between 'deep' and 'shallow' cyclones throughout this study is based purely on the vertical organisation of the system rather than incorporating intensity information. This is in contrast to some previous studies which have used the term 'deep' to describe a strong or intense subset of cyclones which may or may not have upper level development. This method does not require a unique match, so an individual upper low could potentially be associated with multiple surface lows during its track. This allows the classification of both surface and upper lows independently into 'deep' and 'shallow' subsets.

The locations of all cyclones on a given day were expanded such that they influence the surrounding 10°, consistent with the observed radius of cyclone-associated rainfall (Hawcroft et al 2012, Utsumi et al 2017, Booth et al 2018). The region of cyclone influence was calculated on the model grid, so will differ slightly between models based on their resolution. A point was classified as influenced by a deep cyclone on a given day if it was influenced by either a deep surface or a deep upper cyclone; a shallow surface cyclone if it was influenced by a shallow surface cyclone but no deep cyclone; and a shallow upper cyclone if there was a shallow upper cyclone but no deep or shallow surface system. Testing showed that there was little overlap between these three categories.

Cyclones from each model were associated with daily rainfall on the model grid, with the exception of ERA5, which was first bilinearly regridded to 1° to increase analysis speed, with cyclone rainfall data available at Pepler (2021b). This was achieved by converting the three cyclone categories into a binary mask used to multiply the daily rainfall prior to calculating a monthly sum for each category. In addition to total cyclone-related rainfall, we calculate the number of days with extreme rainfall. This is defined as the number of days greater than or equal to the local 99.7th percentile calculated from all days during the 1979–2005 period, which would be exceeded once per year on average in the current climate. Rain days were defined as days when rainfall was at least 1 mm, to exclude days with very low rainfall which are over-estimated in climate models (Stephens et al 2010), and the rain rate for a given cyclone category was calculated by dividing the total rainfall by the number of days. Mean values over the present (1979–2005) and future (2070–2099) periods were calculated for each model on the model grid prior to bilinearly regridding to a common 1° grid to allow the production of ensemble-mean figures.

Latitude mean changes in cyclone rainfall or frequency were calculated by taking the sum of all cyclones (or all cyclone rainfall) in a given latitude or latitude band in the future climate, and dividing by the sum across the whole latitude band in the current climate. Statistical significance of latitudinal mean changes is assessed for each model using a two-tailed t-test on annual (or seasonal) data, with results considered significant for p < 0.05. The latitude bands used for analysis were based on the relative dominance of deep cyclones in figure 1, and defined as the polar regions (poleward of 75°); the extratropics (45–75°) where deep cyclones produce at least 50% of rainfall; the subtropics (15–45°) where deep cyclones are less common but remain an important cause of heavy rainfall; and the tropics (15° S to 15° N). Although results are presented globally for completeness, the focus of this study's analysis and conclusions are on extratropical regions including the midlatitudes and subtropics, for which the cyclone identification scheme is optimised; such methods are less effective at identifying tropical cyclones (equatorward of 20°), particularly at climate model resolutions (e.g. Dowdy and Catto 2017).

Zoom In Zoom Out Reset image size

Using model-dependent thresholds that give the same total number of cyclones across the globe, the CMIP5 models produce broadly similar spatial patterns of cyclone frequency to those based on ERA5 reanalysis data (figure 1), with highest frequencies in the extratropics. However, when cyclones are identified using MSLP, the CMIP5 models consistently generate too many cyclones over northern hemisphere land areas, particularly near elevated topography such as the Himalayas. This is likely a consequence of the poor representation of topography in coarse-resolution models, and issues in the extrapolation of model pressure to sea level. In comparison, cyclones are less frequent than in ERA5 over the global oceans, particularly in the southern hemisphere midlatitudes.

In comparison to surface cyclones, upper cyclones are less influenced by topography and show less longitudinal variation in frequency. The CMIP5 models are well able to simulate these features, with consistently lower biases than the models produce for surface cyclones. This suggests that an increased focus on upper cyclones may help to improve the reliability of future cyclone projections (e.g. Dowdy et al 2014). The CMIP5 models consequently have fewer deep cyclones in the southern hemisphere when compared to ERA5, and a higher frequency of shallow surface cyclones in the northern hemisphere (figure 2, supplementary figure S1 (available online at stacks.iop.org/ERL/16/054044/mmedia)).

