Introduction

This page describes two studies of convection exhibiting different characteristics; one over N. America associated with the formation of severe tornadoes, the other over central Africa. The N.America case has strong large scale forcing whereas the central African case is driven by the diurnal cycle. The role of convection in these cases is quite different.

Both cases use a forecast from the same date/time (initial conditions).

In these case studies, you will carry out a control forecast, followed by any number of suggested sensitivity experiments.

These cases were used in the 2014 OpenIFS user workshop at the University of Stockholm.
A Linux 'virtual machine' with a complete copy of the Stockholm workshop exercises is available on request from openifs-support@ecmwf.int.
For more details and copies of handouts, please see the workshop page.

US Tornado convection case (Arkansas)

On the 27 April 7pm local time (00UTC 28 April), tornadoes hit towns north and west of Little Rock, Arkansas killing approx 17 people  (see http://edition.cnn.com/2014/04/28/us/severe-weather/index.html?hpt=hp_c2). On the evening on the 28 April fatal tornadoes occurred over Mississippi ( see: http://www.bbc.co.uk/news/world-us-canada-27199071).

This case study will look at the role of convection and the large scale in these events.

More information can also be found on the ECMWF Severe Event Catalogue 201404 - Convection - Arkansas U.S.

African diurnal deep convection (Central Africa)

Over tropical land masses, incoming radiation strongly heats the surface leading to the development of deep convection and precipitation. Observations show that convective activity and precipitation peak in the late afternoon or early evening. Until very recently, numerical weather prediction models struggled to reproduce this diurnal cycle, often predicting convective activity to peak too early in the day. In this case study, aspects of the convective parameterization scheme can be altered to see how the intensity and the diurnal cycle of convection responds.

On this page...

Note that if using OpenIFS version 38r1 some additional code is needed for the diurnal correction of convection case study.

Please retrieve the modified source code in the file 'src38r1-conv.tgz' from the same directory as the initial conditions as described below.

Initial conditions

Note:  The initial conditions data for the case studies described here are available for OpenIFS 40r1. Please contact OpenIFS Support if you require initial experiment data for more recent model version. 

Case study initial conditions for this case study are provided on the OpenIFS ftp site.

The initial conditions are available at a range of different resolutions. The experiment ids are created at ECMWF and used for identifying the model forecasts on the ECMWF archive system (for those with access).

All of the initial data provided are derived from ECMWF operational analyses (not ERA-Interim). These operational analyses use a horizontal resolution of T1279 and 137 vertical levels. Lower resolutions are spectrally fitted and interpolated to produce balanced initial conditions.

As OpenIFS is a spectral model, the 'T' number refers to the triangular truncation in spectral space. Equivalent grid resolutions are:
T159 ~ 125km resolution, T255 ~ 80km, T511 ~ 40km, T799 ~ 25km, T1279 ~ 16km.

The number of vertical levels is given after the letter 'L' e.g. L62 means 62 vertical levels.

Please note that higher resolutions progressively require more processors and computer memory to run.

Available start dates

We recommend that the initial files from the 27th April are used for a 30hr forecast for these exercises. The initial files from the 22nd are provided for interest in examining a longer forecast lead-time for the N.American case.

ResolutionExpt idStart datesFilenameFile size
T159L62g4a52014/04/22 at 00Zt159l62_g4a5_2014042200.tgz19Mb


2014/04/27 at 00Zt159l62_g4a5_2014042700.tgz19Mb
T255L62g4a42014/04/22 at 00Zt255l62_g4a4_2014042200.tgz51Mb


2014/04/27 at 00Zt255l62_g4a4_2014042700.tgz51Mb
T255L91g4552014/0427 at 00Zt255l91_g455_2014042700.tgz76Mb
T511L62gflf20140422 at 00Zt511l62_gflf_2014042200.tgz190Mb


20140427 at 00Zt511l62_gflf_2014042700.tgz190Mb
T799L62gflg20140422 at 00Zt799l62_gflg_2014042200.tgz441Mb


