SEARCH

SEARCH BY CITATION

Keywords:

  • climatology;
  • ENSO;
  • inter-annual variability;
  • severe thunderstorms;
  • thunderstorm environments;
  • thunderstorms

ABSTRACT

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Datasets and methodology
  5. 3. Verification against rawinsonde profiles
  6. 4. Proximity climatology
  7. 5. Severe thunderstorm environments 1979–2011
  8. 6. Discussion
  9. Acknowledgements
  10. References

Severe thunderstorms present a significant threat to property and life in Australia during the warm season (September to April). However, these relatively infrequent events are poorly understood in terms of frequency and occurrence for much of the continent due to a lack of in-situ observations. With the spectre of a changing climate, there is an increasing need to understand thunderstorms and their impact on Australia, both in the past and for the future. To facilitate this, the relationship between severe thunderstorms and their associated environments is used as a probabilistic proxy for direct observations. To establish these conditions, a proximity climatology of environments was developed for observed severe thunderstorms in Australia during the period 2003–2010 using the ERA-Interim reanalysis. Proximity soundings from the reanalysis for observed severe thunderstorms were used to develop covariate discriminants that identify the increased probability of an environment to produce severe thunderstorms. The covariates use a combination of ingredients describing instability (mixed-layer convective available potential energy) and potential for organized severe convection (deep-layer wind shear). These discriminants have been extrapolated to produce a climatology of environments favourable to the development of severe thunderstorms over the period 1979–2011 from this reanalysis. The inter-annual variations in both the spatial and temporal distribution of convective environments over Australia were analysed, with particular focus on the influence of El Niño-Southern Oscillation (ENSO) on the occurrence of severe thunderstorm environments. These results suggest that while ENSO has a substantial impact on the spatial distribution of severe thunderstorm environments over the continent, the link to frequency is more uncertain.

1. Introduction

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Datasets and methodology
  5. 3. Verification against rawinsonde profiles
  6. 4. Proximity climatology
  7. 5. Severe thunderstorm environments 1979–2011
  8. 6. Discussion
  9. Acknowledgements
  10. References

The environments associated with severe thunderstorm occurrence in Australia are far from being well understood. Limited climatologies of thunderstorm occurrence have considered small portions of the continent, particularly areas that are heavy populated (Yeo et al., 1999; Schuster et al., 2005). Much remains unknown about the frequency of severe thunderstorms outside of the urban areas of the coastal fringe, and even in those areas the exact return time for significant damaging events is far from certain. In the past 20 years, multiple high impact thunderstorm events have affected Australian metropolitan areas including; the 1999 Sydney Hailstorm (Yeo et al., 1999), the 2008 Gap microburst in Brisbane, record setting hailstorms in Perth and Melbourne in 2010, and severe hailstorms on Christmas Day 2011 in Melbourne (Allen, 2012). While 85% of Australia's population lives within 50 km of the coastline (Australian Bureau of Statistics, 2001), the vast inland areas are crucial to agricultural crops that form an integral part of the Australian economy. The difficulty in determining the frequency for the occurrence of severe weather associated with thunderstorms over this spatial area lies in the poor observational record currently available. Severe thunderstorms in Australia can produce very large hail (Schuster et al., 2005), extremely damaging wind gusts (Geerts, 2001) and tornadoes, and any of these phenomena can devastate agricultural production as well as populated areas. With the poor observational record (Griffiths et al., 1993), a method of estimation is required to reduce the bias of observed severe thunderstorm frequency towards populated areas, and to understand the climatological record of severe thunderstorms in Australia. This understanding of the past climatological occurrence is necessary in order to consider the potential impacts of climate variability and change on severe thunderstorm environments, and the resultant frequency of severe thunderstorm occurrence.

Key to addressing this observational inadequacy is a method that is independent of the severe thunderstorm report database. While this database is constantly improving, it leaves much to be desired both in consistency and completeness. Common ‘ingredients’ between all thunderstorm events are the presence of instability, moisture and lift, together with some measure of the environmental winds (Doswell et al., 1996; Doswell, 2001). Using the relationship between these ingredients and thunderstorm occurrence allows them to be applied as a proxy for the probability of a severe thunderstorm occurring. The source of information on these ingredients, however, can be more problematic, requiring some sort of proximity sounding (Potvin et al., 2010). The rawinsonde soundings commonly used for this purpose are not without issues, however, as they need to be recorded within few hours and 200 km of a storm to be considered representative of that storm's environment (Ryan, 1992). To offset these difficulties, an alternative is to use profiles produced from model or reanalysis data, which results in the added advantage of a more uniform spatial and temporal climatology of the aforementioned ‘ingredients’.

Pseudo-proximity soundings of severe thunderstorms can be derived from gridded datasets (Lee, 2002, Allen et al., 2011) to produce detailed proximity data for a greater number of soundings than would be otherwise available using traditional proximity rawinsonde soundings (Brooks et al., 2003; hereafter BLC03). This involves relating observational occurrences of severe thunderstorms to environmental parameters encapsulating the characteristics of the near storm environment; an ingredients-based approach (Doswell et al., 1996). Typically, these environmental parameters include a measure of moist static instability (covering the moisture and instability ingredients) and vertical wind shear. Convective available potential energy (CAPE) and S06 (0–6 km magnitude of shear difference of horizontal wind) are often used as the quantities to describe these environments, with the shear playing an important role in amplifying and sustaining the updraft through storm organization (Rasmussen and Blanchard, 1998, BLC03). These effective ‘observational profiles’ can then be stratified, and from this covariate discriminant relationships between CAPE and S06 that separate the populations are determined using an approach of best probability of detection (POD) and lowest false-alarm ratio. These discriminants can differentiate between severe and non-severe thunderstorm, and severe and significant severe thunderstorm environments (BLC03).

A proximity climatology for severe thunderstorm environments in Australia has already been derived using a high resolution operational forecast dataset (Allen et al., 2011) and observations from the Australian severe thunderstorm database. This climatology was used to derive discriminants appropriate to the Australian region, and these were found to mirror the criteria used by BLC03 with a small degree of variation (Brooks, 2012). We can infer that the physics of the environments that produce severe thunderstorms is reasonably consistent no matter the location, with only the frequency of those environments varying from region to region. The limited extent of the severe thunderstorm database, and also profiles derived from the model, meant that little could be said of the overall thunderstorm environments in Australia and their spatial distribution, nor comment made on inter-annual variability. While Brooks and Dotzek (2006) did consider the worldwide application of the BLC03 discriminant for the period 1970–1999, only passing attention was paid to Australia. Despite this, encouragingly the distribution of environments closely followed the pattern of satellite-derived hail observations (Cecil and Blankenship, 2012; Brooks, 2012). Thus to redress this, ERA-Interim reanalysis data over the defined Australian region were therefore considered for application as not only a proximity dataset for the observational climatology but also as a source of a long period environmental climatology. The concept of ‘warm-season’ severe thunderstorm environments discussed by Allen et al. (2011) is applied in a similar fashion for this current climatology that uses the ERA-Interim reanalysis over the period 1979–2011.

When using reanalysis data for pseudo-profiles of thunderstorm environments, caution must be taken when looking at fields involving strong vertical gradients (BLC03, Thompson et al., 2007). This is particularly applicable to thermal or capping inversions and the initiation of thunderstorms, and thus indirectly to convective inhibition. There are also issues with CAPE determined from the reanalysis being lower than co-located sounding data, likely due to a combination of surface heating, vertical resolution and averaging over the grid squares of the reanalysis (BLC03, Niall and Walsh, 2005; Thompson et al., 2007). Reanalysis data in the past has also been found to be slightly more moist and cooler in the boundary layer in comparison with observations (Betts et al., 1996), while low-level winds in the reanalyses tend to be weaker than those observed (Zwiers and Kharin, 1998). Caution must therefore be taken in applying the results of the proximity climatology to observed soundings quantitatively. However, given that one of the latest and most advanced reanalysis datasets is applied consistently here, and the capability of reanalyses to represent convective ingredients on an inter-annual scale (Brooks et al., 2007; Brooks 2009), to some extent these problems may be moderated compared to earlier reanalysis products.