Zoom In Zoom Out Reset image size

Deep cyclones are most common between 45° and 75°, which we will refer to as the extratropics, where 42% of locations globally are influenced by a deep cyclone at least once on a given day based on ERA5 data (figure 2(a)). Both deep and shallow cyclones are more common in each hemisphere during the cooler months of the year (not shown). Consistent with their high annual frequency, deep cyclones explain 55% of annual rainfall and 70% of extreme rain days above the local 99.7th percentile in the global extratropics in ERA5 (figures 1(b) and (c)). Deep cyclones also play a very important role in precipitation in the midlatitude and subtropical regions between 15° and 45°, which we will refer to as the subtropics. In these regions they explain 11% of days but 23% of rainfall and 36% of extreme rain days using ERA5 data, which is broadly symmetrical between the two hemispheres.

Shallow surface cyclones also explain a larger proportion of extreme rain days in the subtropics than would be expected from their average frequency, whereas shallow upper cyclones with no surface low are less likely to produce notable rainfall. The results based on CMIP5 models are broadly similar to those based on ERA5 reanalysis data, although the models underestimate the contribution of deep cyclones to total and extreme rain in much of the southern hemisphere, while overestimating the frequency of shallow cyclones and their contribution to extreme rainfall in the northern hemisphere midlatitudes, consistent with biases in surface cyclone frequency at these latitudes.

The frequency of deep cyclones is projected to decrease across most of the globe (figure 3(a)). Projected changes are strongest in the southern hemisphere subtropics (15–45° S) where all ten models have a statistically significant decrease in deep cyclone frequency under RCP8.5, with a multimodel mean decline of 20% (range: −28% to −14%). Projected declines in deep cyclone frequency in this region are also statistically significant for all models under a more moderate emission scenario (RCP4.5), with a mean projected decline of −10%.

Zoom In Zoom Out Reset image size

All ten models also project a statistically significant decline in the frequency of deep cyclones in the northern hemisphere subtropics, with a mean decline of −15% under RCP8.5 (range: −20% to −11%). Projected declines in deep cyclone days are smaller but remain statistically significant for all models in the northern hemisphere extratropics (mean: −7%) and southern hemisphere extratropics (mean: −4%), noting that the historical average annual frequencies are higher in these regions. At least eight of the nine models with available data also have statistically significant declines projected for these regions using the RCP4.5 emissions scenario, averaging −9% for the northern hemisphere subtropics, −4% for the northern hemisphere extratropics and −3% for the southern hemisphere extratropics.

In contrast, the frequency of shallow surface cyclones is expected to increase over many land areas (figure 3(b)), particularly during the warmer half of the year (supplementary figure S3). All models except CNRM-CM5 project an increase in the number of shallow surface cyclones in the southern hemisphere subtropics during November–April under RCP8.5, with a mean increase of +10%, although this is only statistically significant for six models. Similarly, eight of the ten models project an increase in shallow surface cyclones in the northern hemisphere subtropics during May–October (mean: +10%), which is statistically significant in seven cases. This projected increase may include more heat lows as land masses warm, although it is likely to include a range of other rain-bearing lows such as small lows embedded in surface troughs (Pepler and Dowdy 2020), which may become increasingly common with warmer sea surface temperatures (Pepler et al 2016b). When assessing all surface cyclones, these opposing changes are conflated, leading to weaker and less robust projected declines in all surface cyclones over land areas, compared to the stronger declines projected for cyclones identified at 500 hPa (supplementary figure S4).

There is a robust decrease projected in the frequency of rain days in the southern hemisphere subtropics (figure 4(a)), with on average 8% fewer rain days in 2070–2099 compared to 1970–2005 under RCP8.5, which is statistically significant for all models. A large part of that decline can be explained by the projected 21% decrease in the number of rain days due to deep cyclones in this region (ensemble range: −15% to −26%). The northern hemisphere subtropics are expected to see 14% fewer deep cyclone rain days by 2070–2099 (ensemble range: −9% to −19%). Nine of the ten models also project a statistically significant decline in the total rainfall from deep cyclones in the southern hemisphere subtropics under RCP8.5 (mean change: −9%), and projected declines are also statistically significant for five models under RCP4.5 (mean change: −5%).