20140427 at 00Zt799l62_gflg_2014042700.tgz441Mb
T1279L62gflh20140422 at 00Zt1279l62_gflh_2014042200.tgz1.1Gb


20140427 at 00Zt1279l62_gflh_2014042700.tgz1.1Gb

To unpack files with .tgz, either use:

tar zxf T159_1999122412_fqar.tgz

or if your tar command does not support compression:

mv T159_1999122412_fqar.tgz T159_1999122412_fqar.tar.gz
gunzip T159_1999122412_fqar.tar.gz
tar xf T159_1999122412_fqar.tar

Download instructions

Example using T255
% mkdir -p runs/convection/t255
% cd runs
% ftp ftp.ecmwf.int
ftp> cd case_studies/convection_USA_Africa
ftp> binary
ftp> get t255l62_g4a4_2014042700.tgz
ftp> quit
% tar zxf t255l62_g4a4_2014042700.tgz
% cd 2014042700
% ls
  ICMCLg4a4INIT  ICMGGg4a4INIT  ICMGGg4a4INIUA  ICMSHg4a4INIT  ecmwf
% ls ecmwf
NODE.001_01.model.1  ifs.start.model.1  namelistfc

Note that the initial conditions unpack into a directory named by the date/time of the forecast start.

The 'ecmwf' directory contains the files produced at ECMWF when this experiment was run:

  • namelistfc : copy this file to 'fort.4' to run the experiment (modify as required)
  • NODE.001_01.model.1 : this is the model output file as run at ECMWF. If your run fails, it may be useful to compare with this file.

Perform control forecast and analysis

The first step is to run the control forecast. Both cases can be studied with a single 30 hour forecast.

Use the initial files dated 27th April 2014 (filenames include 2014042700) starting at 00Z. Some additional initial files are provided for 22nd April and can be used for the N.America tornado case to investigate the impact of lead time on the forecast.

See below for tasks and key questions to address for the control forecast before moving on to the sensitivity experiments.

We suggest starting with horizontal resolution of T255 for these exercises. Higher resolutions can be used for comparison.


Case study: N.America deep convection

On 27 April 2014 7pm local time (00UTC 28 April), tornadoes hit towns north and west of Little Rock, Arkansas.

Key questions and tasks using the control forecast


  1. Understand the weather situation resulting in tornadoes.
  2. Evaluate the control forecast and compare it to the ECMWF reanalysis and observations (perhaps plot hourly precipitation rates).
  3. What is the area of threat according to the control forecast?
    Area of threat is the area where severe weather can expected. This can be identified by plotting parameters such as CAPE (convective available potential energy), CIN (convective inhibition) and 850-hPa equivalent potential temperature. 
  4. How does the convective adjustment process takes place and and what is the role of large scale forcing (why and where it happens)?
    Perhaps plot soundings (tephigrams) (9 pt area average) in threat area before and after the 'tornado event' to see convective adjustment.

 

Case study: African deep convection

Deep convection develops over central Africa as shown on the water vapour satellite image below, with the accompanying ECMWF forecast shown as a false satellite image. Note that local time for the location in central Africa shown on the map is UTC+2hrs.

Key questions and tasks using the control forecast


  1. Understand the weather situation over Africa.
  2. What is the role of large scale in this case (compare with N.America case).
  3. Look at the diurnal variation of key parameters (2m temperature, surface heat fluxes, precipitation, outgoing-longwave-radiation) for location 0N,25E.
    Perhaps plot hourly area average precipitation [-20S to 10N, 10-40E] (can also do this for the sensitivity experiments).
  4. Compare differences in convection profiles between Central Africa and (i) open ocean, and (ii) Amazonia.

Sensitivity experiments

The IFS is highly tuned to give the best forecast over a range of initial conditions. However, it is instructive to try some sensitivity experiments to understand the role of various physical and dynamical processes.

Not all of the suggested experiments are applicable to both cases, indicated in brackets.

