SEARCH

SEARCH BY CITATION

Keywords:

  • heavy precipitation;
  • convergence line;
  • sea breeze;
  • Unified Model

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Case-study: 21 July 2010
  5. 3. Numerical model and experiment design
  6. 4. Simulation results
  7. 5. Summary and discussion
  8. Acknowledgements
  9. References

An investigation is presented of a quasi-stationary convective system (QSCS) which occurred over the UK Southwest Peninsula on 21 July 2010. This system was remarkably similar in its location and structure to one which caused devastating flash flooding in the coastal village of Boscastle, Cornwall on 16 August 2004. Both events were characterised by deep southwesterly flow and saw the repeated development of convective cells along the west coast of the Southwest Peninsula. However, in the 2010 case, rainfall accumulations were around four times smaller and no flooding was recorded. The more extreme nature of the Boscastle case is shown to be related to three factors: (1) higher rain rates, associated with a warmer and moister tropospheric column and deeper convective clouds; (2) a more stationary system, due to slower evolution of the large-scale flow; and (3) distribution of the heaviest precipitation over fewer river catchments.

A numerical simulation of the July 2010 event was performed using a 1.5 km grid length configuration of the Met Office Unified Model. This reveals that convection was repeatedly initiated through lifting along a quasi-stationary boundary-layer convergence line. Sensitivity tests are used to show that this convergence line was a sea-breeze front which temporarily stalled along the coastline due to the retarding influence of an offshore-directed background wind component. Several deficiencies are noted in the 1.5 km model's representation of the storm system, including delayed convective initiation; however, significant improvements are observed when the grid length is reduced to 500 m. These result in part from an improved representation of the convergence line, which enhances the associated low-level ascent, allowing air parcels to more readily reach their level of free convection. The implications of this finding for forecasting convective precipitation are discussed.


1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Case-study: 21 July 2010
  5. 3. Numerical model and experiment design
  6. 4. Simulation results
  7. 5. Summary and discussion
  8. Acknowledgements
  9. References

The so-called First Law of Quantitative Precipitation Forecasting, attributed to C. F. Chappell, states that ‘the heaviest precipitation occurs where the rainfall rate is the highest for the longest time’ (Doswell et al., 1996). Ordinary single-cell thunderstorms may produce heavy rainfall, but they rarely last long enough to give significant localised accumulations. On the other hand, if multiple convective cells repeatedly pass over the same area in rapid succession, extreme rainfall totals can occur. This process may be associated with ‘back-building’ mesoscale convective systems (MCSs), in which the upstream development of new cells matches the system's downstream translation speed (Chappell 1986; Schumacher and Johnson 2005). It may also be associated with repeated topographically forced initiation of cells over the same location (e.g. Maddox et al., 1978). Both situations result in a quasi-stationary convective system (QSCS) which can locally produce extreme rainfall accumulations.11 Depending on the intensity of the rain produced, the duration of the system, the distribution of rainfall over different drainage basins, and the characteristics of these drainage basins (e.g. antecedent moisture, slope, soil porosity, vegetation cover), flash flooding may occur (Davis 2001). Worldwide, many severe flash floods have been attributed to QSCSs attributed to QSCSs (e.g. Maddox et al., 1978; Shepherd and Colquhoun et al., 1985; Petersen et al., 1999; Romero et al., 2000; Golding et al., 2005; Ducrocq et al., 2008; Zhang and Zhang, 2012).

For a storm system to become quasi-stationary, local conditions must remain conducive to the development of deep, moist convection for an extended period of time. Thus, there must be a continuous supply of moisture and instability, and a persistent mechanism to lift parcels of air to their level of free convection (LFC). Additionally, the wind direction in the cloud layer must remain roughly constant so that cells maintain a consistent track over time. In cases where convective initiation is related to local topography (e.g. forced orographic lifting), a steady low-level wind may also ensure that new cells repeatedly form in the same location. Slow evolution of the large-scale flow can therefore be favourable to the development of QSCSs. However, mesoscale and storm-scale processes often govern the exact location and timing of convective initiation (Doswell 1987). Thus, understanding the mechanism by which a particular QSCS forms requires datasets with high spatial and temporal resolution. Investigations of these events have therefore typically relied on remotely sensed observations (from ground-based radar and satellites) and numerical simulations.

The existing body of work on QSCSs is dominated by case-studies of extreme flash-flood-producing events in the USA (e.g. Maddox et al., 1978; Petersen et al., 1999; Schumacher and Johnson, 2008) and the Mediterranean region (e.g. Romero et al., 2000; Ducrocq et al., 2008; Miglietta and Regano et al., 2008). Comparatively few studies have investigated these systems in the UK. One oft-cited piece of work is that of Golding et al. (2005; hereinafter GCM05), which examined the ‘Boscastle storm’ of 16 August 2004. This QSCS formed along and just inland of the west coast of the UK Southwest Peninsula (Figure 1), and remained stationary for several hours, resulting in rainfall totals which exceeded 200 mm over a narrow swath of land (Burt 2005). The steep and rocky local catchments rapidly channelled this water downstream, leading to devastating flooding in the coastal settlements of Boscastle and Crackington Haven (Figure 1 shows locations).

thumbnail image

Figure 1. Map of the UK Southwest Peninsula showing orography height (m, shading) and locations mentioned in the text. Thick lines mark county boundaries and county names are shown in upper case. Diamonds show the locations of the two weather radars which provide coverage for this region: Predannack (P) and Cobbacombe Cross (CC).

Download figure to PowerPoint

GCM05 investigated the Boscastle case using available observations and numerical simulations with a high-resolution (Δx =1 km) version of the Met Office Unified Model (UM). They found that deep convection was initiated and maintained by a persistent, narrow convergence line which developed along the coastline during the day. Based on the results of sensitivity tests, the authors concluded that this convergence line was ‘a sea-breeze front whose position was determined by a subtle balance between the gradient wind direction, retardation and backing of the wind over land, and differential heating’. They also suggested that the modest instability in this case favoured ‘closely packed storms with weak downdraughts that did not distort the coastal convergence line’. The intensity of the rainfall, which was estimated to have exceeded 500 mm h−1 for brief periods (Burt 2005), appeared to be the result of high tropospheric humidity, sustained by large-scale ascent, which promoted unusually high precipitation efficiencies.

While the Boscastle storm was a rare and extreme event, the development of persistent convergence lines and associated convective showers over the UK Southwest Peninsula is a relatively common occurrence. Monk (1987) and Hand (2005) both noted the tendency for lines of convective cloud, co-located with a well-defined zone of near-surface convergence, to develop along and downwind of the Southwest Peninsula in southwesterly flow. These features occur most frequently during the spring and summer from around midday to early evening (i.e. when the land is warmer than the sea), suggesting that sea-breeze circulations play an important role in their formation. However, other factors, such as differential surface roughness may also be significant, as suggested by GCM05.

In this article, we present an analysis of a QSCS which was remarkably similar to the Boscastle storm in terms of its location and structure, but significantly less severe in terms of its impact. The storm occurred on 21 July 2010 and produced maximum (radar-derived) rain accumulations of 50 mm in 3 h, with no reports of flooding. This case provides an excellent opportunity to investigate the factors which distinguish a severe (i.e. flash-flood-producing) QSCS from a non-severe QSCS, without the complications associated with comparing events in different geographical locations. It also allows us to build on the work of GCM05 and Monk (1987) by examining a range of factors (differential surface heating, differential surface roughness, orography, and convective outflow) which might influence the formation and maintenance of convective lines in the Southwest Peninsula. The latter objective is achieved through a series of simulations using the UK Variable-resolution (UKV) configuration of the UM. Although the horizontal resolution of this model (Δx =1.5 km over the UK) is slightly coarser than that employed by GCM05, the two are sufficiently similar to allow for a meaningful comparison of results.The accuracy of the model simulations is assessed through detailed comparisons with radar-derived surface rainfall data. Furthermore, we investigate the impact of increased horizontal resolution on forecast accuracy via a simulation with 500 m grid spacing.

The rest of this article is organised as follows. Section 2 describes the synoptic setting on 21 July 2010, the evolution of the QSCS, and the resulting precipitation distribution. The differences between this event and the Boscastle case are then discussed. Section 3 describes the UM and the UKV configuration, along with our simulation methodology. In section 4, a control run is first presented and compared to observations. Results from a series of sensitivity experiments are then used to demonstrate the mechanism by which convective cells repeatedly developed in the same location. Following this, results from the 500 m grid-length simulation are examined. Finally, section 5 presents a summary and discussion of our findings.

