Professional Documents
Culture Documents
UM Version : 10.1
Last Updated : 2014-12-05 (for vn10.0)
Owner : Jonathan Wilkinson
Met Office
FitzRoy Road
Exeter
Devon EX1 3PB
United Kingdom
This document has not been published; Permission to quote from it must be obtained from the Unified Model
system manager at the above address
UMDP: 035
The Thunderstorm Electrification Parametrization Scheme
Contents
1 Introduction 2
1.1 Processes represented by the Electric Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Processes not represented by the Electric Scheme . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Model resolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
5 Discharge 6
Acknowledgments 9
References 10
1 Introduction
This document describes the science contained within the ‘thunderstorm electrification parametrization scheme’
in the UM. The scheme is closely related to the microphysics (see UMDP-026 ).
The thunderstorm electrification (or electric) scheme attempts to provide a basic insight into the timing and
location of lightning activity for numerical weather prediction. Previous forecasts of thunderstorm activity were
often based on bulk atmospheric parameters (e.g. CAPE) or on tephigram profile which indicated deep con-
vection. The lightning scheme should attempt to try and parametrize the charging and discharging processes
that actually occur within thunderstorms, rather than link them to bulk atmospheric variables, at least with the
convection-permitting forecast models.
The electric scheme is currently an extension of the microphysics (or large-scale precipitation) scheme, in that
it relies heavily on the microphysical processes to predict where clouds will charge. However, it has been
decided to develop an ’electric scheme’ rather than simply add lightning diagnostics to the microphysics code.
This recognises that processes other than cloud microphysics can lead to lightning (e.g. the electrification of
volcanic ash) and allows for the scheme to be extended in the future to account of other atmospheric electric
processes (e.g. the global electric circuit).
A number of processes are not represented by the scheme. This is because the scheme is initially designed to
run quickly and to provide forecasts of lightning risk for operational numerical weather prediction. It is anticipated
that more complex versions of the thunderstorm scheme will become available in future versions of the UM.
The representation of the processes listed below will be examined for possible inclusion during future scheme
development.
The following processes that occur within thunderstorms are not yet represented by the scheme:
• Cloud-to-cloud and intracloud discharges
• Representation of different charged layers
• Strike polarity
• Transient Luminous Events (e.g. sprites and blue jets)
• Triggered of lightning by bodies not directly related to the thunderstorm (e.g. rockets and helicopters).
In addition, the electric scheme does not yet represent atmospheric electrical phenomena other than those
within thunderstorms (examples of processes not represented include the global atmospheric electric circuit
and the ionosphere).
The current version of the electric scheme has been set up to run with high-resolution convection-permitting
models only, which we define here as any numerical weather prediction model with a horizontal grid resolution
of 5 km or less.
The current electric scheme only works with the convective-permitting models because the flash rate schemes
used require prognostic graupel to be turned on. This is only produced in large quantities in strong updraughts,
which in turn require high-resolution models.
Future releases of the scheme are likely to include lightning prediction forecasts for the lower-resolution models,
but this has not yet been implemented. This will involve linking the lightning prediction scheme to some variables
taken from the convection and cloud schemes
In the high-resolution convective-permitting models, the convection scheme has either a very small impact or is
switched off completely. Therefore, in such cases a definition of whether a model grid point is ‘a storm’ or not
has to be made from other properties. One method of determining whether a convective cell is a storm can be
done by determining whether the cell contains graupel, as this is a prerequisite for lightning development. This
can be done simply using graupel water path (GWP), defined as
Z T OA Z T OA
GW P = GW C dz = ρ qg dz, (1)
0 0
where TOA is the top of the atmosphere (limit of the highest model layer). In reality, most graupel will exist at
altitudes of less than 15 km, so the graupel water path is in fact an integral through the cloudy layer. Examining
data from the UKV model, it was noted that a GWP threshold of 200 g m−2 was sufficient to determine that a
cell was a storm. Below this figure, lightning flash rates are likely to be zero or extremely low.
A ‘flash rate scheme’ is a method of diagnosing thunderstorm flash rate from variables in the model. A selection
of schemes are described in table 1. Some schemes can be more sophisticated than others. The very basic just
make empirical relationships between cloud top height and flash rate. Others are based on a flux hypothesis,
which relate the upward motion of ice or hail/graupel to lightning flash rate. Finally, yet another method makes
a direct comparison between the storm and a capacitor.