  • What's the impact of the different 'lead times' on the forecast of the convection (i.e. starting from different dates)? (N.America only)
  • What's the impact of resolution on the forecast of the convection? (both)
  • Does reducing the model timestep improve or worsen the forecast? (both)
  • Turn off deep convection (both)

    Do this by editing fort.4, find the namelist block NAMCUMF and add a line:

    LMFPEN=false,            ! disable deep convection

  • Impact of the improved diurnal cycle of convection. (Africa only)
    In this sensitivity experiment, look at the timing of convective and precipitation events by changing how the model parametrizes the diurnal cycle.

    OpenIFS has 3 options for the controlling the diurnal cycle. To change between them:

    - Edit the fort.4 file

    - Find the namelist NAMCUMF and change the parameter RCAPDCYCL accordingly:

    RCAPDCYCL = 2 (default) activates the diurnal cycle using sub-cloud CAPE,

    RCAPDCYCL = 1 diurnal cycle using surface sensible heat flux,

    RCAPDCYCL = 0 reverts the code to a setting before the diurnal cycle for convection was implemented.

  • Increase the precipitation auto conversion rate - what impact does this have? (both)

    Edit the source code to increase the auto conversion rate by 20%

    File: ifs/phys_ec/sucldp.F90, change:

    line 123: RKCONV=1._JPRB/6000._JPRB   ! 1/autoconversion time scale (s)

    to:

    line 123: ! RKCONV=1._JPRB/6000._JPRB   ! 1/autoconversion time scale (s)
    line 124: RKCONV=1.2_JPRB/6000._JPRB    ! default scaled by 20%: 1/autoconversion time scale (s)
  • Impact of the convective time scale adjustment (both)
    An optimization factor in the parametrization is used for tuning the diurnal cycle. This can be altered by changing a value in the model namelist.

    To change the timescale:

    - Edit the fort.4 file

    - Find the namelist  NAMCUMF, parameter RTAUA.

    - The default value is RTAUA=1.

    - Run two sensitivity experiments with values of RTAUA = 0.33 and 3.

    The ratio between the actual cloud base mass flux and the unit (initial) cloud base mass flux:

    \[ \frac{M_{base}}{M^*_{base}} = \frac{PCAPE - PCAPE_{bl}}{\tau} \]

    Look at the amplitude of precipitation.

  • Sensitivity to entrainment rate (both)

    To change the entrainment rate:

    - Edit the fort.4 file

    - Find the namelist block NAMCUMF, parameter ENTRORG

    - The default value is ENTRORG= 1.75E-3

    ENTRORG= 5.8E-4 reduced by factor 3 (mostly shallow convection regime)

    ENTRORG= 5.25E-3 multiplied by factor 3 (mostly deep convection regime)

    Look at the cloud top height, precipitation and eventually changes in temperature and moisture fields with respect to the reference. Note also this is having less impact with the diurnal cycle activated.

Additional questions

  • How important is the correct diurnal cycle of precipitation and radiation for 2m temperature and dewpoint forecast?

Further reading

Journal publications



Comments

The forecasting system at ECMWF makes use of "ensembles" of forecasts to account for errors in the initial state. In reality, the forecast depends on the initial state in a much more complex way than just the model resolution or starting date.  At ECMWF many initial states are created for the same starting time by use of "singular vectors" and "ensemble data assimilation" techniques which change the vertical structure of the initial perturbations.

As further reading and an extension of this case study, research how these methods work.

Acknowledgements

We are especially grateful to Peter Bechtold for suggesting these case studies, assisting in their preparation and for his support during the OpenIFS workshop. We also give a special thanks to: Glenn Carver, Filip Vana and Sandor Kertesz in preparing the material for the OpenIFS user workshop in Stockholm 2014, from which most of the material on this page is derived. We also thank the forecast department for their material on the ECMWF Severe Event Catalogue that was used in preparing these cases. We also thank the participants of the OpenIFS user workshop in Stockholm for their participation and enthusiasm.