2. Case-study: 21 July 2010

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Case-study: 21 July 2010
  5. 3. Numerical model and experiment design
  6. 4. Simulation results
  7. 5. Summary and discussion
  8. Acknowledgements
  9. References

We first examine the 21 July 2010 QSCS event in terms the evolution of the large-scale flow, the life-cycle of the convective system, and the resulting precipitation distribution. We then compare each of these aspects with those for the Boscastle case of 16 August 2004.

2.1. Synoptic setting

The synoptic situation over the British Isles at 0600 UTC (0700 British Summer Time, BST) on 21 July 2010 was characterised by a slow-moving low-pressure system at the surface and a cut-off low at upper levels (Figure 2(a)). The centre of the surface low was located over the northeast coast of England, while a secondary, weaker circulation centre was present over southeast Ireland. The Met Office surface analysis for this time (available online at http://www.wetter3.de/Archiv/archiv_ukmet.html) shows a cold front over the east coast of England, a warm front running northeast from the main low centre to Norway, and a trough line extending south from the secondary circulation centre. Over the Southwest Peninsula, the surface flow was from the southwest, i.e. roughly parallel to the western coastline (Figure 2(a)). Quasi-geostrophic forcing in the region was minimal, due to weak cold air advection and cyclonic vorticity advection aloft (not shown). Furthermore, the peninsula was not positioned under any favourable regions for ascent associated with upper-level jet streaks (not shown). As the day progressed, the cyclone and associated cut-off low aloft moved very slowly northeastward. This resulted in veering winds with time over the Southwest Peninsula such that by 1800 UTC (Figure 2(b)), the surface flow over the west coast was approximately zonal.

thumbnail image

Figure 2. Unified Model 12 km grid length analyses for 0600 and 1800 UTC on (a, b) 21 July 2010 and (c, d) 16 August 2004 showing 500 hPa geopotential height (dm, shading), mean sea-level pressure (hPa, contours), and 10 m wind vectors. (a, b) are from the operational North Atlantic and European (NAE) model of 2010, and (c, d) are from the operational Mesoscale Model of 2004.

Download figure to PowerPoint

The radiosonde ascent from Camborne, Cornwall (Figure 1 shows the location) at 1200 UTC on 21 July 2010 is shown as a tephigram in Figure 3(a), with a surface-based pseudo-adiabatic parcel ascent overlaid. The atmosphere at this time was characterised by an absolutely unstable surface layer, with moist, conditionally unstable air below a weak temperature inversion at 700 hPa, and drier, absolutely stable air aloft. The parcel ascent gives convective available potential energy (CAPE, calculated using virtual potential temperature) of 483 J kg−1 and no convective inhibition (CIN), with the lifting condensation level (LCL) and LFC at 938 hPa (590 m) and the level of neutral buoyancy (LNB) at 453 hPa (6.3 km). Note that, due to relatively high temperature and humidity at the surface, the surface-based parcel is by far the most unstable of the sounding. For example, a parcel initalised with the mixed-layer properties over the lowest 100 m has CAPE of only 122 J kg−1 and an LNB at 576 hPa (4.5 km), though again negligible CIN. Thus, we would expect convection to readily develop but remain rather shallow, with cloud tops generally below 6 km.

thumbnail image

Figure 3. Tephigrams showing the 1200 UTC radiosonde ascents from Camborne on (a) 21 July 2010 and (b) 16 August 2004. Bold solid and dashed lines show the temperature and dewpoint temperature profiles respectively. Thin solid lines show surface-based pseudo-adiabatic parcel ascents, with the resulting CAPE distribution shaded grey. The LCL, LFC, and LNB are also indicated. Wind barbs show speed in knots with half barbs, full barbs, and pennants indicating 5, 10, and 50 knots respectively.

Download figure to PowerPoint

The wind profile in Figure 3(a) shows southwesterly flow over the depth of the troposphere, with a density-weighted mean wind speed in the cloud layer of 12 m s−1. The slight unidirectional shear may have reduced the potential for downdraughts to suppress the convective updraughts.

2.2. Storm evolution and rainfall accumulations

In the moist, conditionally unstable flow over southwest England, convection readily developed during the morning of 21 July. High-resolution visible images from Meteosat Second Generation (not shown) reveal the development of shallow cumuli over much of the Southwest Peninsula between 0700 and 0800 UTC. These clouds rapidly deepened and organised into bands (cloud streets) parallel to the prevailing southwesterly flow. Surface rainfall data from the UK 1 km radar composite produced by the Met Office (Harrison et al., 2000; Harrison et al., 2009; Harrison et al., 2012) shows that, between 0830 and 1000 UTC, numerous precipitating cells formed over the peninsula, in particular along and just inland of the west coast (Figure 4(a)). These cells tracked northeast at a speed of around 11 m s−1, consistent with the calculated cloud-layer mean wind. Over the next two hours, the cells increased in size and coverage, forming an almost continuous line of precipitation along the coastline (Figure 4(b)). The rainfall intensity also increased: around 1049 UTC, the tipping-bucket rain gauge at Boscastle briefly recorded rain rates exceeding 150 mm h−1.

thumbnail image

Figure 4. Radar-derived surface rain rates (mm h−1, shading) over the Southwest Peninsula at various times on (a–f) 21 July 2010 and (g–l) 16 August 2004. Orography is contoured with an interval of 200 m. For both events, the radar data have a grid spacing of 1 km; however, prior to 2007, the interpolation method used to produce the composite caused the apparent resolution to degrade with distance from the radars. Harrison et al. (2009) give details.

Download figure to PowerPoint

Following this, the line remained quasi-stationary for 2 h, showing only slight inland movement between 1200 and 1300 UTC. At 1400 UTC (Figure 4(c)), the storm system was composed of two distinct areas. The first was the main convective line, with heavy precipitation extending from Bodmin Moor into Exmoor. The second, to the southwest of the first, consisted of slightly weaker, isolated cells located closer to the coastline. Animations of the rainfall field show that, in general, the southern cells did not merge with the main line but drifted to the west of it and dissipated. Meanwhile, the cells that made up the main line appear to have initiated farther east, along the centre-line of the peninsula. These cells rapidly intensified as they joined with the main line over Bodmin Moor, then continued northeast, weakening as they approached Exmoor and the Bristol Channel. Several of the more intense cells showed a sudden eastward progression as they approached the northeast end of the convective line (two such cells can be seen protruding from the main line in Figure 4(b)). This movement was likely related to the formation and propagation of cold pools beneath the convective cells, with new initiation occurring along the gust front. Maximum cloud-top heights (derived from Meteosat Second Generation imagery) were around 5.5 km at the northeast end of the line, consistent with the parcel analysis in section 2.1.

After 1400 UTC, the convective line began to move inland, starting at its southwest end with the movement gradually spreading northeast (Figure 4(d)). The model simulations to be presented in section 4 indicate that this movement was due to veering low-level winds associated with the gradual eastward progression of the surface cyclone (Figure 2(a, b)). By 1700 UTC (Figure 4(e)), the line had moved away from the west coast, and extended in an arc from St. Austell Bay to Exmoor. Over the next hour, the system rapidly weakened (Figure 4(f)), eventually dissipating around 1900 UTC.

Figure 5(a) shows gauge and radar-derived rainfall accumulations between 1200 and 1500 UTC (i.e. the period for which the most intense portion of the line was stationary) over part of the Southwest Peninsula's west coast. Typical for a quasi-stationary storm, the precipitation area formed an elongated streak along the direction of cell motion, with sharp rainfall gradients either side (particularly, in this case, on the east side). Peak accumulations of around 50 mm occurred on the northwest slopes of Bodmin Moor—not an insignificant amount of rainfall for a 3 h period, particularly over such fast-response catchments. However, there were no reports of flooding and the effect of the rain on river levels was ‘unremarkable’ (Maggie Summerfield, Environment Agency, personal communication). There was a rapid rise in the level of the River Otter shortly after 1500 UTC noted at the Canworthy Water flood warning station (star in Figure 5(a)), but the level attained happens many times a year. The lack of a significant hydrological response can be explained by the distribution of the heaviest rain across river catchments. Figure 5(a) reveals that the highest accumulations occurred close to the headwaters of several rivers, thereby spreading the runoff across multiple drainage basins. In contrast, in the Boscastle case, the heaviest rain fell to the west of the high ground, over just a handful of small coastal catchments (Figure 5(b)).

thumbnail image

Figure 5. Radar-derived rainfall accumulations (mm, shading) over a portion of the Southwest Peninsula for (a) 1200–1500 UTC on 21 July 2010 and (b) 1200–1600 UTC on 16 August 2004. Note again that, while both images have the same gridded resolution (Δx =1 km), the pixels are generally bigger (2 km2) in the 2004 case due to the change in interpolation method mentioned previously. Triangles indicate the maximum radar-derived accumulations in each case, while circles indicate accumulations from Environment Agency tipping-bucket rain gauges. The highest accumulation in each event is shown in bold. In (a), the diamond indicates the accumulation from the Met Office day (0900–0900 UTC) recording gauge at Lower Moor, and the star shows the location of the Environment Agency's Canworthy Water flood warning station. In (b), note that the value for the Lesnewth gauge (170.0 mm) was determined using the corrected data from Burt (2005). Thin lines denote rivers from the Ordinance Survey GR dataset (http://sharegeo.ac.uk/handle/10672/85).