It should be noted that flash rate schemes differ from ‘explicit electrical schemes’, such as Barthe et al. [2005],
Barthe and Pinty [2007] and Fierro et al. [2013] where the electric charge is actually carried by the model. At
present these schemes are probably too complex to code within the architecture of the Unified Model, but such
schemes are likely to feature in future.
In convection-permitting models, the size of thunderstorms can often be greater than the size of the model grid
box, especially a very-high resolution. Therefore, if a flash rate scheme requires information about the whole
storm, it is necessary to be able to obtain information from a number of model grid columns.
The UM is mostly run in a parallel environment, where the domain is decomposed into a number of sub-domains.
Each domain runs on its own processor, while information that needs to be passed between processors is done
by a communication call, which slows down the model. To avoid communication calls, most physics schemes
are designed to avoid or minimise the need for communication calls to be made.
For the electric scheme, this means that any diagnosis of flash rate that requires properties from a neighbouring
grid box will be slower and more complex than one which can parametrize lightning within the grid column.
The initial electric scheme therefore only includes two relationships that do not require the retrieval of data from
neighbouring grid columns. These are:
• McCaul et al. [2009], described in section 3.3.
• A relationship linking graupel water path with flash rate derived from Deierling et al. [2008], described in
section 3.4.
Using the other relationships in table 1 assuming each grid column is independent of the next tends to over-
predict lightning fluxes in internal testing.
Most flash rate schemes predict ‘total lightning’. To explain why this is and also to define total lightning, it is
necessary to include a short discussion on forms of lightning. Although visual observations of lightning record
many forms (e.g. Sheet lightning, ribbon lightning, forked lightning, bead lightning and very occasionally ball
lightning), remote sensing methods generally are only able to categorise lightning into one of two types:
• Cloud-to-ground (CG) lightning
• Intra-cloud (IC) lightning.
Technically speaking, there is a third category of cloud-to-cloud (CC) lightning, which occurs when there is a
discharge between neighbouring thunderclouds, while IC lightning is a discharge within two parts of a single
cloud. However in practice, they are very difficult to distinguish between the two. Hereafter, like in much of the
literature, it is assumed that IC lightning includes both CC and ‘true’ IC forms.
The ratio of CG to IC lightning varies with many factors. Boccippio et al. [2001] showed that over the continental
United States, the ratio was 2.64–2.94, but with extremes as low as 1.0 over the Rocky Mountains and as high
as 8 or 9 in the Great Plains. However, for individual storms this ratio has been observed to be much higher.
For example, Goodman et al. [1988] found that out of 116 lightning flashes, only 6 ( 5%) were CG, with the
remaining 110 ( 95%) being IC.
Total lightning is defined as the sum of all activity (CG + IC) and total lightning detectors are available. These
are typically used in regions of high levels of damaging lightning activity (e.g. the USA and Brazil). In the USA,
it has been shown that sudden increases in total lightning activity are able to give tens of minutes of warning
of other severe weather [Schultz et al., 2009]. Studies as early as Workman and Reynolds [1949] noted that
IC activity preceded CG lightning, meaning that the measurements of intra-cloud activity can be used to show
when a cell becomes electrically active and a possible danger to aircraft. Given the importance of total lightning
as opposed to just CG, most parametrizations have been therefore developed to forecast total lightning.
The Met Office’s lightning detection network is called Arrival Time Difference network (ATDnet; Gaffard et al.,
2008). It consists of a network of sensors across Europe (most in the UK), which measure Very Low Frequency
(VLF) electromagnetic radiation emitted by each lightning flash. These are typically around 5-20 KHz, with
ATDnet optimised to detect pulses at 13 KHz. ATDnet can detect lightning activity over most of the Western
Hemisphere (and sometimes from even further away). However, the location accuracy is lower, the further the
strike is from Europe). Like similar networks that operate over a long range (examples include the European
Cooperative for lightning detection (EUCLID), or the World Wide Lightning Location Network (WWLLN)), ATDnet
is capable of measuring CG activity. However, ATDnet can only measure some of the IC activity that is taking
place. For example, many of the helicopter-triggered lightning strikes investigated by Wilkinson et al. [2013]
were not detected by ATDnet. This means that comparing the total lightning flash rate schemes to ATDnet data
is not straightforward. However, it should be possible to make a comparison of the locations where the model
predicts strong lightning activity and ATDnet measurements, using a threshold of lightning activity to ensure that
model is representing CG activity.