Zoom In Zoom Out Reset image size

All models project a statistically increase in the global number of heavy rain days under both emissions scenarios, with a mean increase of +95% under RCP8.5 (ensemble range: +57% to +123%) and +44% under RCP4.5 (range: +24% to +55%). The frequency of extreme rainfall increases for all categories, and is dominated by increases in heavy rainfall from non-cyclone days at the tropics and poles, and in heavy rainfall from deep cyclones in the extratropics. The frequency of extreme rain days is expected to increase in most latitudes, even where mean rainfall is projected to decline (figure 4(c)). In the subtropics from 15–45° there is a projected 46% increase in the frequency of extreme rainfall associated with deep cyclones under RCP8.5, despite the projected decline in cyclone frequency, which is statistically significant for all models. This amounts to an average of one extra deep cyclone day with extreme rainfall per decade for each gridpoint in the southern hemisphere subtropics and two extra deep cyclone days with extreme rain per decade in the northern hemisphere subtropics.

In both the midlatitudes and the extratropics, there is a larger projected increase in extreme rainfall from deep cyclones than would be expected from changes in cyclone frequency alone. This is due to a projected increase in the average intensity of cyclone rainfall. Averaged across all areas of the globe with at least one cyclone per year, the average daily rain rate on deep cyclone days is projected to increase by 27% under a high emissions scenario (+15% under a moderate emissions scenario). Similar increases in rainfall intensity are projected for shallow surface and shallow upper cyclones, with a smaller increase in rainfall intensity for days with no cyclone detected (ensemble mean: +15%). This means that the proportion of deep cyclones which are associated with extreme rain is projected to increase globally from 0.6% in the historical simulations to 0.9% in 2070–2099 under a moderate emissions scenario, and 1.2% under a high emissions scenario.

These results are consistent with an increase in atmospheric moisture availability on a warming planet associated with the Clausius–Clapeyron relationship of 7% K⁻¹, which can contribute to increased intensity of heavy rainfall events (Schneider et al 2010). These results are also consistent with previous studies which have projected an increase in the intensity of cyclone-related rainfall in several regions of the globe (Pepler et al 2016b, Yettella and Kay 2017, Zhang and Colle 2017, Raible et al 2018, Cavicchia et al 2020, Reboita et al 2020). Projected increases in rainfall intensity are largest in the tropics and close to the poles (supplementary figure S5), but more mixed in the midlatitudes, where the average rain rate is likely to decline in some areas on days with no cyclone identified, consistent with projected declines in rain days.

The increase in extreme rain days associated with deep cyclones is projected to occur broadly across the northern hemisphere, as well as across large parts of the southern hemisphere (figure 5). Projected increases are largest in parts of the North Pacific and northwest Atlantic, noting that the relatively short historical period used in this study (1979–2005) may not capture long-term climatological variability in cyclone occurrence in these regions (Varino et al 2019). Little change or a small increase in cyclone-related extreme rain days is projected for many southern hemisphere land areas even in locations where the total number of deep cyclones is expected to decline, highlighting a strong increase in the intensity of rainfall per deep cyclone in those cases. This is particularly notable in areas where SSTs are projected to warm most strongly, such as off the east coast of Australia, where rain rates are expected to increase for all cyclone categories (supplementary figure S5). In contrast to the number of extreme rain days, projected declines in deep cyclone numbers in these regions are likely to result in fewer rain days and lower total rainfall. These results clearly demonstrate the value in considering cyclones over a range of levels, particularly the occurrence of deep cyclones, when considering future projected changes, as their large contribution to mean rainfall and extremes may result in major impacts on water availability and flood risk (Seneviratne et al 2012).

Zoom In Zoom Out Reset image size

There is also a projected increase in the frequency of extreme rainfall from shallow surface cyclones in the midlatitudes and extratropics. In addition to an increase in the overall frequency of shallow surface cyclones over land areas (figure 3(b)), the likelihood that a shallow surface cyclone will produce extreme rain is also expected to increase across the globe, from 0.55% to 1.1% of cyclones. This increase in extreme rainfall from shallow surface cyclones is particularly notable around the east coasts of continents (including near the Gulf Stream, the Tasman Sea, and east of South America) where rapidly warming sea surface temperatures may contribute to additional moisture and latent heat release for convective systems that can contribute to the occurrence of extreme rainfall (Pepler et al 2016a).