Download figure to PowerPoint

2.3. Comparison to Boscastle case

Figures 2–6 provide a comparison of the 21 July 2010 case and the Boscastle case, in terms of the large-scale situation, the evolution of the convective systems, and the resulting precipitation. A detailed discussion of the Boscastle case is outside the scope of this paper, but can be found in Burt (2005), GCM05, and Golding (2005). Here, we focus on the main similarities and differences between the two cases.

On 16 August 2004, a slow-moving weakly baroclinic low pressure system was again affecting the UK, but in this case, the system was positioned around 1000 km farther west, over the Eastern Atlantic (Figure 2(c, d)). As was the case on 21 July 2010, southwesterly flow was present over the depth of the free troposphere at 1200 UTC (Figure 3(b)), with weak unidirectional shear. However, the large-scale evolution on 16 August 2004 did not act to significantly turn the wind with time; thus, deep, southwesterly flow was maintained throughout the day. Figure 3(b) shows an approximately moist neutral temperature profile with high humidity throughout the troposphere. GCM05 suggested that this deep moist layer was the result of large-scale ascent associated with an upper-level jet streak (not shown). The surface-based parcel ascent in this case shows a much deeper layer of instability extending up to the tropopause at 250 hPa, with CAPE of 1318 J kg−1. Note that this value is considerably larger than that quoted by GCM05 (170 J kg−1) which was determined using a 500 m mixed-layer parcel. However, they noted that some mature cells in the Boscastle convective line did reach the tropopause.

Figure 4 provides a comparison of the evolution of the 21 July 2010 QSCS and the Boscastle QSCS. The similarity in the location and structure of the two systems is striking; however, there are several important differences. First, in the Boscastle case, convection initiated later: the first precipitating cells along the west coast of the Peninsula did not appear until around 1100 UTC. This suggests that initially CIN was too high and/or lifting was too weak for parcels to reach their LFC, giving a longer period for instability to grow through surface heating. Second, the rain rates in the convective cores of the Boscastle system were considerably higher. Comparing Figures 3(a) and (b), we note several features of the thermodynamic environment which may have favoured more intense precipitation on 16 August 2004:

  • Higher specific humidity throughout the troposphere. This indicates the presence of more water vapour available for condensation and precipitation formation (26 mm of precipitable water compared with 20 mm in the 2010 case).

  • Higher relative humidity at mid-levels. This will have reduced the detrimental effects of entrainment on cloud liquid water content (and buoyancy) favouring higher precipitation efficiencies.

  • Deeper warm and cold cloud layers (below and above the freezing level respectively). This may have simultaneously increased both warm-rain and ice-phase precipitation formation allowing for a more efficient collection of cloud droplets.

  • Higher CAPE. This will have enhanced updraught velocities, increasing the vertical flux of water vapour and thus the rate of cloud water production.

Despite high low-level humidity and reduced potential for precipitation evaporation, the intense rain rates in the Boscastle case appear to have resulted in rapid downdraught formation, giving rise to bowing segments in the convective line to the northeast of Bodmin Moor (visible in Figure 4(i, j, k)). The final important difference between the two cases was that the Boscastle storm remained stationary for a longer period of time. As previously noted, in the 2010 case the convective line began to move inland after 1400 UTC. In contrast, the Boscastle storm remained in place until 1630 UTC when it was swept northeast by a separate area of convection (visible southwest of the main line in Figure 4(k)). This difference appears to be related to the persistence of deep southwesterly flow in the Boscastle case, compared to veering flow in the 2010 case.

The result of these differences was a far more extreme rainfall event on 16 August 2004. While radar-derived totals for the Boscastle storm reached just over 110 mm between 1200 and 1600 UTC, corrected data from the Environment Agency's tipping-bucket rain gauge at Lesnewth (Burt 2005) shows an accumulation of 170 mm for this period (Figure 5(b)). This underestimation by the radar does not appear to have occurred on 21 July 2010; the total for the rainfall day (0900 UTC, 21 July to 0900 UTC, 22 July) at the Met Office Lower Moor gauge (indicated by a diamond in Figure 5(a)) agrees well with the nearby radar maximum. The difference in radar–gauge agreement between the two cases may be related to changes in the processing of radar data between 2004 and 2010 (Harrison et al., 2009), although it should be noted that the ZR relationship remained constant during this time. Another possibility is that attenuation effects were more severe in the Boscastle case due to the higher rain rates.

To further illustrate the differences in precipitation between the two cases, we present Figure 6 which compares rain rate and accumulation time series for the radar grid point with highest accumulation on 21 July 2010 with those for the Lesnewth gauge on 16 August 2004. This illustrates the relative impact of higher rain rates and longer rain duration in the Boscastle case. Extrapolation of the 2010 data suggests that, had the storm persisted as long as the Boscastle QSCS, peak accumulations would have reached around 90 mm. This equates to roughly 30% of the difference between the two events. Meanwhile, average rain rates in the Boscastle case were around 30 mm h−1 higher than in the 2010 case (47 mm h−1 compared to 16 mm h−1). Of course, we cannot know how intense the convective system on 21 July 2010 would have become had it remained stationary for longer. In the Boscastle case, the heaviest rain occurred after 1430 UTC, with over half the total (around 100 mm) falling in the 50 min from 1455 to 1545 UTC.

thumbnail image

Figure 6. Time series of rain rate (bars) and rain accumulation (lines) for the point of maximum radar-derived rainfall accumulation on 21 July 2010 (dark grey) and for the Environment Agency's tipping bucket gauge at Lesnewth, Cornwall on 16 August 2004 (light grey). A heuristic correction has been applied to the Lesnewth data to account for under-reading during periods of intense rainfall (Burt, 2005, gives details). Crosses on the y-axis show the average of rain rates ≥0.2 mm (the resolution of the tipping-bucket gauge) for each case.

Download figure to PowerPoint

In summary, we note that, while the 21 July 2010 QSCS showed clear similarities to the Boscastle QSCS of 16 August 2004, differences in the intensity, duration, and distribution of precipitation gave rise to very different impacts, with no recorded flooding in the former case and a devastating flash flood in the latter. Both events were characterised by deep southwesterly flow, which appears to favour the formation of QSCSs over the Southwest Peninsula (Monk, 1987; Hand, 2005). However, it is clear that subtle differences in the properties of this flow, as well as its evolution over time, can dramatically alter the severity of the convective systems which develop.

3. Numerical model and experiment design

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Case-study: 21 July 2010
  5. 3. Numerical model and experiment design
  6. 4. Simulation results
  7. 5. Summary and discussion
  8. Acknowledgements
  9. References

In order to investigate the mechanisms controlling the continued redevelopment of convection along the Southwest Peninsula west coast on 21 July 2010, a series of simulations were carried out using version 7.3 of the Met Office UM. This section describes the model and the design of the simulations.

3.1. The Unified Model

The UM is a suite of numerical modelling software developed by the UK Met Office for simulating the atmosphere and other Earth-system components on a range of space- and time-scales. The UM dynamical core solves the fully compressible, non-hydrostatic, deep-atmosphere equations using a semi-implicit, semi-Lagrangian predictor–corrector scheme (Davies et al., 2005). In the horizontal, the model uses a regular latitude–longitude grid with Arakawa-C staggering. For limited-area configurations, the pole of the grid is rotated such that the domain is approximately centred on the Equator, in order to minimise variations in grid length across the domain. In the vertical, the model uses a terrain-following hybrid-height coordinate with Charney–Philips staggering. The vertical grid spacing is smallest close to the surface, in order to better resolve boundary-layer processes, and increases approximately quadratically with height. Parametrization schemes are used to represent a variety of subgrid-scale processes, including cloud condensation and evaporation (Smith 1990), cloud and precipitation microphysics (Wilson and Ballard 1999), radiation (Edwards and Slingo 1996), surface exchange (Essery et al., 2003), local and non-local boundary-layer mixing (Lock et al., 2000), and convection (Gregory and Rowntree 1990). High-resolution (Δx ∼1 km) configurations also use a horizontal subgrid mixing scheme based on the approach of Smagorinsky (1963), while for some research configurations (typically those with Δx ≤500 m) this scheme is also applied in the vertical, in place of the boundary-layer scheme.