McCaul et al. [2009] parametrize the total lightning flash rate to be 95% due to the mixed-phase region of the
cloud (r1 ) and 5% due to a larger, widespread production of lightning (r2 ), which probably includes activity from
the anvils of thunderstorms. Hence, the total flash rate is simply a combination of both factors:
r = 0.95 r1 + 0.05 r2 (2)
Factor r1 is parametrized as a linear function of the upward flux of graupel at the −15◦ C level
r1 = k1 wqg [−15C], (3)
while r2 is a linear function of the sum of the graupel water path and total ice water paths
r2 = k2 ( GW P + T IW P ), (4)
where TIWP includes all forms of ice that are not graupel and is derived as in equation 1. This means that if the
model is run with the second ice prognostic (ice crystals; described in UMDP-026 ), both variables are included.
McCaul et al. [2009] use values of 0.042 for k1 and 0.20 for k2 , however these are available as an input to the
scheme and so may be changed by the user.
Deierling et al. [2008] derive from radar data a relation which links the total lightning flash rate per unit time to
the mass of precipitating graupel at altitudes above the −5◦ C level (pg ). When converted to a flash rate per
second1 this can be expressed as
f = 6.0 × 10−10 pg − 0.27. (5)
However for most situations, a very large mass of graupel ( 108 kg ) is required to produce a sensible lightning
flash rate; this is because Deierling et al. [2008] used the mass of graupel in a radar pulse volume. However,
we can derive a more general linear relationship for this
Once a flash rate is determined, it is important for any unused flash (or flash potential) to be allowed to be
advected within the model. Consider a storm with flash rate r and lifetime l. The total flashes for this storm is
therefore given by integrating the flash rate over the storm lifetime:
Z t=l
F= r dt (7)
t=0
Typical flash rates for different storms are shown in table 2. It can be noted that for many storms, the flash rate
implies that less than one flash per model timestep2 occurs.
It is possible just to output the instantaneous flash rate, but by itself this is not sufficient enough to predict
whether a storm will produce lightning or not. For example, a flash rate of 0.001 s−1 could potentially be one of
several possibilities, such as:
1 The original Deierling et al. [2008] paper expressed a flash rate per minute
2 typically under 100 seconds for high-resolution models
Table 2: Typical flash rates of different types of thunderstorms found across the globe and typical number of
flashes in a storm lifetime.
1. The peak of a non-lightning producing storm, which although has some charging taking place, it is insuffi-
cient to produce lightning before the clouds dissipate
2. A storm, which is increasing in power, with flash rate higher at the next timestep, which produces lightning
later in its lifetime.
3. A dissipating storm, which has produced lightning in the past, and will not produce any further lightning.
4. As item 3, but has charged significantly in the last few timesteps and the flash rate here is sufficient to ‘tip
it over the edge’ and produce lightning this timestep.
It is impossible to know from flash rate alone which (or none) of these categories the storm is in. Simply
intergrating the flash rate over the model timestep and accumulating these (in the same way as a precipitation
accumulation is done) has been suggested. However, this would produce non-integer lightning values (e.g. 0.2
of a flash), which is not very realistic. Also, as a storm could potentially move over several grid boxes whilst
the charge is building up, it would lead to an unphysical ‘trail’ of flash values following a storms track. Another
option would be to use a random number generator to determine whether a storm would produce lightning in
this timestep. Although this should work, it does not allow the storm to gradually build up charge over a time.
So, the option chosen is to advect any unused ‘flash potential’ with the storm using a prognostic variable.
At the end of each timestep, the total flashes is determined from equation 7, which is a real number. The integer
component of this number is discharged (see section 5) and the remainder, fr (by definition, a number not less
than zero and always less than 1.0) is advected. In order to do this, the flash potential is ‘attached’ to the storm.
First of all, a constant value, κ is defined for each model column as
fr
κ= , (8)
qgc
where Z T OA
qgc = qgraup . (9)
0
Next, the flash potential at each grid point in the column is determined as
Pt = κ × qgraup . (10)
This is then advected through the model’s dynamical core to produce the flash potential at the next timestep
(Pt+1 ). Naturally, this process is not important in cases where the flash rate is high as more than 1 flash or more
per model timestep will occur. But, in the majority of UK storms, this is not the case and allowing a gradual build
up to a lightning flash makes sense.