Although cyclone climatologies have been previously found to be sensitive to the choice of tracking method used, particularly due to differences in the number of weak cyclones (Neu et al 2013, Ulbrich et al 2013, Rudeva et al 2014, Grieger et al 2018) our results are robust to a range of parameter choices including cyclone intensity thresholds and minimum distance criteria. However, as the southern hemisphere midlatitudes are a key region where deep cyclones and their associated rainfall is projected to decline, the underestimation of deep cyclones by CMIP5 models could potentially mean that projected future declines in total rainfall are also underestimated in these regions.

By the end of the 21st century, climate models project a robust decline in the frequency of vertically well-organised 'deep' cyclones, particularly in the southern hemisphere midlatitudes, although an increase in average cyclone rain rates means there may simultaneously be an increase in cyclone-related heavy rain. The average rainfall intensity on deep cyclone days increases by 27% in this study, which is broadly similar to what could be expected from thermodynamics alone based on 3 °C–4 °C of global warming and the Clausius–Clapeyron relationship of 7% K⁻¹ (Schneider et al 2010, Raible et al 2018). At the same time, there is a projected increase in the frequency of shallow cyclones identified only at the surface, particularly over land areas and during the warm half of the year, which could reflect an increased incidence of heat lows as the land surface warms.

While previous studies have projected future declines in cyclone frequency in the southern hemisphere midlatitudes (Grieger et al 2014, Pepler et al 2016b, Reboita et al 2020) and increases in cyclone-related extreme rainfall (Yettella and Kay 2017, Zhang and Colle 2017, Raible et al 2018), such studies have focused purely on surface cyclones. The results in this paper highlight the importance of considering upper cyclones in projections (Dowdy et al 2014), due to both their stronger projected changes and their better representation by coarse-resolution global models. We also highlight the value in using a multi-level approach when tracking cyclones, rather than only considering a single level, as vertically well organised (deep) cyclones are critically important for total rainfall and heavy rain events in the subtropics and extratropics (Lim and Simmonds 2007, Lakkis et al 2019, Pepler and Dowdy 2020). The projected increase in shallow surface cyclones over land areas means that studies which have projected cyclones based on MSLP alone may underestimate potential future declines in those cyclones which bring the majority of rainfall, and miss important changes in the structure and characteristics of cyclones influencing land areas.

This project is funded by the Earth Systems and Climate Change Hub of the Australian Government's National Environmental Science Program (NESP), and was assisted by resources from the Australian National Computational Infrastructure (NCI). The authors would like to thank Andrew Brown, Irina Rudeva and two anonymous reviewers for their helpful comments on this paper.

The raw gridded data used for this study (6-hourly sea level pressure, daily 500hPa GPH, and daily total precipitation) were accessed from the Australian National Computational Infrastructure (NCI), and are publicly available from the data providers. ERA5 data is directly available from the European Centre for Mid-Range Weather Forecasting via the Copernicus Climate Change Service at https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-pressure-levels?tab=overview, while CMIP5 data is available from the respective modelling centres and the Earth System Grid Federation at https://esgf-node.llnl.gov/projects/cmip5/.

The cyclone tracks generated as part of the study, as well as the key input files for the cyclone tracking, are available at https://doi.org/10.6084/m9.figshare.13393172. Post-processed grids of cyclone-associated rainfall for the 1979–2005 and 2070–2099 are available at https://doi.org/10.6084/m9.figshare.13393840.

The cyclone tracking used in this paper is available freely online at https://cyclonetracker.earthsci.unimelb.edu.au/, and the associated input files are available with the output track files at 10.6084/m9.figshare.13393172. Example wrapper scripts for converting netCDF files to suitable input formats and post-processing output files are available at https://github.com/apepler/cyclonetracking

Data analysis and figure generation was performed using the R and NCL programming languages. Code used to generate intermediate files, figures, and data for this paper are available at https://github.com/apepler/DeepCycloneTracking