verif 0.4.0

Features

• Deterministic metrics such as MAE, bias, correlation, RMSE (e.g. -m mae)

• Threshold-based metrics such as the false alarm rate, ETS, EDI, Yule’s Q (e.g. -m ets)

• Probabilistic metrics such as brier score, PIT-histogram, reliability diagrams (e.g. -m bs)

• Special plots like Taylor diagrams (-m taylor), quantile-quantile plots (-m qq).

• Plot scores as a function of date, lead time, station elevation/lat/longitude (e.g. -x date)

• Show scores on maps (-type map)

• Subset the data by specifying a date range and lat/lon range (-llrange 5,10,58 60)

• Export to text (-type text)

• Options to adjust font sizes, label positions, tick marks, legends, etc (-labfs 14)

• Anomaly statistics relative to a baseline like climatology (-c climfile.txt)

• Output to png, jpeg, eps, etc and specify dimensions and resolution (-f image.png -dpi 300)

For a full list, run verif without arguments.

Installation Instructions

Download the source code of the latest version: https://github.com/WFRT/verif/releases/. Unzip the file and navigate into the extracted folder. verif requires python as well as the python packages numpy, scipy, and matplotlib. The python package basemap is optional, but provide a background map when verification scores are plotted on a map. NetCDF4/HDF5 is not required, but will make reading of NetCDF files faster. If this package is not found, the NetCDF capabilities from scipy will be used instead.

Ubuntu

Install the required pacakges:

sudo apt-get install python-numpy python-scipy python-matplotlib python-setuptools

Then install verif by executing the following inside the extracted folder:

sudo python setup.py install

This will create the executable /usr/local/bin/verif. Add this to your PATH environment variable if necessary (i.e add export PATH=/usr/local/bin/:$PATH to ~/.bashrc). If you do not have sudo privileges do:

sudo python setup.py install --user

This will create the executable ~/.local/bin/verif. Add the folder to your PATH environment variable.

Mac OSX

Install python, numpy, scipy, and matplotlib, then install verif by executing the following inside the extracted folder:

sudo python setup.py install

verif will then be installed /usr/local/share/python/ or where ever your python modules are installed (Look for “Installing verif script to <some directory>” when installing). Add the folder to your PATH environment variable.

Examples

Fake data for testing the program is found in ./examples/. There is one “raw” forecast file and one bias-corrected forecast file (where a Kalman filter has been applied). Here are some example commands to test out:

verif examples/raw.txt examples/kf.txt -m mae
verif examples/raw.txt examples/kf.txt -m ets
verif examples/raw.txt examples/kf.txt -m taylor
verif examples/raw.txt examples/kf.txt -m error
verif examples/raw.txt examples/kf.txt -m reliability -r 0
verif examples/raw.txt examples/kf.txt -m pithist

Text-based input

To verify your own forecasts, the easiest option is to put the data into the following format:

# variable: Temperature
# units: $^oC$
date     offset id      lat     lon      elev     obs      fcst   p10
20150101 0      214     49.2    -122.1   92       3.4      2.1    0.914
20150101 1      214     49.2    -122.1   92       4.7      4.2    0.858
20150101 0      180     50.3    -120.3   150      0.2      -1.2   0.992

Any lines starting with ‘#’ can be metadata (currently variable: and units: are recognized). After that is a header line that must describe the data columns below. The following attributes are recognized:

• date (in YYYYMMDD)

• offset (forecast lead time in hours)

• id (station identifier)

• lat (in degrees)

• lon (in degrees)

• obs (observations)

• fcst (deterministic forecast)

• p<number> (cumulative probability at a threshold of 10)

obs and fcst are the only required columns. Note that the file will likely have many rows with repeated values of offsetid/lat/lon/elev. If station and lead time information is missing, then verif assumes they are all for the same station and lead time. The columns can be in any order.

Deterministic forecasts will only have “obs” and “fcst”, however probabilistic forecasts can provide any number of cumulative probabilities. For probabilistic forecasts, “fcst” could represent the ensemble mean (or any other method to reduce the ensemble to a deterministic forecast).

Available metrics

Here is a list of currently supported metrics. Note that the plots that are possible to make depend on what variables are available in the input files.

Proposed NetCDF input

We are working on defining a NetCDF format that can also be read by verif. Here is our current proposal, based on the NetCDF/CF standard:

netcdf format {
dimensions :
   date    = UNLIMITED;
   offset  = 48;
   station = 10;
   ensemble = 21;
   threshold = 11;
   quantile = 11;
variables:
   int id(station);
   int offset(offset);
   int date(date);
   float threshold(threshold);
   float quantile(quantile);
   float lat(station);
   float lon(station);
   float elev(station);
   float obs(date, offset, station);              // Observations
   float ens(date, offset, ensemble, station);    // Ensemble forecast
   float fcst(date, offset, station);             // Deterministic forecast
   float cdf(date, offset, threshold, station);   // Accumulated prob at threshold
   float pdf(date, offset, threshold, station);   // Pdf at threshold
   float x(date, offset, quantile, station);      // Threshold corresponding to quantile
   float pit(date, offset, station);              // CDF for threshold=observation

global attributes:
   : name = "raw";                                // Used as configuration name
   : long_name = "Temperature";                   // Used to label plots
   : standard_name = "air_temperature_2m";
   : Units = "^oC";                               // Used to label axes
   : Conventions = "verif_1.0.0";
   }

Copyright and license

Copyright © 2015-2016 UBC Weather Forecast Research Team. verif is licensed under the 3-clause BSD license. See LICENSE file.