A difficulty present in this method of characteristic parameters is a lack of inclusion of a lift or initiation requirement for any thunderstorm to develop. Typically, this requires a factor to overcome the latent convective inhibition (CIN, Tailleux and Grandpeix, 2001) and promote the convective process. The relative small-scale features that can promote convection further complicate understanding of this initiation process (Doswell, 1987, Doswell et al., 1996). These features include orography, mesoscale features, drylines, shortwave troughs and diurnal heating. Each of these processes are generally poorly represented by lower resolution models, as they are of a sub-grid scale and often parameterized. In addition, the poor rendition of vertical gradients noted in reanalysis data makes actually understanding the energy required for convection difficult. Accordingly, it is not possible to comment on the actual occurrence of thunderstorms based on these environments, and we can therefore only comment on the occurrence of favourable environments for severe storm development.

The extension of the climatology of environments to a multi-decadal period allows consideration of the influence of both El Niño-Southern Oscillation (ENSO) (Rasmusson and Wallace, 1983) and northern Australian sea surface temperatures (SSTs) on inter-annual variations of the environments. Certain ENSO cycles have been found to enhance hailstorm activity in the Sydney region (Leigh and Kuhnel, 2001). This is especially the case during La Niña, or the rising Southern Oscillation Index (SOI) between July and November and falling thereafter, with a much higher probability of exceeding the mean annual number of hailstorms. In southeast Queensland the higher numbers of severe thunderstorm days were found to be 60% more likely in neutral ENSO years than La Niña years, while El Niño was found to be the second most favourable pattern for severe thunderstorm occurrence (Yeo, 2005). Problematically, these studies have been regionally limited, and relied on relationships to observational climatologies (e.g. the Sydney region: Kuhnel, 1998, Leigh and Kuhnel, 2001, and southeast Queensland: Yeo 2005). These studies have concluded that the most favourable ENSO phase for severe thunderstorms appears to be neutral, while La Niña produced the lowest frequency. Despite the relative weakness of the ENSO signal in southeast Australia compared to inland regions (Ropelewski and Halpert, 1987), both ENSO and northern Australian SSTs have an impact on the moisture and circulation pattern over the Australian continent, and thereby severe thunderstorm environments by way of ingredients. Thus given the potential for continental scale impacts due to ENSO phase, we investigate this further to ascertain its effect on thunderstorm environments across the country.

This article is structured as follows: Section 'Datasets and methodology' describes the three databases that have been produced and analysed in this study. Section 3 examines the performance of the gridded reanalysis set as proximity data to rawinsonde observations from across Australia. Section 4 analyses the proximity climatology produced from the ERA-Interim reanalysis, and places the covariate discriminants in context of international studies and the high resolution Australian climatology. Section 4 also tests the inter-annual frequency of environments in comparison to the observational severe thunderstorm dataset and the temporal distribution of these environments. Section 5 describes the inter-annual variability of Australian severe thunderstorm environments over the period 1979–2011 and investigates the relation of these environments to ENSO and northern Australia SSTs. Finally, Section 6 provides the conclusions and potential future applications of this climatology.

2. Datasets and methodology

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Datasets and methodology
  5. 3. Verification against rawinsonde profiles
  6. 4. Proximity climatology
  7. 5. Severe thunderstorm environments 1979–2011
  8. 6. Discussion
  9. Acknowledgements
  10. References

2.1. Severe thunderstorm dataset

A database of 1550 severe thunderstorm events was produced from the Bureau of Meteorology (BoM) national severe thunderstorm records, significant weather summaries and other verifiable thunderstorm events for seven warm-seasons from March 2003 to April 2010 (Figure 1(a)). This database and its formulation is further described by Allen et al. (2011), and corresponds to the reports for the high-resolution proximity used in that study. Reports within the database are stratified into two categories: severe thunderstorm reports; with any one or more of hail in excess of 2 cm, wind gusts in excess of 90 km h−1 and any tornadic event; and significant severe thunderstorm reports; which are defined as 5 cm or greater hail size, winds exceeding 120 km h−1 and tornadoes exceeding F2 intensity (Hales, 1993; Doswell, 2001). We note that in this study, the reports correspond to severe thunderstorms, and not to reports of severe weather. This distinction is achieved by attributing any report made within 40 km and 2 h to the thunderstorm event first reported, similar to the approach of Kelly et al. (1985). The convective warm-season used is taken to be between September and April, with events outside this period scrutinized for the production of large hail. We acknowledge that this may remove some events that may otherwise have been classed as severe. However, these particular events are infrequent and uncommon over the summer period in Australia.

image

Figure 1. (a) Locations of reports of warm season severe and significant severe thunderstorms for Australia from the severe thunderstorm dataset produced by Allen et al. (2011), over the period March 2003 to April 2010. (b) The Australian region chosen for the climatology. Stars indicate the location of Australian rawinsonde stations, with the Brisbane Airport rawinsonde site (B), and the Moree rawinsonde site (M) chosen for station cases. The dashed area is referred to as the East Australian Region (EAReg), which extends between 39°S and 20°S, and 144°E and 154.5°E. The small box corresponds to the area displayed in panel (c) of this figure. (c) The Brisbane region, showing a single thunderstorm report (black circle), the location of ERA-Interim gridpoints (pluses), and the areas over which the reanalysis gridpoint values are averaged.

Download figure to PowerPoint

2.2. Proximity datasets and approach

Rawinsonde atmospheric profiles were used to verify ERA-Interim reanalysis data and were derived for 16 stations (Figure 1(b)). The stations were selected over a range of inland and coastal areas scattered across the continent to maximize the available number of comparisons and reliability of records. This allowed comparison of stations from the sub-tropics to the mid-latitudes, where the greatest frequency of severe convection is located, thereby allowing the sampling of a greater range of environments. As the complete comparison was not particularly informative for some variables, two stations were identified to examine specific environmental factors and the performance of the reanalysis in coastal and inland locations. The stations chosen were Brisbane Airport, a near coastal station on the east coast with the near highest thunderstorm occurrence, and Moree, an inland station located to the west of the coastal mountain range with a comparable number of high CAPE environments (Figure 1(b)). Soundings were removed during the quality control process to address missing levels and values for the rawinsonde data. Owing to station recording frequency and limitations surrounding the launch of balloons in the event of failure, only 0000UTC soundings were retained in this filtering process. These data were then interpolated onto the ERA-Interim pressure level grid between the surface and 100 hPa for the calculation of sounding parameters.

To examine the influence of using the ERA-Interim dataset, a brief comparison was considered to the MesoLAPS numerical weather prediction model, a 12.5 km horizontal resolution model with 29 vertical levels nested within the Limited Area Prediction System (LAPS) model (Puri et al., 1998). Further details of the steps and filtering process used to produce the comparison of this model dataset to the rawinsonde data are detailed in the study of Allen et al. (2011).

The European Centre for Medium-Range Weather Forecasts (ECMWF) has produced three high quality gridded datasets of meteorological parameters from operational forecast models run while ingesting observational data for climatological study over the late 20th century. The latest of these ECMWF Re-Analysis (ERA) projects is the ERA-Interim reanalysis, a state of the art 0.75° horizontal resolution dataset that has temporal resolution every 6 h, 29 relevant vertical pressure levels and a global gridded dataset (Dee et al., 2011). In this study, a coarse resolution 1.5° ERA dataset is used, owing to the availability and the resolution being closer to climate models, an intended later application of this climatology. Pseudo-proximity soundings were constructed from pressure, temperature, mixing ratio, and horizontal wind in both the meridional and zonal direction. In comparison to other reanalyses, the ERA-Interim is considered to provide a better overall moisture profile, and high levels of performance in data assimilation. Unlike the earlier NCEP reanalyses that have known problems with reliability in the Southern Hemisphere (Bromwich et al., 2007; Allen et al., 2010), the ERA-Interim dataset provides a high quality dataset for creating gridded pseudo-proximity soundings.