Among its many applications, the UM is the Met Office's operational numerical weather prediction (NWP) model and is used to produce global and regional deterministic and ensemble forecasts up to six days ahead. At version 7.3, the deterministic nested suite consisted of four different configurations—Global, North Atlantic and European (NAE), UK 4 km (UK4), and UK Variable-resolution (UKV)—each producing four forecasts per day. The UKV model was used for the present investigation and is discussed further in the next section.

3.2. The UKV configuration

The UKV model is a limited-area, variable-resolution configuration of the UM. It was developed to improve the resolution of forecasts over the UK without the need for an intermediate-resolution model to properly treat boundary condition data from the 12 km grid length NAE model (Tang et al., 2012). The UKV horizontal domain consists of three sections: a coarse-resolution (Δx =4 km) outer frame, a fine-resolution (Δx =1.5 km) inner domain, and a variable-resolution transition area between (Figure 7). In the vertical, the model has 70 levels with a top at 40 km. At UM version 7.3, the operational UKV was run at 0300, 0900, 1500, and 2100 UTC each day, with initial and boundary conditions provided by an NAE run initialised 3 h earlier. A data assimilation cycle operated from T − 2 to T + 1 (where T is the forecast run time), which included variational assimilation of surface- and satellite-derived 3D cloud fractions (Renshaw and Francis 2011) and latent heat nudging for the assimilation of radar-derived surface rain rates (Lones and Macpherson 1997).

thumbnail image

Figure 7. UKV model domain and orography height (m, shading). The inner and outer solid boxes show the limits of the constant-resolution interior domain and the variable-resolution transition zone respectively. The dashed box shows the domain used for our simulations. The dotted lines show true latitudes and longitudes.

Download figure to PowerPoint

A key feature of the UKV model is that it treats convection explicitly, i.e. without the use of a parametrization scheme. Since numerical models can only accurately represent processes larger than several grid lengths, individual convective cells (and in particular, their updraughts and downdraughts) are still significantly under-resolved with 1.5 km grid spacing. To truly capture the turbulent nature of deep, moist convection one must apply large-eddy simulation (LES) techniques and use grid lengths of 100 m or less (Bryan et al., 2003). Despite this, equation image(1 km) grid length configurations of the UM have been shown to provide substantial benefit in quantitative precipitation forecasting for convective situations in the UK compared to lower-resolution configurations with parametrized convection (e.g. Roberts and Lean, 2008; Lean et al., 2008).

3.3. Simulation strategy

Simulations of the 21 July 2010 QSCS have been carried out using the UKV configuration of the UM at version 7.3. Note that at the time of the event, the Met Office was actually running a later version (7.6); however, this was not available for the present investigation. The 0400 UTC operational UKV analysis (the output of the model's 3 h data assimilation cycle) was used as initial conditions (ICs) while lateral boundary conditions (LBCs) were provided by the 0000 UTC NAE model forecast. In addition to a control simulation, a number of sensitivity tests were carried out in order to isolate the mechanisms responsible for the repeated initiation of convective cells along the peninsula coastline. These are discussed in detail in section 4.2. A run with 500 m grid spacing was also performed and is described in section 4.3. In each case, the model was integrated forward for 15 h (to 1900 UTC).

In order to minimise undesirable feedbacks on the large-scale flow in the sensitivity runs and improve computational efficiency, all simulations were performed on a smaller domain nested within the full UKV model but with the same resolution (Δx =1.5 km). This domain, shown by the dashed box in Figure 7, consists of 240×240 grid points which correspond exactly to points on the full UKV model grid (to eliminate the need for interpolation of the initial analysis). A single run of the full UKV model was used to provide LBCs for the nested domain at half-hour intervals. The control run and all sensitivity tests were then run on the nested domain using the same initial and boundary condition data. Comparison between the output of the full UKV run and the control run (not shown) revealed some slight differences in storm evolution and precipitation accumulations; however, these do not affect the main findings of this investigation.

It should be noted that our 1.5 km configuration had a different treatment of subgrid vertical mixing from the operational UKV model. The latter uses the Lock et al. (2000) boundary-layer scheme which includes both local and non-local mixing, while our configuration used the Smagorinsky-type scheme. This setting may not be optimal at 1.5 km grid spacing; however, it has been used successfully in even coarser-resolution studies of convection by Holloway et al. (2012). For our purposes, it allowed for a cleaner comparison with the 500 m run, which by default uses the Smagorinsky scheme.

4. Simulation results

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Case-study: 21 July 2010
  5. 3. Numerical model and experiment design
  6. 4. Simulation results
  7. 5. Summary and discussion
  8. Acknowledgements
  9. References

The following section presents results from the various simulations of the 21 July 2010 QSCS. Model data are presented on the rotated pole grid used in UK limited-area configurations of the UM. Where a direct comparison between model and radar data is required, the latter are bilinearly interpolated to the model grid.

4.1. Control simulation

Figure 8 shows the evolution of the surface precipitation field in the control simulation. This can be directly compared with the observed evolution shown in Figure 4(a–f). The model appears to have captured the repeated development of convective cells along the west coast of the peninsula during the late morning and early afternoon, and their subsequent inland propagation. However, there are some notable deficiencies in its representation of both the timing and structure of the storm system. These are further illustrated by Figures 9–11. Figure 9 compares the observed and simulated rainfall accumulations for 0900–1800 UTC, Figure 10 compares the observed and simulated rain intensity along the peninsula coastline for the same period using Hovmöller diagrams, and Figure 11 shows histograms of the observed and simulated rain rates for the entire simulation period and domain.

thumbnail image

Figure 8. Evolution of surface rain rates (mm h−1, shading) in the control simulation. Model orography is contoured with an interval of 200 m. The times shown match those for the radar images in Figure 4(a–f).

Download figure to PowerPoint

thumbnail image

Figure 9. Rainfall accumulations (mm) between 0900 and 1800 UTC on 21 July 2010 from (a) the radar and (b) the control simulation. Crosses mark the point of maximum accumulation. Boxes show the area used to produce the Hovmöller diagrams in Figure 10. These boxes both originate at the same point, are 200 km long and 10 km wide, and are orientated such that they pass through the points of maximum accumulation.

Download figure to PowerPoint

thumbnail image

Figure 10. Hovmöller plots of rain rate (mm h−1) at 5 min temporal resolution between 0900 and 1800 UTC on 21 July 2010 from (a) the radar and (b) the control simulation. These were computed along the boxes shown in Figure 9 with values averaged over the short axes (10 km width). Dotted lines show the locations of the maximum rainfall accumulation indicated in Figure 9. No radar data were available for 1615 and 1620 UTC. Note that the contour values in this figure are different from those in all other rain rate and rain accumulation plots.

Download figure to PowerPoint

thumbnail image

Figure 11. Histogram showing percentage contribution of rainfall rates (in 1 mm h−1 bins up to 60 mm h−1) to grid points with rainfall accumulations ≥20 mm, computed over the entire control simulation domain and time period (0400–1900 UTC). Data are shown for the radar (black), control simulation (dark grey), and 500 m grid length simulation (light grey). Before processing, the radar data were interpolated to the model grid and the 500 m data were smoothed to the control simulation resolution using a 3×3 boxcar average at each grid point.

Download figure to PowerPoint

We first note that the model initiates convection late, with the first precipitating cell appearing at 1000 UTC–over an hour later than in the radar data (Figure 10). The length of time (and thus the distance) between successive cells is also greater in the model. Consequently, the storm system fails to achieve the continuous, linear structure seen in the radar images. The cells themselves are smoother than those observed and too large, particularly during the mature stage of their evolution (Figure 8). Furthermore, they evolve too slowly in terms of the intensity of rainfall they produce (Figure 10). As noted in section 2.2, the observed storms developed rapidly and produced heavy precipitation over Bodmin Moor where they joined the main convective line. As they approached the northeast end of the line in Devon, they generally weakened and became less organised (Figure 4(a–f)). By contrast, the modelled storms produce only light rainfall over Bodmin Moor and do not peak in intensity until they reach North Devon. Beyond this, they continue to grow laterally and weaken only slightly as they move across the Bristol Channel into Wales (Figure 8). The net effect of these differences on the accumulated rainfall is shown in Figure 9. For the 9 h period considered, the accumulation pattern associated with the QSCS is fairly well captured; however, the maximum is less by a factor of two (25 mm compared to 50 mm) and shifted around 100 km to the northeast (Figure 10). This shift is primarily due to the slower development of cells in the model, which also results in reduced along-line accumulation gradients on the upstream (southwest) side of the precipitation maximum. We might expect the difference in maximum accumulation to be greater given the wide spacing between successive cells in the model; however, this appears to have been at least partly compensated for by overly intense precipitation in the convective cores (Figure 10). Indeed, Figure 11 reveals that the simulation has a substantial positive bias in rain rates when compared to the radar observations. For those grid points with accumulations ≥20 mm, too much precipitation is associated with rain rates between 10 and 25 mm h−1, and too little is associated with rain rates below 10 mm h−1.