5 Discharge
At each model timestep, the electric scheme integrates the flash potential (Pt−1 ) in each model grid column
which has been passed in from the previous timestep, providing an integrated flash potential (It−1 ). This is
then added to the flash rate (r) determined by the flash rate scheme (section 3) to produce an integrated flash
potential for this timestep:
It = It−1 + r ∆t (11)
The integer portion of It at each grid box is ‘discharged’ as a flash, while the remainder fr is prognosed to the
next time step. The number of lightning flashes is a 2D field and as yet, no distinction is made between CG and
IC forms of lightning.
Appendix I: STASH
The following table contains details of the STASH codes available through Section 21 (the Electric scheme).
Acknowledgments
Thanks to members of the Met Office microphysics working group for their assistance and feedback in prepara-
tion of the scheme. Thanks to Paul Field for initiating the lightning work in the UM. Thanks go to Alec Bennett
(Bristol Industrial and Research Associates Limited) and Christopher Emersic (University of Manchester) for
useful discussions regarding atmospheric electricity and thunderstorm electrification.
References
C. Barthe and J.-P. Pinty. Simulation of a supercellular storm using a three-dimensional mesoscale model with
an explicit lightning flash scheme. J. Geophys. Res., 112(D06210), 2007. doi:10.1029/2006JD007484.
(Referenced on page 3.)
C. Barthe, G. Molinié, and J.-P. Pinty. Description and first results of an explicit electrical scheme in a 3D cloud
resolving model. Atmos. Res., 76:95–113, 2005.
(Referenced on page 3.)
J. M. L. Dahl, H. Höller, and U. Schumann. Modeling the flash rate of thunderstorms. Part I: framework. Mon.
Weather Rev., 139:3093–3111, 2011a.
(Referenced on page 3.)
J. M. L. Dahl, H. Höller, and U. Schumann. Modeling the flash rate of thunderstorms. Part II: implementation.
Mon. Weather Rev., 139:3112–3124, 2011b.
(Referenced on page 3.)
W. Deierling, W. A. Petersen, J. Latham, S. Ellis, and H. J. Christian. The relationship between lightning activity
and ice fluxes in thunderstorms. J. Geophys. Res., 113(D15210), 2008. doi:10.1029/2007JD009700.
(Referenced on pages 3, 4, 5, and 8.)
A. Fierro, E. Mansell, D. MacGorman, and C. Ziegler. The implementation of an explicit charging and discharge
lightning scheme within the WRF-ARW model: Benchmark simulations of a continental squall line, a tropical
cycle and a winter storm. Mon. Weather Rev., 2013. doi:10.1175/MWR-D-12-00278.1, in press.
(Referenced on page 3.)
C. Gaffard, A. J. Bennett, N. C. Atkinson, J. Nash, E. Hibbett, G. Callaghan, P. Taylor, and P. Odams. Observing
lightning from the ground over large areas of the globe. In Proceedings WMO Technical Conference on
Instruments and Observing Methods, WMO: St. Petersburg, 2008.
(Referenced on page 4.)
S. J. Goodman, D. E. Buechler, P. D. Wright, and W. D. Rust. Lightning and precipitation history of a microburst-
producing storm. Geophys. Res. Lett., 15:1185–1188, 1988.
(Referenced on page 4.)
E. W. McCaul, S. J. Goodman, LaCasse. K. M., and D. J. Cecil. Lightning threat using cloud-resolving model
simulations. Weather and Forecasting., 24:709–729, 2009.
(Referenced on pages 3, 4, 5, and 8.)
C. Price and D. Rind. A simple lightning parametrization for calculating global lightning distributions. J. Geophys.
Res., 97:9919–9933, 1992. doi:10.1029/92JD00719.
(Referenced on page 3.)
Christopher J. Schultz, Walter A. Petersen, and Lawrence D. Carey. Preliminary development and evaluation
of lightning jump algorithms for the real-time detection of severe weather. J. Appl. Meteorol. Climatol., 48:
2543–2563, 2009.
(Referenced on page 4.)
J. M. Wilkinson, H. Wells, P. R. Field, and P. Agnew. Investigation and prediction of helicopter-triggered lightning
over the North Sea. Meteorol. Appl., 20:94–106, 2013. doi:10.1002/met.1314.
(Referenced on page 4.)
E. J. Workman and S. E. Reynolds. Electrical activity as related to thunderstorm cell growth. Bull. Amer.
Meteorol. Soc., 30:142–144, 1949.
(Referenced on page 4.)