Variables calculated using the improved algorithm included Most Unstable, a 50 mb surface Mixed-Layer and Surface-Based CAPE fields (MUCAPE, MLCAPE and SBCAPE), 50 mb surface Mixed-Layer CIN (MLCIN), sounding-related parameters, lifted indices, 2–4 km lapse rates, and bulk magnitude vertical wind shears for both 0–6 and 0–1 km (S06, S01). In addition to improve visualization of the differences between model and rawinsonde CAPE, the derived variable Wmax (inline image) was applied for 'Severe thunderstorm dataset' Using this suite, a comparison to identical variables calculated using rawinsonde data was considered for the most relevant variables for thunderstorm environments.

3. Verification against rawinsonde profiles

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Datasets and methodology
  5. 3. Verification against rawinsonde profiles
  6. 4. Proximity climatology
  7. 5. Severe thunderstorm environments 1979–2011
  8. 6. Discussion
  9. Acknowledgements
  10. References

The 0000UTC rawinsonde data were used to assess the performance of reanalysis data to produce the variables to describe severe convective environments. While in the past, the reanalysis data have been considered to be relatively weak in their resolution of surface temperature and boundary layer, the ERA-Interim reanalysis has a comparable number of vertical levels to the MesoLAPS model. Three variables that were considered to be important based on the discriminants used to this point and known weaknesses in the reanalyses (BLC03) were MLCAPE, S06 and MLCIN. The rawinsonde and reanalysis comparisons were also considered against the performance of the high resolution MesoLAPS to examine the influence of horizontal resolution and the variations from operational mesoscale model forecasts.

MLCAPE for the MesoLAPS model shows a large degree of scatter, with a tendency for the values to be overestimated for cases of MLCAPE greater than 10 J kg−1 (Figure 2(a)). In contrast, the MLCAPE for ERA-Interim pseudo-profiles was characterized by less scatter than for MesoLAPS, and had a comparison that was closer to a one-to-one ratio (Figure 2(b)). Consideration of the rawinsonde and model averages suggests that the performance of ERA-Interim pseudo-soundings is closer to the rawinsonde dataset. This is illustrated by an improvement of nearly 100 J kg−1 in the root mean square difference (RMSD) in the all sounding station case, and a smaller variation in the average MLCAPE values (Table 1). Individual stations display variability in the performance with higher RMSD for both Brisbane and Moree for the ERA-Interim (Figure 3(a) and (b)). This is likely to be partially related to the location of the nearest reanalysis gridpoint to the actual rawinsonde station, with larger distances increasing the potential RMSD. Another factor that influences these comparisons is the timing of synoptic systems within the reanalysis, which while similar to the observed environment, may differ on the mesoscale due to resolution. It has been previously noted that there is a tendency to under-estimate MLCAPE when profiles are calculated using reanalysis datasets (e.g. NCEP/NCAR Niall and Walsh, 2005; NCEP II Thompson et al., 2007). In contrast, the ERA-Interim tends to slightly overestimate when compared to rawinsonde profiles, albeit less than the MesoLAPS dataset. This overestimation is accompanied by a smaller mean difference between the rawinsonde and reanalysis as compared to earlier reanalyses and additionally by higher correlations. This suggests that the ERA-Interim reanalysis data can provide a reasonable picture of MLCAPE profiles for Australia with a small positive bias, but with some limitations in coastal regions where topographic boundaries expose the spatial resolution.

image

Figure 2. Comparison of rawinsonde soundings and both the MesoLAPS and ERA-Interim proximity soundings for all cases (4988 soundings for MesoLAPS, 3697 soundings for the ERA-Interim reanalysis) for: (a) Wmax as derived from values of MLCAPE for MesoLAPS profiles, (b) as (a) except derived using MLCAPE from ERA-Interim Profiles, (c) and (d) as (a) and (b) except for 0–6 km Bulk Vertical Wind Shear (S06), (e) and (f) as (a) and (b) except Mixed-Layer parcel Convective Inhibition (MLCIN). The dashed line shows one-to-one ratio.

Download figure to PowerPoint

Table 1. Statistical comparison of ERA-Interim and MesoLAPS pseudo-profile variables to rawinsonde observations for 0000UTC soundings for non-zero rawinsonde CAPE cases
 CorrelationRMSDRad. AvgModel Avg
  1. All MesoLAPS includes comparisons for proximity soundings at all 16 rawinsonde stations (4988 soundings), All ERA-Interim for the reanalysis dataset to rawinsonde (3697 soundings), Brisbane ERA includes comparisons for Brisbane Airport (marked B, Figure 1b, 598 soundings) and Moree ERA for comparisons to Moree Aerodrome (marked M, Figure 1b, 389 soundings). For MLCIN, all non-zero cases in the presence of non-zero CAPE are considered (3499 soundings MesoLAPS, 2834 ERA-Interim soundings). Correlation is the linear regression correlation coefficient between the respective model variables and the rawinsonde, RMSD is the root mean square difference, Rad. Avg is the rawinsonde average, and Model Average refers to the average for the respective comparison datasets (ERA-Interim and MesoLAPS).

MLCAPE
All MesoLAPS0.68421255328
All ERA-Interim0.51328234240
Brisbane ERA0.66377283332
Moree ERA0.57415355376
S06
All MesoLAPS0.686.01515
All ERA-Interim0.874.61414
Brisbane ERA0.804.91113
Moree ERA0.843.81513
MLCIN
All MesoLAPS0.53563712
All ERA-Interim0.77555025
image

Figure 3. Comparison of rawinsonde to ERA-Interim profiles for (a) Wmax as derived from values of MLCAPE at Brisbane (598 soundings) and (b) Moree (389 soundings). (c) As (a) except S06 at Brisbane, and (d) as (b) except S06 at Moree. A line for the one-to-one ratio is shown.

Download figure to PowerPoint

As demonstrated by Allen et al. (2011), the MesoLAPS forecast model output tended to perform well producing profiles of the vertical wind shear over the depth of the atmosphere (Figure 2(c)). The ERA-Interim in comparison is even better than the MesoLAPS, with a smaller RMSD and a slight tendency to under-estimate the values of S06. While the averages for MesoLAPS and rawinsonde data are very similar, the higher RMSD reflects the larger scatter, related in some part to the forecast nature of the data (and resultant errors) and the slight tendency for the model to over-estimate shear. In the case of the individual stations, Brisbane has a greater degree of scatter, with the reanalysis tending to overestimate the values of S06. In contrast, the RMSD for Moree is much lower, with an under-estimating bias for the strength of shear environments. Again, these comparisons may be influenced by the proximity of the nearest reanalysis gridpoint to the rawinsonde station. Another factor in the Brisbane case may be the influence of sea-breezes and other small scale features, with that particular station being located very close to the coast. Thus the difference that occurs in the reanalysis climatology is likely to be in the 0–1 km layer as noted by previous studies (Zwiers and Kharin, 1998; BLC03; Allen et al., 2011), where vertical gradients may not be well captured by the resolution. Outside of the coastal regions, S06 is well represented by the reanalysis dataset, making it an effective variable for use in any discriminant. This result agrees well with the comparisons made between rawinsonde and reanalysis-derived proximity soundings in the United States (Lee, 2002).

Representation of CIN presents a regular problem in many model datasets, and the values of MLCIN may not reflect observations. Both MesoLAPS and the ERA-Interim tend to do a poor job of representing MLCIN (Figure 2(e) and (f)), with both datasets having large scatters from the rawinsonde calculations and high RMSD (as much as 50 J kg−1). In the case of MesoLAPS, the bias is a severe under-estimation of the MLCIN value, while for the reanalysis there is a somewhat lower difference between the profiles and the comparable rawinsonde values. In both datasets, the potential difference is greater than what would be considered the minimum requirement for initiation (approximately 25 J kg−1). This reflects a real problem in identifying environments that are likely to result in severe convection. This is particularly a problem in locations where, for much of the time, the environments may sustain large MLCAPE under a strong capping inversion such as the tropics and may be an explanation for the south Texas problem noted by BLC03. Accordingly, MLCIN is treated with caution in considering favourable environments for this study, and we acknowledge the potential for biases in the favourable MLCAPE environmental fields due to this problem.