Clearly there are some significant deficiencies in the representation of the 21 July 2010 QSCS in our control simulation. Some of these may be due to inadequate horizontal resolution, a possibility which is explored in section 4.3 using a 500 m grid length simulation. However, a comprehensive investigation of all model biases is outside the scope of this work. Much research is ongoing into the ability of high-resolution configurations of the UM (and other operational models) to accurately forecast convective precipitation. Here, we note that, while the simulation is far from perfect, it successfully captures the key process for QSCS development: the repeated generation of convective cells in roughly the same location. We therefore turn our attention to the initiation mechanism.

Based on the findings of GCM05 and Monk (1987), we would anticipate that lifting along a boundary-layer convergence line was responsible for the repeated initiation of convection in the present case. An examination of the 10 m horizontal divergence field from the control simulation (Figure 12) confirms this to be the case. Over the course of the morning, areas of strong convergence (divergence < −0.001 s−1) develop along portions of the western coastline (Figure 12(a, b)). These gradually expand and join up, forming a quasi-continuous line by the early afternoon (Figure 12(c)) which subsequently moves inland (Figure 12(d)). The inland movement of the line after 1400 UTC appears to be due to a gradual veering of the background flow associated with the eastward progression of the surface low-pressure system. Figure 12 shows that convective cells repeatedly develop and track along the northwest side of the convergence line and remain bound to it as it moves inland. Vertical cross-sections taken across the coastline (not shown) reveal that the low-level convergence is associated with an overturning circulation, approximately 1 km in depth, superimposed on the background wind field, with vertical velocities up to around 1 m s−1. At particular times and points along the line, this lifting was clearly sufficient for parcels to reach their LFC, initiating deep convection. The resulting cells were then advected northeast, parallel to the convergence line, which continued to supply them with moist, potentially buoyant air.

thumbnail image

Figure 12. Wind vectors and divergence (10−4s−1, shading) at 10 m and surface rain rate greater than 1 mm h−1 (black contours, stippling) in the control simulation at (a) 1000, (b) 1200, (c) 1400, and (d) 1600 UTC. This figure is available in colour online at wileyonlinelibrary.com/journal/qj

Download figure to PowerPoint

Figure 12 shows that the convergence line is associated with a change in wind direction from southerly/south-southwesterly on the southeast (land) side to southwesterly/west-southwesterly on the northwest (sea) side. One might speculate that this results from frictional backing of the flow over land; however, in some locations the wind just offshore clearly veers towards the land. This veering is particularly pronounced at 1400 UTC (Figure 12(c)) at the southwest end of the coastline where it creates a stream of divergent flow emanating from the northern tip of the Land's End Peninsula (a feature that was also present in the Boscastle simulations of GCM05; their Figure 12). These observations are consistent with the idea of the convergence line as a sea-breeze front: higher temperatures over land result in a pressure gradient directed from sea to land which in turn produces an onshore flow component (Miller et al., 2003, give a review of the sea-breeze system). To verify this hypothesis and determine the relative importance of the land–sea temperature contrast and frictional effects (as well as other potential influences), a series of sensitivity tests were performed. These are the subject of the next section.

4.2. Sensitivity tests

To investigate the origins of the simulated convergence line, we consider four factors which are known to generate and modulate regions of boundary-layer convergence: differential surface heating, differential surface roughness, orography, and convective outflow. For the Boscastle case, GCM05 found that a positive land–sea temperature difference was critical to the formation of the convergence line, suggesting that it was a sea-breeze front. Orography, meanwhile, was shown to slightly modulate the precise location of the line and the resulting distribution of precipitation. The authors also suggested the importance of frictional backing of the flow over land in creating an offshore flow component which balanced the sea breeze, maintaining the convergence line in place, and storm-generated outflow in distorting the convergence line at its northeast end. However, these factors were never formally addressed through sensitivity tests. Leoncini et al. (2012) also performed simulations of the Boscastle case using a 1 km grid length version of the UM, including a run without the land–sea roughness contrast. This showed that, contrary to the hypothesis of GCM05, differential surface roughness only modulated the convergence line and was not a necessary condition for its formation.

Table 1 details how each of the sensitivity tests in the present investigation was carried out. Note that the methodology employed to remove the land–sea temperature contrast (the WEAKSUN run) is different from that in GCM05. Specifically, GCM05 fixed the land surface temperature and fluxes to values typical of nearby sea points, whereas we have simply reduced the solar constant. Our approach reduces insolation of the land surface, in turn reducing surface fluxes and thus boundary-layer air temperatures. Sea surface temperatures, on the other hand, are fixed to climatological values, so fluxes and temperatures over sea points are not directly affected. As will be shown, the result is that the low-level land–sea air temperature difference remains negative throughout the simulation. While our approach is less direct than that of GCM05, it had the advantage of being very simple to implement in the model.

Table 1. Details of the sensitivity tests performed.
NameFactor under investigationMethodology
WEAKSUNDifferential surface heatingSolar constant reduced to 400 W m−2
SAMEROUGHDifferential surface roughnessRoughness length for momentum over land and sea fixed to 4×10−5m
NOOROGOrographyLand height over Southwest Peninsula set to 1 m
NOOUTFLOWConvective outflowLatent cooling due to rain evaporation and melting of snow switched off

Figure 13 shows the impact of each sensitivity test on the low-level wind and divergence fields at 1400 UTC (cf. Figure 12(c)). Surface precipitation is also shown; however, it is important to note that slight changes in the instantaneous position and size of the convective cells cannot be considered indicative of a systematic response to a particular change in model set-up. The WEAKSUN run immediately stands out in Figure 13 due to the complete disappearance of the convergence line. Consistent with this, the region of divergent flow emanating from the Land's End Peninsula is no longer present, and the winds along much of the coastline have a reduced westerly component. This confirms the hypothesis that the veering flow offshore is a response to differential heating of the land and sea; i.e. it is part of a sea-breeze circulation.

thumbnail image

Figure 13. As Figure 12, but for each of the sensitivity runs at 1400 UTC: (a) WEAKSUN, (b) SAMEROUGH, (c) NOOROG, and (d) NOOUTFLOW. Black boxes show the area for which the time series in Figures 14 and 15 were calculated. This figure is available in colour online at wileyonlinelibrary.com/journal/qj

Download figure to PowerPoint

thumbnail image

Figure 14. Time series of four quantities from the control simulation (black, solid) and each of the sensitivity runs—WEAKSUN (dark grey, solid), SAMEROUGH (light grey, solid), NOOROG (dark grey, dashed), NOOUTFLOW (light grey, dashed)—computed over the box shown in Figure 13: (a) difference between mean 1.5 m temperatures over land and sea points; (b) mean 10 m zonal wind component over sea points; (c) number of points with 10 m wind divergence less than −5 × 10−4 s−1; and (d) mean surface rain rate. Rain rates from the radar (black, dotted) are also shown in (d). Data are plotted with a time resolution of 10 min.

Download figure to PowerPoint

In contrast, the impact of the other sensitivity tests is relatively minor. As one would expect, reducing the land roughness (Figure 13(b)) results in higher wind speeds and veering (or rather, reduced frictional backing) of the flow. These two changes have counteracting effects on the convergence line: faster winds enhance convergence with the onshore flow along the west coast, while clockwise turning of the wind reduces it. The net effect appears to be small. Thus, contrary to the conclusions of GCM05 but in agreement with Leoncini et al. (2012), frictional effects over land are not necessary for the development of this type of quasi-stationary convergence line. Flattening the orography also has a minor influence on the convergence line (in agreement with GCM05 and Leoncini et al., 2012), though it does of course reduce small-scale variations in the divergence field over land (Figure 13(c)).

In the control simulation, convective outflow is apparent as localised areas of strong divergence coinciding with precipitating cells (Figure 12). These are clearly absent in the NOOUTFLOW run (Figure 13(d)), but this again has little overall impact on the convergence line. Animations of the divergence field for the control run reveal that convective outflow may have locally enhanced and distorted the convergence line; however, it was too weak to substantially influence the evolution of the line or the associated convection. As noted in section 2.2, in reality several cells at the northeast end of the line showed a sudden eastward movement, presumably associated with propagating cold pools. The failure of the control simulation to capture this occurrence perhaps relates to the wide spacing between convective cells. This will have allowed outflow to spread out in both the along-line and cross-line directions, whereas in reality, adjacent cold pools may have merged, restricting motion to only the cross-line direction. Note that, in the NOOUTFLOW run, other areas of convection within the simulation domain are quite strongly affected by the absence of latent cooling. In particular, intensification of a system to the northwest of the peninsula alters the low-level flow here giving rise to the convergence lines and precipitation area visible at the northern edge of Figure 13(d). However, these features do not appear to influence the convection over the Peninsula.

