metrics.crpss

This class is for calculating the CRPSS (Continous Ranked Probability Skill Score) following the paper of Goddard et al. (2012)

author:Sebastian Illing contact:sebastian.illing@met.fu-berlin.de

CRPSS Module

class metrics.crpss.Crpss(output='/tmp/crpss/output', output_plots='/tmp/crpss/plots', decadals='1960, 1965, 1970, 1975, 1980, 1985, 1990, 1995, 2000', variable='tas', project='baseline0', product1='output', institute1='mpi-m', model='mpi-esm-lr', experiment='decadal', leadtimes='1, 2-9', observation='HadCrut', observation_ensemble='*', ensemblemembers='*', maskMissingValues=True, result_grid=False, cache='/tmp/crpss/cache', baseDir='./', timeFreq='mon', basic_output=True, bootstrapSwitch=False, bootstrap_number=2, leadtimes_mode='yearly', level=None, lonlatbox='', fieldmean=False, zonalmean=False, input_part='input1', colormap='goddard', observation_type='REANALYSIS', reanpath='', months=None)

Class to calculate the CRPSS. Call method “analyze” to get results

Parameters:

• output – Path where the results are saved
• output_plot – Path where the plots are saved
• decadals – a list with decadal experiments, i.e. [1960,1960,...,1995]
• variable – CMOR-Name of variable like ‘tas’ (Near Surface Temperature)
• project – CMOR-Name of the project like CMIP5 or BASELINE1
• product1 – CMOR-Parameter
• institute1 – CMOR-Name of the institute like MPI-M
• model – CMOR-Name of the used model like MPI-ESM-LR
• experiment – Experiment name like decs4e or historical
• leadtimes – Leadtimes to analyze, ie. 1,2-9
• observation – Observation or Reanalysis Experiment, HadCrut or ERA-Int or path to observation file
• ensemblemembers – Comma separated string of esemble members like ‘r1i1p1,r2i1p1,...’
• maskMissingValues – Boolean value. Whether you want to mask missing values or not
• result_grid – Griddescription of resultgrid like r72x36
• cache – Path for cachedir during analysis
• baseDir – Path of Class to find default files
• timeFreq – Timefrequency of the files, ie ‘mon’ or ‘day’
• basic_output – Boolean - If true only basic out is produced (CRPSS_ES and LESS).
• bootstrapSwitch – If you want calculating Significance using Bootstrap methods
• bootstrap_number – Number of Bootstrap runs
• leatimes_mode – ‘monthly’ or ‘yearly’ calculation
• level – For 3D-Files –> Select a single level
• lonlatbox – If you want to select a specific lonlat-Box
• fieldmean – Boolean - If you want to calculate field means
• zonalmean – Boolean - If you want to calculate zonal means
• input_part – String - Will be added to output path and cache file
• colormap – Matplotlib Colormap used for ploting
• observation_type – NOT USED ANYMORE!!!
• reanpath – NOT USED ANYMORE!!!
• months – Usefull if your analysing monthly and your experiment name has a month included

prepareInput()

This method is to search and prepare the input files. Integrated methods: 1. Searching files 2. Remapping 3. Selecting lonlatbox 4. Calculating fieldmean 5. Calculating zonalmean

The prepared files are stored in “self.inputRemapped” and “self.observationRemapped”

analyze()

Main function to calculate the CRPSS after Goddard et al. (2012) It also calculates the ESS and LESS after Kadow et al. (2015) The following steps are performed:

1. Temporal Smoothing
2. Calculating ensemble means of hindcasts
3. Calculating cross-validated means of ensemblemembers
4. Calculating anomalies of hindcast and observations
5. Removing conditional bias
6. Calculating mean ensemble variance and reference STD
7. Calculating climatological standard deviation
8. Calculating CRPS for mean ensemble variance and for reference STD
9. Calculating CRPSS
10. Calculating Ensemble Spread Score (ESS) and Logarithmic ESS
11. Plotting

removeConditionalBiasNew(hindcast, observations, ensembleMembers)

Remove the conditional bias from the ensemble members. Following Murphy 1988 and Goddard et al. 2012

Parameters:

• hindcast – dict of hindcast ensemble means
• observations – dict of observation anomalies
• ensembleMembers – dict of hindcast ensemble members

Returns:

dict of bias corrected ensemble members

_removeConditionalBias(hindcast, bias, ensembleMembers)

This is the single process version of removeConditionalBias

getCrps(hindcast, observation, variances, tag)

Calculates crpss for dictionarys after Gneiting and Raferty (2007)

Parameters:

• hindcast – dict of anomaly conditional bias corrected hindcast
• observation – dict of anomaly observation
• variances – dict of Variances you want to test
• tag – name tag

Returns:

dict of crps

_calcCrps(hindcast, observation, variance, tag)

Method calculates the CRPS after the formula by Gneiting and Raferty (2007).

Parameters:

• hindcast – anomaly conditional bias corrected hindcast
• observation – anomaly observation
• variance – Variance or STD you want to test
• tag – name tag

Returns:

crps file

getCrpss(crps1, crps2, tag)

Calculates the crpss out of two different crps dicts.

Parameters:

• crps1 – dict
• crps2 – dict
• tag – name tag

Returns:

crpss filepath

getReferenceStd(hindcast, observation)

Calculate Reference STD after Goddard et al. Calculates the Root Mean Squared Error between hindcast and observation and standardizes by n-2

Parameters:

• hindcast – dict of anomalies
• observation – dict of observation anomalies

Returns:

reference STD

bootstrap(crps, crpsRef, tag, crpss, bootstrap_number=500, plot_range=[-1, 1], colorbar='RdBu_r')

Bootstrap function to calculate significance crosses for crpss and other metrics. It also plots the fields with significance crosses

_multiBootstrap(crps, crpsRef, tag, crpss, i)

Single process version used in crpss.bootstrap

_multiEnsspreadBootstrap(ensAnoms, anomalies, tempSmoothedObs, i)

TODO: Not working!!!

bootstrapEnsembleSpreadscore(ensAnoms, anomalies, tempSmoothedObs, spreadscore, bootstrap_number, spreadscore_ln)

Bootstrapping method for ESS and LESS

checkEnsembleError(ensList)

Checks if enough ensemblemembers are selected for analysis Crpss needs at least 2 ensemblemembers. And is only reasonable for more