4. Proximity climatology

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Datasets and methodology
  5. 3. Verification against rawinsonde profiles
  6. 4. Proximity climatology
  7. 5. Severe thunderstorm environments 1979–2011
  8. 6. Discussion
  9. Acknowledgements
  10. References

A difficulty when using reanalysis as a proximity dataset is selecting the environment that most closely reflects the environment realized by the storm. To select gridpoints other than the nearest can be somewhat problematic, as the spatial distances between reanalysis gridpoints may be the difference between an oceanic or an inland location. This can result in a completely different environment than what was actually proximal to the convection. To produce proximity soundings from this gridded dataset, we consider a report (Figure 1(c)) that occurs between gridpoints (each of which corresponds to an average of the environment of the area surrounding that gridpoint). To ensure, this selection does not unfairly bias the climatology, any given report has its proximity sounding selected by choosing the maximum MLCAPE of any of the surrounding four gridpoints, provide those gridpoints occur over land. This results in a sounding being within 3 h and 200 km of the report for which it is considered to be proximal.

To ensure the selection of representative warm season soundings, reports with proximity soundings of CAPE less than 100 J kg−1 were removed to address the issue of poorly sampled environments. Given the relatively small sample of severe and significant severe thunderstorm reports, it is not surprising that only 879 cases make it through the screening of CAPE greater than 100 J kg−1 (Figure 4). This results in no meaningful discriminant being possible from the ERA-Interim climatology without substantially increasing the sample size. To place this in context, however, for the 32-year climatology, the number of environments which exceed a determined threshold at any given gridpoint is sought, where the choice of gridpoint is not of concern. As all gridpoints over the continent are considered, we reduce the potential of an environment ‘miss’. Accordingly we can apply the previously determined discriminants to ascertain whether the values determined from MesoLAPS and BLC03 are usable with this reanalysis. As MesoLAPS tends to overestimate the CAPE of environments in general, we chose to use the slightly lower significant severe (hereafter SigSEV) discriminant from Allen et al. (2011):

  • display math(1)
image

Figure 4. Distribution of proximity soundings (897 soundings) for severe thunderstorm reports (dots) and significant severe thunderstorm reports (stars) in log(CAPE)-log(S06) phase space for the ERA-Interim reanalysis. Discriminant lines shown correspond to the discriminants determined for SEV environments based on POD (Equation (2)) and SigSEV environments (Equation (1)).

Download figure to PowerPoint

This provides a value more appropriate for the lower-resolution reanalysis data, and can be considered the equivalent discriminant for significant severe determined by BLC03. The second lower threshold for severe thunderstorm (hereafter SEV) environments is also applied; the discriminant given by the equation:

  • display math(2)

These discriminants were then tested using POD and false alarm ratio (FAR) for the ERA-Interim dataset. The effect of using the coarser resolution dataset is noticeable, with the POD for SEV environments (Table 2) dropping to only three-quarters of the MesoLAPS proximity climatology (Allen et al., 2011). The difference between the BLC03 and Allen et al. (2011) discriminants in terms of FAR/POD spread is very similar, with the BLC03 achieving a slightly higher POD. As noted by Brooks (2012), given the similarity between these two discriminants and the development of the Australian proximity climatology, we chose to use the Allen et al.'s (2011) discriminants (SEV, SigSEV) as representative of environments conducive to severe thunderstorm development for the climatology.

Table 2. Probability of Detection (POD) and False Alarm Ratio (FAR) for SEV (Equation (2)), and SigSEV (Equation (1)) discriminants and significant severe environments from severe as determined using the Brooks et al. (2003) discriminant
 PODFAR
Severe0.756N/A
Significantly severe0.5830.471
Brooks0.6360.523

4.1 Spatio-frequency comparison – the first 7 years

To test the spatial distribution of the environment climatology, the observational severe thunderstorm database is broken down into seasonal sets of latitude and longitude. To make this comparison, a reanalysis equivalent grid of 150 km is placed over the continent and the number of reports in each bin counted. This gives a spatial and frequency depiction of severe thunderstorm observations that can then be compared to the frequency of SEV environment occurrence determined from ERA-Interim using the environmental discriminants (Figure 5(a)). These reports are used as an effective verification of the performance of the climatological discriminants. In addition, this also permits analysis of the systemic under-reporting of severe thunderstorm occurrence for Australia, particularly where frequency is unknown in regional areas.

image

Figure 5. (a) Spatial distribution of the observed frequency of severe thunderstorm reports per year over the period September 2003 to April 2010. Frequency is based on an ERA-Interim equivalent grid of 150 km placed over the continent and the number of reports in each grid cell. (b) Gaussian smoothed contours of ERA-Interim SEV environments per year for the seven warm seasons September 2003 to April 2010. Frequency is based on the number of days per year where severe thunderstorm environments exceed the SEV discriminant (Equation (2)) for pseudo-soundings at each gridpoint of the ERA-Interim reanalysis for Australia. Favourable environments must also satisfy the conditions: MLCIN < 25 J kg−1, S06 > 7.5 m s−1, 700–500 hPa Lapse rate > 6.5 K km−1, and occur over land.

Download figure to PowerPoint

The frequency of reports closely mimics the distribution of rural centres of inland Australia (Figure 5(a)). On the coastal fringe, particularly around the metropolitan centres of the eastern seaboard, the influence of Australia's mainly coastal population on the reporting of severe thunderstorm phenomena is apparent. Particularly, high densities of reports are found near Brisbane and Sydney, with the largest density contours encompassing the region from just south of Rockhampton on the east coast, to Melbourne in the south. Reports to the west of the Great Dividing Range on the east coast are uncommon apart from the few larger towns and cities to be found there. In comparison, the peak severe environment occurrence extends from west of Sydney, through the Hunter Valley along the coast towards the Northern Rivers, through to the hinterland of Queensland near Mackay.

In contrast to the reports distribution, the climatology of favourable SEV environments extends further inland from southern New South Wales and further to the northwest through the Atherton tableland and west towards the Northern Territory (Figure 5(b)). A second peak is found on the northwest Pilbara coast and southwest of Darwin. This likely coincides with the monsoon season and the warm SSTs surrounding the northern coastline. These environments for northern Australia are likely the result of the poor representation of CIN as noted in 'Severe thunderstorm dataset', similar to the problem noted by BLC03 in extreme southern Texas. In the observed atmosphere, these SEV environments are normally impossible to initiate, with CIN through the period leading up to the wet season often 200 J kg−1 or more. This results in a potential environment that is unlikely to be broken at these latitudes due to a lack of strong lifting to overcome the CIN. Intriguingly, however, this area is identified to be associated with satellite-derived hail observations (Cecil and Blankenship, 2012), suggesting that there may be some degree of reality to some of these environments. Inland from Perth through the southwestern part of the continent, a number of SEV environments are identified that do not appear in the reports. The relatively sparse population of the Western Australian wheatbelt and a lack of observers means that there is likely under-reporting in this area. This problem, however, is not confined to inland Western Australia; a systemic under-reporting of severe thunderstorms is present over much of the continent. Some of the disparity between SEV environments and reports exists due to environments occurring that may never initiate or realize a thunderstorm. One of the few locations that have a similar number of severe thunderstorm reports and SEV environments is southern Victoria, and particularly the area around Melbourne. We hypothesize that this results from the relatively dense population in this state relative to others in Australia. A second factor is the strong lifting mechanisms available at the higher latitude increasing the likelihood of severe thunderstorms being realized when the environment occurs. Tasmania does seem to cause some problems with this explanation, with a number of thunderstorms environments that are not matched by regular reports. This may be a result of the relatively low population density across most of the island relative to where the favourable SEV environments occur. Six reports are from Tasmania, and three are instances of significant severe storms, which would suggest that when environments occur there is a tendency for storm events to result, similar to the observations for Melbourne.