Figure 14 summarises the evolution of each simulation in terms of a number of key variables: mean land–sea temperature difference at 1.5 m, mean 10 m zonal wind over the sea, number of grid points with ‘strong’ 10 m wind convergence (divergence ≤−5 × 10−4), and mean surface rain rate. To focus attention on the area of interest, each of these has been computed over the box shown in Figure 13. As we would expect, the land–sea temperature difference (Figure 14(a)) follows the diurnal cycle of surface heating over land, increasing during the morning and early afternoon, peaking around 1400 UTC, and then decreasing again thereafter. However, in the WEAKSUN run, values remain negative throughout the day. Some localised areas of positive land–sea temperature difference do occur (not shown), but on average the low-level air over this part of the Southwest Peninsula remains cooler than that over the sea. Note that the higher temperatures in the NOOROG run are purely a result of the lower land elevation. The low-level zonal wind over the sea (Figure 14(b)) increases throughout the day, partly in response to the evolution of the large-scale flow (section 2.1), but also due to veering associated with the land–sea temperature difference. The absence of the latter effect in the WEAKSUN run is evident, with a reduced westerly component during most of the day.

Regions of strong convergence exist at the start of the simulations (Figure 14(c)) due to land breezes, with cool air descending down the hills of the peninsula and moving out across the sea. The land breezes decay during the subsequent hours as insolation warms the land, reversing the thermal pressure-gradient acceleration; however, this process is retarded in the WEAKSUN run. In the other simulations, regions of strong convergence again start to form after 0830 UTC, associated with the development of the sea-breeze circulation. These are slightly stronger in the SAMEROUGH simulation due to stronger winds over land. Convergence peaks between 1330 and 1500 UTC, coincident with the development of heavy precipitation (Figure 14(d)), and decays thereafter as the line moves inland and out of the box. In contrast, in the WEAKSUN run, convergence remains weak, increasing only slightly between 1400 and 1600 UTC with the passage of a transient feature associated with the base of the surface pressure trough (Figure 14(c)). The resulting lack of forcing, combined with reduced instability, prevents the development of heavy convective precipitation (Figure 14(d)).

4.3. 500 m grid length simulation

In section 4.1, it was noted that the control simulation shows a number of deficiencies in its representation of the 21 July 2010 QSCS. These include late initiation of convection, cells that are too large, intense, and widely spaced, and slow convective evolution. It has also been noted that with 1.5 km grid spacing, convective storms are still significantly under-resolved. One might therefore anticipate that increasing the resolution would improve the model's representation of this event. To test this hypothesis, a simulation with 500 m grid spacing was performed using a standard research configuration of the UM. Compared to our 1.5 km control simulation, the only difference in model physics is higher critical relative humidity, which is used by the cloud scheme to diagnose condensation and evaporation (Smith 1990).

The same domain, nested within the full UKV model, was used in this simulation but with triple the horizontal resolution (718×718 grid points). The vertical grid was not altered. Orography and other ancillary data were initially kept at the same resolution and bilinearly interpolated to the 500 m grid. However, it was found that this simple interpolation method concentrated the curvature of the orography field at the original (UKV model) grid points, creating spurious regions of low-level convergence and divergence. To alleviate this problem, the interpolated orography data was smoothed using a 3×3 boxcar moving average. As in previous simulations, the 0400 UTC operational UKV analysis was used as initial conditions, LBCs were provided by the full UKV run, and the model was integrated forward for 15 h.

Figure 15 compares the evolution of the 500 m run to that of the 1.5 km control run in an almost identical manner to Figure 14, the only difference being that Figure 15(c) shows mean convergence rather than the number of ‘strong’ convergence points. This change is necessary due to the greater number of grid points and the overall increase in convergence discussed below. The impact of the resolution change on the land–sea temperature difference (Figure 15(a)) and zonal wind over sea points (Figure 15(b)) is relatively small. Both are slightly enhanced during the late morning and early afternoon in the 500 m run. This appears to be the result of reduced cloud cover (presumably associated with the higher critical relative humidity), which leads to enhanced short-wave heating of the land surface and a slightly stronger sea-breeze circulation. Far more dramatic changes, however, are seen in the convergence and precipitation fields. Throughout the simulation, but especially from 0900 to 1600 UTC, convergence is enhanced in the 500 m run (Figure 15(c)). The rapid increase in convergence around 0900 UTC is shortly followed by the development of heavy precipitation (Figure 15(d)). When compared to the radar observations, the 500 m run shows vast improvements in the timing and rate of convective development, although it overdoes the area-averaged rainfall intensity.

thumbnail image

Figure 15. Similar to Figure 14, but comparing the control simulation (black, solid) and the 500 m grid length simulation (grey, solid). Here (c) shows the mean convergence, i.e. the mean of all points with divergence < 0.

Download figure to PowerPoint

To better illustrate the changes in precipitation, wind, and divergence, we present Figure 16, which shows a snapshot of these fields at 1400 UTC in the 500 m run. Comparison of Figure 16(a) with Figure 4(c) (observed rainfall) and Figure 8(c) (control simulation rainfall), shows that increasing the horizontal resolution has somewhat improved the model's representation of the structure of the storm system: cells are more numerous, more closely packed, and have enhanced fine-scale structure. Animations reveal that the storm evolution, including the eastward propagation and weakening of cells at the north end of the line, is also more in line with observations. Furthermore, Figure 11 demonstrates that the positive bias in rain intensity, while still present, is significantly reduced in the 500 m run. Despite these improvements, the representation remains far from perfect. This may be attributed at least partly to the convection still being under-resolved (Bryan et al., 2003), but there are of course many other potential sources of model error. Previous experience with high-resolution configurations of the UM has shown that explicit convection can be quite sensitive to parametrization settings, particularly those related to cloud and precipitation microphysics and subgrid mixing (Clark 2006). For the present case, recent tests with a new model version (7.8) suggest some sensitivity to the choice of vertical mixing scheme. Specifically, use of the boundary-layer scheme, rather than Smagorinsky mixing, improves the representation of the convective system, though less dramatically than the 500 m run (not shown).

thumbnail image

Figure 16. Output from the 500 m simulation at 1400 UTC: (a) Surface rain rate (mm h−1), and (b) 10 m wind vectors and divergence (10−4 s−1, shading). This figure is available in colour online at wileyonlinelibrary.com/journal/qj

Download figure to PowerPoint

Turning to the divergence field (Figure 16(b); cf. Figure 12(c)), the most striking change is in the scale and magnitude of the maxima and minima. Most significantly for this case, the convergence line is stronger and narrower with values around −5×10−3s−1 over a width of just 3–5 grid points. The wind field meanwhile is changed very little. This shows that, as we increase the resolution, the horizontal scale over which the wind varies decreases accordingly, allowing for enhanced convergence/divergence and associated vertical motions. Cross-sections (not shown) confirm the presence of stronger low-level ascent and an associated deepening of the boundary layer along the convergence line. For the present case, this change is significant in determining the timing and pattern of convective initiation.

5. Summary and discussion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Case-study: 21 July 2010
  5. 3. Numerical model and experiment design
  6. 4. Simulation results
  7. 5. Summary and discussion
  8. Acknowledgements
  9. References

We have presented an analysis of a quasi-stationary convective system which formed over the UK Southwest Peninsula on 21 July 2010. This system showed remarkable similarity to the flash-flood-producing Boscastle storm of 16 August 2004. In both events, convective cells repeatedly developed and moved along and just inland of the peninsula's west coast, producing intense precipitation over a narrow swath of land. However, maximum rainfall accumulations were approximately four times smaller in the 2010 case and no flooding was recorded. This difference is related to three factors: the intensity of the rainfall, the duration of the convective systems, and the distribution of the rainfall across drainage basins. In the Boscastle case, average rain rates were around three times higher than those in the 2010 case. A comparison of soundings for the two cases suggests a more favourable environment for intense precipitation on 16 August 2004, with greater precipitable water, higher mid-level relative humidity, a deeper layer of instability, and higher CAPE. The Boscastle storm also remained quasi-stationary for around 90 min longer than the 2010 storm due to slower evolution of the wind field: in the latter case, veering low-level flow caused the convective system to move inland several hours before it dissipated. Finally, slight differences in the location of the two storms meant that, in the Boscastle case, the heaviest rainfall was distributed over fewer river catchments, further enhancing the hydrological response. At the synoptic scale, both events were characterised by a slow-moving, weakly baroclinic cyclone; however, in the 2010 case this feature was over the UK while in the Boscastle case it was centred around 1000 km farther west. Despite this difference, both situations gave rise to deep southwesterly flow over the Southwest Peninsula which appears to be crucial to the development of the QSCSs.

