Heat wave duration index (HWDI) based on daily temperature data#

The World Meteorological Organization (WMO), defines a heat wave as five or more consecutive days of prolonged heat in which the daily maximum temperature is higher than the average maximum temperature by 5°C (9°F) or more.

To measure the heat wave index, one of recommendation of the Intergovernmental Panel on Climate Change (IPCC) was to use the heat wave duration index (HWDI). This index has been defined as as the maximum period in each year of at least 5 consecutive days where the maximum temperature is at least 5°C warmer than the daily climatology.

How-to?#

To calculate the HWDI, you need to have timeseries daily maximum temperature data and tools to calculate the index. Please follow below procedure to get started.

Tools#

  1. Windows Subsystem for Linux (WSL) - https://docs.microsoft.com/en-us/windows/wsl/install. Since CDO is not available for Windows, the easiest way to use it is via the WSL. If you are on Linux/macOS, then you are good to go.

  2. Anaconda or Miniconda

  3. Climate Data Operator (CDO) - https://code.mpimet.mpg.de/projects/cdo, installed using conda install -c conda-forge cdo. Please create a new environment to install CDO, just to make sure the CDO will not breaking your current python environment.

Data#

  1. Global high resolution (30 arc sec ~ 1km) daily maximum temperature data (tasmax), 1979 - 2016, available as 1 month data per 1 file, from CHELSA - https://chelsa-climate.org/, downloaded from https://data.isimip.org/datasets/92b05291-fbe5-4ed2-b3df-29ff0cded9f2/.

Case#

California, US

Step-by-step#

Lets start!

NOTE

  • To follow below step, you need to download CHELSA tasmax data from above link. Total size for tasmax data is 1.2TB, please make sure you have bandwidth and unlimited data package

  • Or you can proceed directly to step 2, using pre-computed result for step 1 which is available in folder 01_tasmax, with size 9.75GB. (upload is in progress)

  • Also make sure you have plenty of storage. Step 2 and 3 will produce intermediate output each around 40GB, step 4 and 5 each around 80GB, step 6 around 2GB and step 7 around 430MB.

  1. Clip global data using bounding box

    for fl in ./00_chelsa_global_tasmax/*.nc; do cdo sellonlatbox,-8.3,-7.6,12.3,12.9 $fl ./01_tasmax_/usa_california_`basename $fl`; done

  2. Merge monthly data into annual

    for year in {1979..2016}; do cdo mergetime ./01_tasmax/usa_california_chelsa-w5e5v1.0_obsclim_tasmax_30arcsec_global_daily_${year}??.nc ./02_tasmax_annual/usa_california_chelsa_daily_tasmax_${year}.nc; done

  3. Delete data who has 29th February.

    According to https://code.mpimet.mpg.de/boards/2/topics/9522, dealing with leap years in climate model output has caused many scientists to stumble, since there are just about 25% February 29ths in the time series so that the sample size for February 29th is just not big enough. Therefore, remove it from the time series before analysing it.

    for year in {1979..2016}; do cdo -delete,month=2,day=29 ./02_tasmax_annual/usa_california_chelsa_daily_tasmax_${year}.nc ./03_tasmax_del29feb/usa_california_chelsa_daily_tasmax_${year}.nc; done

    below script also works, using del29dfeb

    for year in {1979..2016}; do cdo -del29feb ./02_tasmax_annual/usa_california_chelsa_daily_tasmax_${year}.nc ./03_tasmax_del29feb/usa_california_chelsa_daily_tasmax_${year}.nc; done

  4. Convert Kelvin to degree Celsius and don’t forget to change the variable (here tasmax) units, too. Combining operators:

    for year in {1979..2016}; do cdo -b 32 -setattribute,tasmax@units="degC" -addc,-273.15 ./03_tasmax_del29feb/usa_california_chelsa_daily_tasmax_${year}.nc ./04_tasmax_celsius/usa_california_chelsa_daily_tasmax_${year}.nc; done

  5. Merge all nc files result from point 4 into single nc.

    cdo mergetime ./04_tasmax_celsius/usa_*.nc ./05_tasmax_all/usa_california_chelsa_daily_tasmax_1979_2016.nc

  6. Calculate the mean TXnorm of daily maximum temperatures for any period used as reference

    cdo ydrunmean,5,rm=c ./05_tasmax_all/usa_california_chelsa_daily_tasmax_1979_2016.nc ./06_tasmax_meanofreference/usa_california_chelsa_daily_tasmaxnorm_ref_1979_2016.nc

    Other alternative is to use 90th percentile of daily max temperature of a five windows centred. If you are interested to compute the running multi-year daily 90th percentile over all input timesteps for a running window of five days, use:

    cdo ydrunmin,5 infile minfile 
cdo ydrunmax,5 infile maxfile 
cdo ydrunpctl,90,5 infile minfile maxfile outfile

  7. Calculate annual heat wave duration index w.r.t mean of reference period

    for year in {1979..2016}; do cdo eca_hwdi ./04_tasmax_celsius/usa_california_chelsa_daily_tasmax_${year}.nc ./06_tasmax_meanofreference/usa_california_chelsa_daily_tasmaxnorm_ref_1979_2016.nc ./07_hwdi/usa_california_chelsa_daily_hwdi_${year}.nc; done

Result#

  1. 2008

    HWDI_2008

    Reference:

  2. 2013

    HWDI_2008

    Reference:

  3. 2016

    HWDI_2008

    Reference:

References#

  1. https://www.wcrp-climate.org/etccdi

  2. http://etccdi.pacificclimate.org/list_27_indices.shtml

  3. https://code.mpimet.mpg.de/projects/cdo/embedded/cdo.pdf

  4. https://code.mpimet.mpg.de/projects/cdo/embedded/cdo_eca.pdf

  5. https://data-infrastructure-services.gitlab-pages.dkrz.de/tutorials-and-use-cases/use-case_climate-extremes-indices_cdo.html

  6. https://en.wikipedia.org/wiki/List_of_heat_waves

CDO operators