The seasonal cycle over the EAReg was compared between reports and the reanalysis SEV environments normalized by the number of gridpoints (Figure 6(a)). The pattern is well replicated particularly through the peak season (October to January). The early and late season values tend to be somewhat over-represented by the reanalysis in proportion, perhaps as a result of the cases where CAPE magnitude or the potential for capping to be overcome are overly optimistic. Clearly, reports over the EAReg are not uniform in distribution, and the reports from a particular system that produces SEV environments may span more than a single reanalysis gridpoint and be distributed over a much larger area. This explains the difference in magnitude for the maximum seasonal frequency from reports as compared to environments over the EAReg, where single or multiple large-scale environments produce multiple reports over the region.

image

Figure 6. (a) Monthly inter-annual variability of severe thunderstorm reports (grey; 1163 reports) as compared to ERA-Interim SEV environments (black) for the EAReg over the period 2003–2010. The outer bounds correspond to the maximum and minimum number of occurrences in any given month, with the mean indicated by the horizontal bar. The frequency of ERA-Interim environments for the respective months is normalized by the number of land gridpoints over the region (68) multiplied by the number of longitudinal points (9). (b) Temporal distribution of severe thunderstorm reports over the EAReg by nearest hour in local time for the period March 2003 to April 2010 (black line and small circles) and equivalent temporal distribution of the total SEV environments over Australia produced from the ERA-Interim reanalysis at the respective analysis times (large black circles). Times correspond to daylight eastern savings time, or equivalently 1800,0000,0600,1200UTC, respectively.

Download figure to PowerPoint

The diurnal distribution of reports (Figure 6(b)) indicates that severe thunderstorms are observed between 3 pm and 9 pm local time, with the greatest peak around 5 pm, or just after the peak of the diurnal cycle of surface temperature. For the reanalysis, as there are only four available analysis times, we see a similar distribution (with the times corresponding to east coast daylight savings local time that apply during the summer) with a peak for the 5 pm/0600UTC analysis. However, given CIN less than 25 J kg−1 is a condition for an SEV environment to be considered favourable, it is clear that a larger number of environments are occurring at the 11 am and 11 pm (0000,1200UTC) analysis timeframes. This would suggest an apparent lack of CIN, particularly for the morning sounding (11 am/0000UTC). For the purposes of this analysis, a single environment is determined from there being at least one severe favourable thunderstorm pseudo-sounding analysis time in any particular day (i.e. if there are four analysis soundings with a favourable environment only one is counted for that gridpoint on the day).

5. Severe thunderstorm environments 1979–2011

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Datasets and methodology
  5. 3. Verification against rawinsonde profiles
  6. 4. Proximity climatology
  7. 5. Severe thunderstorm environments 1979–2011
  8. 6. Discussion
  9. Acknowledgements
  10. References

5.1. Mean CAPE and S06 distribution

An important aspect of severe thunderstorm environments is the distribution of mean fields of CAPE and shear and their evolution over the course of the convective season. To examine this, mean monthly CAPE and S06 are combined into three composite periods based on pattern similarity: Early (October, November), Mid (December, January) and Late (February, March). The Early season is dominated by high mean S06, with the sub-tropical jet present over southern Australia (Figure 7(d)). A small area of higher mean shear is identified to be pushing along the coastal fringes and over the Great Dividing Range, likely a result of regular troughing over the area and the continental/oceanic air mass interaction. By the Mid season, the highest mean S06 pattern has shifted south in conjunction with the pole-ward shift of synoptic systems during the Austral summer (Figure 7(e)). Shear over this part of the season remains less than 15 m s−1 over much of the EAReg and southeast. However, over Victoria this value tends to remain longer, finally disappearing during the Late season when values exceeding this become rare over the entire continent.

image

Figure 7. Smoothed mean bi-monthly environmental CAPE for 0600UTC over the Australian region for (a) October and November (Early), (b) December and January (Mid) and (c) February and March (Late). (d)–(f), as for respective mean CAPE periods, except for 0–6 km Bulk Vertical Wind Shear (S06). Minimum CAPE contour is 25 J kg−1, and contours are 0–100 at intervals of 25, 100, 200–1000 at intervals of 200.

Download figure to PowerPoint

In contrast, the mean environmental CAPE peaks in tropical regions associated with the moisture field, and shifts southwards over the course of the warm season with available moisture (Figure 7(a)–(c)). This is particularly prevalent over both coasts, where advection of moisture pole-ward via the passage of troughs is common. CAPE tends to stay higher along the eastern coastal fringe than the west, which may be related to the influence of the continental divide in effect shielding the dry surface flow from the inland areas. CAPE remains limited for the southeast of Australia during the Early season, with values typically less than 100 J kg−1. During November, a southwest surge of mean CAPE occurs in Western Australia, and the mean CAPE increases inland over Queensland and extends west of the divide. As the cycle reaches Mid-season, peak mean CAPE extends over the north of Australia, linked to the arrival of moisture in the Northern Monsoon, and extends down the east and west coasts of the continent. In the Late season, mean CAPE over south central Australia diminishes associated with retreating moisture, while CAPE over southeast Queensland begins to abate during March.

5.2. Severe thunderstorm environment frequency

Considering the 7-year climatology in the context of the longer period, we find that the period 2003–2010 was unusually active on the east coast. In particular, more favourable SEV environments occurred in the southeast Queensland and northeast New South Wales near coastal areas compared with the period 1979–2011 (Figure 8(a)). Also noticeably different from the 7-year climatology, there is a greater occurrence of environments further inland on average for the period 1979–2011. This difference is pronounced both through more SEV environments over the western half of the continent, and environments occurring further inland through the EAReg. A similar decrease in frequency over the east coast in particular is found for the SigSEV population with little variation to the distribution between the two periods (not shown). The period 2003–2010 was dominated by long-term drought conditions in eastern Australia, particularly the areas inland of the Great Dividing Range, and this would seem to have produced far fewer environments inland, but a greater number in the coastal fringe. With the addition of a number of wet years and a longer period picture of the environments, there is a greater number of inland SEV environments.

image

Figure 8. (a) Smooth distribution of favourable SEV environments per year for warm seasons over the period September 1979 to April 2011. Frequency is based on the number of days per season where SEV environments exceed the severe discriminant (Equation (2)) for pseudo-soundings at each gridpoint of the ERA-Interim reanalysis for Australia. Favourable environments must also satisfy the conditions: MLCIN < 25 J kg−1, S06 > 7.5 m s−1, 700–500 hPa Lapse rate > 6.5 K km−1, and occur over land. (b) as (a) except for environments which exceed the SigSEV discriminant (Equation (1)). (c) As (a) except contours of MLCAPE environments exceeding 1000 J kg−1 in the presence of at least 5 m s−1 S06. (d) As (a) except contours of S06 exceeding 15 m s−1 in the presence of MLCAPE of at least 10 J kg−1.

Download figure to PowerPoint

SigSEV environments follow a similar distribution to the SEV climatology, albeit at a lower frequency, particularly in the southern latitudes, and not at all in Tasmania (Figure 8(b)). These environments occur with higher prevalence in the band from west of Sydney towards Canberra, and extending along the eastern coastline to near Cairns, as far inland as Emerald, and as far west as Moree and Dubbo. The highest number of environments in this area is through the Northern Rivers and southeast Queensland, peaking near Gladstone, correlating well with the hail estimates of Cecil and Blankenship (2012). A second area of these SigSEV environments in the northwest of Australia occurs in a similar location to SEV environments, with a similar peak frequency to those observed in southeast Queensland. The reality of these SigSEV environments is far from certain, with strongly sheared environments uncommon at that latitude, and CAPE dominating the formative mechanism (Figure 8(c) and (d)). At least in part, some of this population is a result of the aforementioned issues with CIN in the ERA-I and the south Texas problem noted by BLC03 when using reanalysis data, or potentially un-observed severe instances (Cecil and Blankenship 2012). While climatologies of thunderstorm occurrence derived using lightning suggest high frequency in this area (Kuleshov et al., 2002), it is difficult to identify significant severe thunderstorms without an extensive database of reports for the area. Thus, the nature of the resultant storms and their relationship to environments in the northwest of Australia remains somewhat unknown at this stage.