Numerical simulations of the 21 July 2010 event were performed using a 1.5 km grid length configuration of the Met Office Unified Model. A control simulation successfully captured the repeated development of convective cells along the coastline, but failed to accurately represent the narrow, linear structure of the storm system. The model also showed a substantial positive bias in instantaneous rain rates and underestimated the storm-total precipitation due to wide spacing between successive convective cells. Despite these biases, the simulation was suitable for investigating the mechanism by which the QSCS formed. As in the Boscastle case, convective initiation was maintained by lifting along a quasi-stationary boundary-layer convergence line. Sensitivity tests were performed to determine the mechanisms controlling this feature. In agreement with the findings of GCM05 for the Boscastle case, the convergence line was shown to be the result of a balance between the background flow over land and the near-surface component of a sea breeze circulation along the west coast. However, in contrast to a hypothesis put forward by GCM05, frictional turning of the wind over land was not found to be necessary for this process to occur. Furthermore, the effects of latent cooling-produced storm outflow and the orography of the Southwest Peninsula were not significant in the 2010 case.

To investigate the impact of enhanced horizontal resolution on the modelled storm system, a simulation with 500 m grid spacing was performed. This showed marked improvements in the timing of convective initiation, the structure of the convective system, and the rainfall intensity. Critical to the improvements in convective initiation was an increase in the strength of the convergence line, which allowed low-level air parcels to more readily reach their LFC. This change can be attributed directly to an improved representation of sharp horizontal wind gradients. Observations of boundary-layer convergence lines (e.g. Wilson and Schreiber, 1986; Wilson et al., 1992) reveal that, in reality, the width of these features ranges from around 0.5 to 5 km. It is therefore not surprising that the 1.5 km model failed to adequately resolve the convergence line, particularly when we consider the additional smoothing generated by numerical diffusion and the model's subgrid mixing parametrization.

This final finding is important as it suggests that in situations where boundary-layer convergence is the dominant mechanism of convective initiation, the highest resolutions currently used operationally may still be insufficient for quantitative precipitation forecasting. Barthlott et al. (2010) reached a similar conclusion based on simulations of a convergence line-forced thunderstorm observed during the Convective and Orographically induced Precipitation Study (COPS; Wulfmeyer et al., 2008). They used the German Weather Service's COSMO–DE model with horizontal grid lengths of 2.8 and 1 km. Both runs failed to predict the storm because simulated updraughts along the convergence line were too weak for parcels to overcome CIN. Significant improvements in forecasting convection and its associated hazards (e.g. flash flooding) are anticipated in the next decade, with the introduction of convective-scale ensemble prediction systems (e.g. Clark et al., 2012) and continuing advances in the assimilation of high-resolution remotely sensed observations (e.g. Renshaw and Francis, 2011). However, in certain meteorological situations, improved prediction might only be achieved with the use of even higher resolutions (Δx <1 km). Of course, the computational requirements for such configurations are vast and, in the near future, resources may be better spent on other modelling developments, such as those mentioned above. Thus, for the current generation of high-resolution NWP models, efforts may be required to refine the treatment of horizontal diffusion so that artificial smoothing of convergence zones (and other sharp gradients) is minimised.

Returning to the issue of QSCSs, this study and GCM05 highlight the potential significance of quasi-stationary sea-breeze fronts as a mechanism by which convection may be repeatedly initiated in one area. The basic ingredients for such a feature—a positive land–sea temperature difference and an offshore-directed wind component—are no doubt quite common. However, the balance between the two is delicate, as evidenced in the present case by the sudden inland movement of the convergence line following a subtle shift in the background flow. Based on a synthesis of many previous numerical investigations of sea breezes, Crosman and Horel (2010) suggested that an offshore geostrophic wind greater than 4–8 m s−1 but less than 6–11 m s−1 could cause a sea-breeze front to stall at the coastline. However, in the present case, the offshore wind component was only around 1–2 m s−1. This discrepancy may be related to the relatively small land–sea temperature difference (1–2 °C), but also to the existence of a strong along-shore wind component. Historically, the along-shore component of the background wind has been considered of secondary importance to the cross-shore component which strongly modulates the ability of the sea breeze to move inland or even form (Crosman and Horel 2010). However, this may in part be because the majority of numerical investigations of these interactions have considered infinite coastlines, either through the use of two-dimensional models or three-dimensional models with periodic boundary conditions in the along-shore direction. We hypothesise that, in the case of a finite-length coastline (e.g. a peninsula) with a strong along-shore background wind, the sea-breeze circulation will be weaker (at least near the upstream end of the coastline) because the offshore air is being continually replenished and therefore cannot fully adjust to the thermally driven pressure gradient. Thus, for a given land–sea temperature contrast, a weaker offshore-directed background wind component would be required to balance the sea breeze and create a quasi-stationary convergence line.

In future work, we hope to further investigate quasi-stationary sea-breeze fronts using high-resolution idealised simulations. Specifically, we would like to determine the region of Vg–ΔT parameter space (where Vg is the low-level geostrophic wind and ΔT is the land–sea temperature difference) for which these features can form. Currently, we are constructing a climatology of QSCSs in the UK which will provide valuable information about the frequency of these storms and their relation to coastlines and other topographic features.

Acknowledgements

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Case-study: 21 July 2010
  5. 3. Numerical model and experiment design
  6. 4. Simulation results
  7. 5. Summary and discussion
  8. Acknowledgements
  9. References