To ascertain the origin of the high climatological peaks in the severe and significant severe distributions, the spatial distributions of high MLCAPE and high S06 environments was considered separately (Figure 8(c) and (d)). These conditional environments allow the estimation of the contribution from the individual ingredients to the occurrence of high-end (significant) severe environments. MLCAPE environments exceeding 1000 J kg−1 (conditional on S06 being greater than 5 m s−1) are found to be exceedingly rare south of 30°S on an annual basis, with a slight extension down the east coast. The peak occurrence of high CAPE environments is found in the northwest of Australia, with another high environmental frequency found throughout eastern Queensland. This would suggest that CAPE is the dominant factor in the relationship through these areas.

Some questions remain regarding the realization of storms from these environments in the northern parts of Australia, particularly whether the CIN is sufficiently low for initiation to occur. S06 plays a far stronger role throughout eastern Australia and through the southwestern and southeast corners of Australia, with the distribution of these environments following more closely the SEV and SigSEV distributions. The greatest frequency of environments with S06 in excess of 15 m s−1 (in the presence of at least 10 J kg−1 MLCAPE) occurs along the east coast, stretching from 20°S southwards to Victoria. A second region extends across the southwest, which is coincident with the favourable severe environments that stretch through Western Australia. These distributions suggest that the frequency of the individual ingredient's contribution to the environmental product vary substantially over the continent, highlighting the importance of considering covariates that include both S06 and CAPE.

5.3. Bi-monthly severe thunderstorm environments

The mean bi-monthly distribution of SEV environments was considered to establish the pattern variations over the course of the warm season. During September, environments tend to be dominated by low CAPE and are highly sheared, particularly over Victoria and along the coastal fringe/dividing range in the east (not shown). There is also an area of environments that occur over southwest Western Australia associated with the passage of the strong frontal systems in conjunction with low CAPE environments. In the Early season, the southward extent of moisture along with seasonal warming begins to extend from northwest Australia across the top end and along the Great Dividing Range and east coast (Figure 9(a)). Generally, less than two environments occur in these regions during the Early part of the season.

image

Figure 9. Smooth distribution of mean bi-monthly SEV environments over the period 1979–2011 for (a) October and November (Early), (b) December and January (Mid) and (c) February and March (Late). Frequency is based on the number of days per month where environments exceed the SEV discriminant (Equation (2)) for pseudo-soundings at each gridpoint of the ERA-Interim reanalysis for Australia. Favourable environments must also satisfy the conditions MLCIN less than 25 J kg−1, S06 greater than 7.5 m s−1, 700–500 hPa Lapse rate greater than 6.5 K km−1 and occur over land.

Download figure to PowerPoint

With troughing over inland Western Australia and seasonal moisture, environments become more frequent over the inland area, and these continue from the Mid season into February (Figure 9(b) and (c)). As the season progresses, these environments tend to be more commonly found in the coastal regions of Western Australia owing to the dry air present in desert areas within the inland regime. Peaks in northwestern Australia appear to span the summer months, though to some extent this is likely to be related to the high seasonal CAPE from available moisture, combined with modest shear. In particular, this is noticeable for December through March with mean shear generally less than 5 m s−1. This is coincident with CAPE likely not being controlled by the aforementioned issues, with under-representation of capping and CIN due to vertical resolution.

Over the far southeast the season tends to peak earlier, from Early to Mid-season with frequency reducing rapidly thereafter. Over southeast Queensland, the season begins to increase to ten environments per season from November through January. This tends to suggest a significantly overestimated frequency of environment/report yield, where the climatology of reports tends to remain below this value. This again is possibly related to the presence of environmental mixed air from the inland arid regions, as well as capping inversions not being adequately detected. There is also likely some relationship to a lack of initiating features for thunderstorm environments during this part of the year for the region outside of orographic inducement. In this region, the Early season tends to be dominated by the shear, while the Late season is more dependent on CAPE. This suggests that the lack of severe observations during the late part of the season in southeast Queensland has more to do with a lack of organized severe storms due to accompanying weak shear, in conjunction with the decreased frequency of initiation.

5.4. Influence of ENSO and inter-annual variability

The difference between the 2003–2010 climatology and the 1979–2011 climatology in terms of annual mean environments suggests an inter-annual variation in severe thunderstorm occurrence over Australia. Regarding frequency, there seems little correspondence between the ENSO cycle and the overall occurrence of environments (Figure 10(a)). For example, 2010/2011 was one of the strongest La Niña events on record and yet the number of events in 1985/1986, which had a neutral ENSO pattern, eclipses this environment total. There is also no certainty that neutral conditions will entail a greater number of SEV environments, though there appears to be some relation to SigSEV environment occurrence with the two climatological peaks occurring in 1985/1986 and 1989/1990. Even when comparing the strongest La Niña and El Niño events, the differences over the Australian region vary, with the strong El Niño event of 1982/1983 producing as many events as all but the strongest La Niña. There is substantial inter-annual variability over the period, however, unusually there is very little inter-annual variability between 1992/1993 and 1999/2000. After this period, we see a strong oscillation in the frequency between 2001/2002 and 2010/2011, culminating in the secondary climatology peak.

image

Figure 10. Inter-annual variability of the total number of environment days per season for warm seasons 1979–2011. (a) Total number of environments in Australia per warm season exceeding the SEV discriminant (Equation (2), top line), and number of environments exceeding the SigSEV discriminant (Equation (1), bottom line). Significant ENSO events (El Niño, La Niña, Neutral) over the period are shown. (b) As (a) except total number of environments over the EAReg. Total number of possible environments per season for land gridpoints over the Australian Region is 79 860, while over the EAReg 21780 are possible.

Download figure to PowerPoint

Considering the total occurrence over one of the areas most influenced by the ENSO in eastern Australia (the EAReg, Figure 1(b)), there seems to be even less differentiation between some of the strong events. For example, the 1997/1998 strong El Niño event produced a similar number of SEV and SigSEV environments to the 1998/1999 strong La Niña event. Instrumental records suggest that the 1997/1998 El Niño was unusually wet for eastern Australia. However, if the four other significant El Niño events over the climatological period are considered, each case over the EAReg is associated with four of the lowest severe thunderstorm environment frequencies (1982/1983, 1991/1992, 1994/1995, 2002/2003). This is particularly noticeable for the weak El Niño in the 2002/2003 season, which coincided with and contributed to incredibly dry conditions over eastern Australia.

In contrast, three of the highest annual SEV environment totals correspond to strong La Niña events (1998/1999, 2008/2009, 2010/2011). This suggests that Neutral or La Niña events tend to produce the greatest number of SEV environments for the EAReg, while El Niño conditions produce the fewest. This is in contrast to the observations in the Brisbane area, where Neutral or El Niño are generally considered the most favourable for severe thunderstorm occurrence (Yeo, 2005). If correlations to an SOI composite are considered, the only significant correlation between the inter-annual variation of SEV environments and ENSO is for a June to December average of monthly SOI with a significant positive correlation (0.36) to the frequency of SEV environments over the EAReg. Despite numerous tests of smaller sub-regions involving the east coast and southern sections of the EAReg, no significant correlations were identified for any SOI composite.

With little evidence for modulating the inter-annual frequency over Australia, the potential for ENSO to modify the spatial occurrence of SEV environments is examined. Considering the spatial distribution of 2 three-season composite sets for La Niña and El Niño years, there are some regions with significant departures from the mean pattern in both cases (Figure 11). The significance of these differences from the mean sample was examined using a bootstrapping technique. Considering random 3-year combinations with replacement of the 32 season samples, the 90th (hereafter significant increase) and 10thh percentile (hereafter significant decrease) were determined. The differences in frequency of the three season La Niña (1988/1999, 1998/1999, 2010/2011) and El Niño (1982/1983, 1991/1992, 1994/1995) composites were then analysed for significance (Figure 11). Along the east coast during El Niño periods, the distribution of environments is shifted north as the moisture decreases in quality through the southeastern corner of the continent. This is reflected by significant decreases over a number of gridpoints for much of the New South Wales coast and along the Great Dividing Range. Further north, through southeast Queensland there is an increase in the frequency of SEV environments compared to a normal year, however, the increase is not significant. Central Australia remains relatively dry (although the frequency of environments in this area is relatively small on average), while there is also a non-significant increase in environments in the northern parts of the continent. For much of Western Australia during El Niño years, the frequency of environments significantly increases, extending from near Perth as far east as the border, through much of the agricultural wheatbelt to approximately 23°S. The recognized tele-connection in the southwest corner of Australia and much of Western Australia during El Niño (Allan et al., 1996) results in an enhanced amount of moisture over this area, and this results in a substantial increase in the number of favourable SEV environments.

image

Figure 11. (a) Composite of the number of SEV environments exceeding the severe discriminant (Equation (2)) for the three El Niño warm seasons determined for Southern Oscillation Index < −8 (1982/1983, 1991/1992, 1994/1995). Stippling indicates areas for which the number of environments in the composite exceeds the 90th percentile. Lines indicate gridpoints at which the average for the composite is lower than the 10th percentile. (b) As (a) except for three La Niña warm seasons determined for Southern Oscillation Index > +8 (1988/1999, 1998/1999, 2010/2011).

Download figure to PowerPoint

The abundant available moisture during La Niña years results in a larger area of favourable environments extending further inland and south along the east coast of the continent. The greatest influence on SEV environments occurs through southern New South Wales and northern Victoria, with significant increases for the area near Melbourne, across Gipsland and extending north towards Echuca, Hay and east to Wagga Wagga. The mechanism in this area is likely the increase in rainfall to the north increasing the moisture flux over the area and reducing the infiltration of dry continental air that commonly prevails in the southeast. When the trough systems arrive over the area, they are met with a favourable boundary layer environment to promote severe thunderstorm development. In southeast Queensland and northeast New South Wales, there is a lower frequency of SEV environments, with a shift in the east coast peak further north to extend between Cairns and Bundaberg, and further inland than observed during El Niño years. There is also a notional significant increase across Cape York, however, this is probably related to the increased moisture without the equivalent rise in CIN. Southwest Australia varies little from the climatological mean during La Niña years.

6. Discussion

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Datasets and methodology
  5. 3. Verification against rawinsonde profiles
  6. 4. Proximity climatology
  7. 5. Severe thunderstorm environments 1979–2011
  8. 6. Discussion
  9. Acknowledgements
  10. References

The temporal differences between the total number of SEV and SigSEV environments in years with similar ENSO conditions would suggest that Australia's severe thunderstorm environment is influenced by other climatological patterns over the region. While ENSO conditions clearly modulate the spatial distribution and frequency of environments to some extent, the availability of moisture seems to be the most important control on the occurrence of severe thunderstorm environments across the continent. This is particularly true through the inland and southern portions of the continent. Other factors also influence the synoptic pattern that result in the instability response, and vertical wind shear required to produce conditions favourable to severe thunderstorms. A second plausible factor influencing the source for the moisture over the continent would appear to be the SSTs around northern Australia. With the moist air mass over the continent originating from the Coral Sea, the Arafura Sea and along the northwest shelf, the relative temperature of this water also likely contributes to the inter-annual variability of severe thunderstorm environments. It is therefore not surprising that a correlation exists between Northern Australian SST anomalies and the frequency of SEV environments for December through February. However, the negative correlation that is present indicates a different mechanism than expected (−0.42). This suggests that, with higher SSTs over the waters north of Australia during the monsoonal phase there tends to be a decrease in the frequency of environments. This could partially be explained by increased cloud cover, or too much available moisture resulting in an excess of convection, none of which are given the opportunity to severe before encountering competition. This relative warmth of SSTs would also suggest an influence from the Indian Ocean dipole, particularly over the northwestern shelf. As the frequency of mid-latitude storm systems with associated troughs and air mass fluxes which provide the necessary lift and ingredients for severe convective development are other factors, it is not surprising that ENSO, at least on the continental scale, cannot entirely explain the incidence of SEV or SigSEV environments.

Within this article, the first reanalysis-derived climatology for Australian severe thunderstorm environments for the period 1979–2011 has been presented. While it is clear that this environmental climatology is unable to offer a definitive occurrence of severe thunderstorm events over the continent, it does offer a detailed picture of the annual pattern, seasonal cycle and inter-annual variability of the environments and parameters that are favourable for the occurrence of severe thunderstorms. The climatology also points strongly to the potential for model environmental data to be used to estimate how favourable a region may be to severe thunderstorm events without having an extensive reports dataset, as noted by BLC03.

It has also been identified via comparison to rawinsonde-derived profiles that reanalysis data seem to provide a better approach to estimating the occurrence of severe thunderstorm environments than the operational forecast model used previously. However, it is clear that the relationships identifying the environments associated with severe convection require further attention and analysis using a more expansive observational dataset over a longer period. Thus far, the discriminant relationships have only been considered for a relatively small number of reports for both Australia and the United States, (BLC03, Allen et al., 2011) over periods that may contain large variation in climatic patterns that are different from the climatological norm. As can be observed from both this work and others internationally (Kaltenböck et al., 2009; Gensini and Ashley, 2011), there is merit in considering how well a discriminant applies to a reanalysis dataset prior to considering occurrence of SEV or SigSEV environments. The tailoring of the discriminants to suit any potential environmental bias in the dataset is essential, particularly if any application is to be made to climate models or other datasets with known difficulties in considering these environments.

Little to no trend can be identified for SEV or SigSEV environments over Australia during the past 32 years, with a high degree of inter-annual variability in environments precluding any solid result. Without a longer climatological period, it remains difficult to quantify any climatological trend in either MLCAPE or S06, or to consider a possible seasonal shift in climatological frequency. The same is also true for considering the relationship to ENSO, where a larger sample of events may reveal a more concrete link. While SEV environments extend over most of the continent, the mechanisms that support them vary depending on latitude, with CAPE producing a greater number of environments north of 23°S, and wind shear being the dominant ingredient to produce severe and significant severe thunderstorms over the southeast, southwest and through the climatological peak along the east coast.

The success in applying this climatology suggests a potential for the application of environmental discriminants to a wide variety of datasets for different parts of the globe. Clearly, while attention needs to be paid to the observational/environmental relationship, this presents an opportunity to generate a picture of potential severe thunderstorm occurrence over areas that are sparsely populated or may not have high quality observations (e.g. South Africa). However, it is well known from a severe weather forecasting point of view that no magic bullet exists to discriminate perfectly between severe and non-severe thunderstorm environments (Doswell and Schultz, 2006), and thus any such analysis will always be subject to a probabilistic interpretation rather than directly being able to quantify the actual occurrence of severe thunderstorms. The potential to produce extended climatologies for situations where there was a lack of observations suggests that investigation using other historical reanalyses, or future simulations using regional and global climate models is a realistic goal when considering severe thunderstorm environments.

Acknowledgements

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Datasets and methodology
  5. 3. Verification against rawinsonde profiles
  6. 4. Proximity climatology
  7. 5. Severe thunderstorm environments 1979–2011
  8. 6. Discussion
  9. Acknowledgements
  10. References

The authors are grateful for the contributions and suggestions of the reviewers in improving the final manuscript. We also appreciate the suggestions and comments provided by Dr Harold Brooks and his input and discussions during this work. We are also grateful to the ECMWF for providing the reanalysis data used in this study. This research was supported in part by funding from the Australian Research Council Centre of Excellence for Climate System Science (grant CE110001028).