RAW was supported through a National Environment Research Council (NERC) studentship (reference no. NE/I528569/1) with CASE support from the Met Office. DJK's contribution was partially supported by a Canadian National Science and Engineering Council grant NSERC/RGPIN 418372-12. Nimrod radar data and UM analyses were provided by the British Atmospheric Data Centre (BADC), and atmospheric sounding data were provided by the University of Wyoming. We thank Maggie Summerfield, Sarah Alcock and Suzanne Long of the Environment Agency for supplying rain-gauge data and information regarding river levels, and Stephen Burt for providing the corrected Lesnewth rain-gauge record. Thanks also to Willie McGinty and Emilie Carter for respectively providing the UKV and 500 m configurations of the Unified Model. Further invaluable help with the model simulations was provided by Rosalyn Hatcher, Carol Halliwell, Sylvia Bohnenstengel, Adrian Lock and Jonathan Wilkinson. We also thank Giovanni Leoncini, Terry Davies, Simon Vosper, Stuart Webster, Steve Willington, Dan Suri, Nick Grahame, Brian Golding, and Roy Kershaw for helpful discussions and suggestions. Finally, we acknowledge the two anonymous reviewers whose comments helped us to improve upon an earlier draft of the manuscript.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Case-study: 21 July 2010
  5. 3. Numerical model and experiment design
  6. 4. Simulation results
  7. 5. Summary and discussion
  8. Acknowledgements
  9. References
  • Barthlott C, Schipper JW, Kalthoff N, Adler B, Kottmeier C, Blyth AM, Mobbs S. 2010. Model representation of boundary-layer convergence triggering deep convection over complex terrain: A case study from COPS. Atmos. Res. 95: 172185. DOI: 10.1016/j.atmosres.2009.09.010.
  • Bryan GH, Wyngaard JC, Fritsch JM. 2003. Resolution requirements for the simulation of deep moist convection. Mon. Weather Rev. 131: 23942416.
  • Burt S. 2005. Cloudburst upon Hendraburnick Down: The Boscastle storm of 16 August 2004. Weather 60: 219227.
  • Chappell CF. 1986. Quasi-stationary convective events. In Mesoscale Meteorology and Forecasting. Ray PS. (ed.) 289310. Amer. Meteorol. Soc: Boston, MA.
  • Clark AJ, Weiss SJ, Kain JS, Jirak IL, Coniglio M, Melick CJ, Siewert C, Sobash RA, Marsh PT, Dean AR, Xue M, Kong F, Thomas KW, Wang Y, Brewster K, Gao J, Wang X, Du J, Novak DR, Barthold FE, Bodner MJ, Levit JJ, Entwistle CB, Jensen TL, Correia J. 2012. An overview of the 2010 Hazardous Weather Testbed Experimental Forecast Program spring experiment. Bull. Amer. Meteorol. Soc. 93: 5574.
  • Clark PA. 2006. ‘High-resolution model progress and plans’. Science Advisory Committee Paper No. 11.9, Met Office: Exeter, UK. http://www.metoffice.gov.uk/learning/library/publications/science/met-office-scientists/mosac.
  • Crosman ET, Horel JD. 2010. Sea and lake breezes: A review of numerical studies. Boundary-Layer Meteorol. 137: 129. DOI: 10.1007/s10546-010-9517-9.
  • Davies T, Cullen MJP, Malcolm AJ, Mawson MH, Staniforth A, White AA, Wood N. 2005. A new dynamical core for the Met Office's global and regional modelling of the atmosphere. Q. J. R. Meteorol. Soc. 131: 17591782.
  • Davis RS. 2001. Flash flood forecast and detection methods. In Severe Convective Storms. Doswell III CA. (ed.) Meteorol. Monogr. 28: 481525. Amer. Meteorol. Soc: Boston, MA.
  • Doswell III CA. 1987. The distinction between large-scale and mesoscale contribution to severe convection: A case study example. Weather Forecasting 2: 316.
  • Doswell III CA, Brooks HE, Maddox RA. 1996. Flash flood forecasting: An ingredients-based methodology. Weather Forecasting 11: 560581.
  • Ducrocq V, Nuissier O, Ricard D, Lebeaupin C, Thouvenin T. 2008. A numerical study of three catastrophic precipitating events over southern France. II: Mesoscale triggering and stationarity factors. Q. J. R. Meteorol. Soc. 145: 131145.
  • Edwards JM, Slingo A. 1996. Studies with a flexible new radiation code. I: Choosing a configuration for a large-scale model. Q. J. R. Meteorol. Soc. 122: 689719.
  • Essery RLH, Best MJ, Betts RA, Cox PM, Taylor CM. 2003. Explicit representation of subgrid heterogeneity in a GCM land surface scheme. J. Hydrometeorol. 4: 530543.
  • Golding B. 2005. ‘Boscastle and North Cornwall post-flood event study: Meteorological analysis of the conditions leading to flooding on 16 August 2004’. Forecasting Research Technical Report No. 459. Met Office: Exeter, UK. http://www.metoffice.gov.uk/archive/forecasting-research-technical-report-459.
  • Golding B, Clark PA, May B. 2005. The Boscastle flood: Meteorological analysis of the conditions leading to flooding on 16 August 2004. Weather 60: 230235.
  • Gregory D, Rowntree PR. 1990. A mass flux convection scheme with representation of cloud ensemble characteristics and stability-dependent closure. Mon. Weather Rev. 118: 14831506.
  • Hand WH. 2005. Climatology of shower frequency in the British Isles at 5 km resolution. Weather 60: 153158.
  • Harrison DL, Driscoll SJ, Kitchen M. 2000. Improving precipitation estimates from weather radar using quality control and correction techniques. Meteorol. Appl. 7: 135144.
  • Harrison DL, Scovell RW, Kitchen M. 2009. High-resolution precipitation estimates for hydrological uses. Proc. Inst. Civil Eng.–Water Manag. 162: 125135.
  • Harrison DL, Gaussiat N, Pierce C, Norman K. 2012. Radar products for hydrological applications in the UK. Proc. Inst. Civil Eng.–Water Manag. 165: 89103.
  • Holloway CE, Woolnough SJ, Lister GMS. 2012. Precipitation distributions for explicit versus parametrized convection in a large-domain high-resolution tropical case study. Q. J. R. Meteorol. Soc. 138: 16921708.
  • Jones CD, Macpherson B. 1997. A latent heat nudging scheme for the assimilation of precipitation data into an operational mesoscale model. Meteorol. Appl. 4: 269277.
  • Lean HW, Clark PA, Dixon M, Roberts NM, Fitch A, Forbes R, Halliwell C. 2008. Characteristics of high-resolution versions of the Met Office Unified Model for forecasting convection over the United Kingdom. Mon. Weather Rev. 136: 34083424.
  • Leoncini G, Plant RS, Gray SL, Clark PA. 2013. Ensemble forecasts of a flood-producing storm: Comparison of the influence of model-state perturbations and parameter modifications. Q. J. R. Meteorol. Soc. 139: 198211.
  • Lock AP, Brown AR, Bush MR, Martin GM, Smith RNB. 2000. A new boundary-layer mixing scheme. Part I: Scheme description and single-column model tests. Mon. Weather Rev. 128: 31873199.
  • Maddox RA, Hoxit LR, Chappell CF, Caracena F. 1978. Comparison of meteorological aspects of the Big Thompson and Rapid City flash floods. Mon. Weather Rev. 106: 375389.
  • Miglietta MM, Regano A. 2008. An observational and numerical study of a flash-flood event over southeastern Italy. Nat. Hazard Earth Sys. 8: 14171430. DOI: 10.5194/nhess-8-1417-2008.
  • Miller STK, Keim BD, Talbot RW, Mao H. 2003. Sea breeze: Structure, forecasting, and impacts. Rev. Geophys. 41: DOI: 10.1029/2003RG000124.
  • Monk GA. 1987. Topographically related convection over the British Isles. In Satellite and Radar Imagery Interpretation, preprints for a workshop held at Reading, 20–24 July 1987. Bader MJ, Waters AJ. (eds). 305324. EUMETSAT: Darmstadt, Germany.
  • Petersen WA, Carey LD, Rutledge SA, Knievel JC, Johnson RH, Doesken NJ, McKee TB, Vonder Haar T, Weaver JF. 1999. Mesoscale and radar observations of the Fort Collins flash flood of 28 July 1997. Bull. Amer. Meteorol. Soc. 80: 191216.
  • Renshaw R, Francis PN. 2011. Variational assimilation of cloud fraction in the operational Met Office Unified Model. Q. J. R. Meteorol. Soc. 137: 19631974.
  • Roberts NM, Lean HW. 2008. Scale-selective verification of rainfall accumulations from high-resolution forecasts of convective events. Mon. Weather Rev. 136: 7897.
  • Romero R, Doswell III CA, Ramis C. 2000. Mesoscale numerical study of two cases of long-lived quasi-stationary convective systems over eastern Spain. Mon. Weather Rev. 128: 37313751.
  • Schumacher RS, Johnson RH. 2005. Organization and environmental properties of extreme-rain-producing mesoscale convective systems. Mon. Weather Rev. 133: 961976.
  • Schumacher RS, Johnson RH. 2008. Mesoscale processes contributing to extreme rainfall in a midlatitude warm-season flash flood. Mon. Weather Rev. 136: 39643986.
  • Shepherd DJ, Colquhoun JR. 1985. Meteorological aspects of an extraordinary flash flood event near Dapto, New South Wales. Austral. Meteorol. Mag. 33: 87102.
  • Smagorinsky J. 1963. General circulation experiments with the primative equations. I: The basic experiment. Mon. Weather Rev. 91: 99164.
  • Smith RNB. 1990. A scheme for predicting layer clouds and their water content in a general circulation model. Q. J. R. Meteorol. Soc. 116: 435460.
  • Tang Y, Lean HW, Bornemann J. 2012. The benefits of the Met Office variable resolution NWP model for forecasting convection. Meteorol. Appl. in press.
  • Wilson DR, Ballard SP. 1999. A microphysically based precipitation scheme for the UK Meteorological Office Unified Model. Q. J. R. Meteorol. Soc. 125: 16071636.
  • Wilson JW, Schreiber WE. 1986. Initiation of convective storms at radar-observed boundary-layer convergence lines. Mon. Weather Rev. 114: 25162536.
  • Wilson JW, Foote GB, Crook NA, Fankhauser JC, Wade CG, Tuttle JD, Mueller CK, Krueger SK. 1992. The role of boundary-layer convergence zones and horizontal rolls in the initiation of thunderstorms: A case study. Mon. Weather Rev. 120: 17851815.
  • Wulfmeyer V, Behrendt A, Bauer H-S, Kottmeier C, Corsmeier U, Blyth A, Craig G, Schumann U, Hagen M, Crewell S, DiGirolamo P, Flamant C, Miller M, Montani A, Mobbs S, Richard E, Rotach MW, Arpagaus M, Russchenberg H, Schlüssel P, König M, Gärtner V, Steinacker R, Dorninger M, Turner DD, Weckwerth T, Hense A, Simmer C. 2008. The Convective and Orographically induced Precipitation Study: A research and development project of the World Weather Research Program for improving quantitative precipitation forecasting in low-mountain regions. Bull. Amer. Meteorol. Soc. 89: 14771486.
  • Zhang M, Zhang D-L. 2012. Sub-kilometer simulation of a torrential-rain-producing mesoscale convective system in east China. Part I: Model verification and convective organization. Mon. Weather Rev. 140: 184201.
  • 1

    Heavy localised precipitation may also be associated with non-stationary linear MCSs where cell motion is approximately parallel to the convective line (the ‘training line, adjoining stratiform’ classification of Schumacher and Johnson, 2005).