References

  1. Top of page
  2. ABSTRACT
  3. 1. Introduction
  4. 2. Datasets and methodology
  5. 3. Verification against rawinsonde profiles
  6. 4. Proximity climatology
  7. 5. Severe thunderstorm environments 1979–2011
  8. 6. Discussion
  9. Acknowledgements
  10. References
  • Allan RJ, Lindesay J, Parker D. 1996. El Niño, Southern Oscillation and Climatic Variability. CSIRO Publishing: Collingwood, Australia; 405.
  • Allen JT. 2012. Supercell storms: Melbourne's white Christmas 2011. Bulletin of the Australian Meteorological and Oceanographic Society 25: 4751.
  • Allen JT, Pezza AB, Black MT. 2010. Explosive cyclogenesis: a global climatology comparing multiple reanalyses. Journal of Climate 23: 64686484. DOI: 10.1175/2010JCLI3437.1
  • Allen J, Karoly D, Mills G. 2011. A severe thunderstorm climatology for Australia and associated thunderstorm environments. Australian Meteorological Oceanographic Journal 61: 143158.
  • Australian Bureau of Statistics. 2001. Census of Population and Housing: Population Growth and Distribution, Australia, cat. no. 2035.0. http://www.abs.gov.au/AUSSTATS/abs@.nsf/Lookup/2035.0Main+Features12001?OpenDocument (Accessed 21 January 2012).
  • Betts AK, Hong SY, Pan HL. 1996. Comparison of NCEP – NCAR Reanalysis with 1987FIFE Data. Monthly Weather Review 124: 14801498.
  • Bromwich DH, Fogt RL, Hodges KI, Walsh JE. 2007. A tropospheric assessment of the ERA-40, NCEP, and JRA-25 global reanalyses in the polar regions. Journal of Geophysical Research 112: D10111. DOI: 10.1029/2006JD007859
  • Brooks HE. 2009. Proximity soundings for severe convection for Europe and the United States from reanalysis data. Atmospheric Research 93: 546553. DOI: 10/1016/j.atmosres.2008.10.005
  • Brooks HE. 2012. Severe thunderstorms and climate change. Atmospheric Research. 10.10.1016/j.atmosres.2012.04.002. in press.
  • Brooks H, Dotzek N. 2006. The spatial distribution of severe convective storms and an analysis of their secular changes. In Chapter 8 Climate Extremes and Society. Cambridge University Press: Cambridge, UK, 24.
  • Brooks HJ, Lee J, Craven J. 2003. The spatial distribution of severe thunderstorm and tornado environments from global reanalysis data. Atmospheric Research 67: 7394.
  • Brooks HE, Anderson AR, Riemann K, Ebbers I, Flachs H. 2007. Climatological aspects of convective parameters from the NCAR / NCEP reanalysis. Atmospheric Research 83: 294305. DOI: 10.1016/j.atmosres.2005.08.005
  • Cecil DJ, Blankenship CB. 2012. Toward a global climatology of severe hailstorms as estimated by satellite passive microwave imagers. Journal of Climate 25: 687703.
  • Dee DP, Uppala SM, Simmons AJ, Berrisford P, Poli P, Kobayashi  S, Andrae U, et al. 2011. The ERA-Interim reanalysis: configuration and performance of the data assimilation system. Quarterly Journal of the Royal Meteorological Society 137: 553597. DOI: 10.1002/qj.828
  • Doswell C. 1987. The distinction between large-scale and mesocale contribution to severe convection: a case study example. Weather and Forecasting 2: 316.
  • Doswell C. 2001. Chapter 1: Severe convective storms – an overview. Severe Convective Storms, Meteorological Monograph, American Meteorological Society 50: 126.
  • Doswell CA, Schultz DM. 2006. On the use of indices and parameters in forecasting severe storms. Electronic Journal of Severe Storms Meteorology 1: 114.
  • Doswell CA III, Brooks HE, Maddox RA. 1996. Flash flood forecasting: an ingredients-based methodology. Weather and Forecasting 11: 560581.
  • Geerts B. 2001. Estimating Downburst-Related Maximum Surface Wind Speeds by Means of Proximity Soundings in New South Wales, Australia. Weather and Forecasting 16: 261269.
  • Gensini VA, Ashley WS. 2011. Climatology of potentially severe convective environments from North American regional reanalysis. Electronic Journal of Severe Storms Meteorology 6: 140.
  • Griffiths D, Colquhoun J, Batt K, Casinader T. 1993. Severe thunderstorms in New South Wales: climatology and means of assessing the impact of climate change. Climatic Change 25: 369388.
  • Hales Jr J. 1993. Biases in the severe thunderstorm database: ramifications and solutions. Preprints, 13th Conference on Weather Forecasting and Analysis, Vienna, American Meteorological Society, 4.
  • Kaltenböck R, Diendorfer G, Dotzek N. 2009. Evaluation of thunderstorm indices from ECMWF analyses , lightning data and severe storm reports. Atmospheric Research 93: 381396. DOI: 10.1016/j.atmosres.2008.11.005
  • Kelly D, Schaefer J, Doswell C III. 1985. Climatology of non-tornadic severe thunderstorm events in the United States. Monthly Weather Review 113: 19972014.
  • Kuhnel I. 1998. The use of a multifactor Southern Oscillation Index for the estimation of annual hailstorm frequencies in the Sydney area. International Journal of Climatology 18: 841858. DOI:10.1002/(SICI)1097-0088(19980630)18:8<841::AID-JOC284>3.0.CO;2-I
  • Kuleshov Y, de Hoedt G, Wright W, Brewster A. 2002. Thunderstorm distribution and frequency in Australia. Australian Meteorological Magazine 51: 145154.
  • Lee J. 2002. Tornado Proximity Soundings from the NCEP/NCAR Reanalysis Data. Unpublished MS Thesis: University of Oklahoma; 61.
  • Leigh R, Kuhnel I. 2001. Hailstorm loss modelling and risk assessment in the Sydney region, Australia. Natural Hazards 24: 171185.
  • Niall S, Walsh K. 2005. The impact of climate change on hailstorms in southeastern Australia. International Journal of Climatology 25: 19331952. DOI: 10.1002/joc.1233
  • Potvin C, Elmore K, Weiss S. 2010. Assessing the impacts of proximity sounding criteria on the climatology of significant tornado environments. Weather and Forecasting 25: 921930.
  • Puri K, Dietachmayer G, Mills G, Davidson N, Bowen R, Logan L. 1998. The new BMRC limited area prediction system, LAPS. Australian Meteorological Magazine 47: 203223.
  • Rasmussen EN, Blanchard DO. 1998. A baseline climatology of sounding-derived supercell and tornado forecast parameters. Weather and Forecasting 13: 11481164. DOI: 10.1175/1520-0434(1998)013<1148:ABCOSD>2.0.CO;2
  • Rasmusson EM, Wallace JM. 1983. Meteorological aspects of the El Nino/southern oscillation. Science 222: 11951202.
  • Ropelewski CF, Halpert MS. 1987. Global and regional scale precipitation patterns associated with the El Niño/Southern Oscillation. Monthly Weather Review 115: 16061626.
  • Ryan CJ. 1992. Dynamical Classification of Australian Thunderstorms. BOM Meteorology Study, No. 40, Monash University, 61.
  • Schuster S, Blong R, Speer M. 2005. A hail climatology of the greater Sydney area and New South Wales, Australia. International Journal of Climatolology 25: 16331650.
  • Tailleux R, Grandpeix J. 2001. A generalized Convective Inhibition Energy. Quarterly Journal of the Royal Meteorological Society 30: 120.
  • Thompson R, Mead C, Edwards R. 2007. Effective storm-relative helicity and bulk shear in supercell thunderstorm environments. Weather and Forecasting 22: 102115.
  • Yeo C. 2005. Severe thunderstorms in the Brisbane region and a relationship to the El Niño Southern Oscillation. Australian Meteorological Magazine 54: 197202.
  • Yeo S, Leigh R, Kuhnel I. 1999. The April 1999 Sydney Hailstorm. Australian Journal of Emergency Management 14: 2325.
  • Zwiers F, Kharin V. 1998. Changes in the extremes of the climate simulated by CCC GCM2 under CO2 doubling. Journal of Climate 11: 22002222.