Home
Verif is a command-line tool for assessing the quality of weather forecasts for point locations. It can also compare forecasts from competing weather models or that have different post-processing methods applied to them.
The program reads files with observations and forecasts in a specific format (see the "Arranging my own data" section of this tutorial). The input files contain information about dates, forecast lead times, and locations such that statistics can be aggregated across different dimensions. To ensure a fair comparison among files, Verif will discard data points where one or more forecast systems have missing forecasts. Since Verif is a command-line tool, it can be used in scripts to automatically create verification figures.
Thomas Nipen ([email protected])
This tutorial uses Verif version 1.3.0, which was released March 10, 2024. The tutorial will cover many of Verif's more than 90 metrics/diagrams and 70 command-line options. The tutorial will not focus on the motivation behind each metric. For this, we recommend reading the following excellent resource on forecast verification: WWRP/WGNE Joint Working Group on Forecast Verification Research website. Verif supports many of the metrics found on that webpage.
You can also watch the Verif introduction presentation given at the 2017 American Meteorological Society annual general meeting or read more in our article published in BAMS.
To do the tutorial you need to download the short-range wind speed dataset: MEPS.nc and ECMWF.nc. Alternatively, create your own dataset or have a look at other available datasets.
The tutorial dataset files include wind speed forecasts and observations for 202 stations across Norway. The MEPS.nc file contains forecasts for a regional NWP model used at MET Norway, which has 30 ensemble members and a grid resolution of 2.5 km. ECMWF.nc contains forecasts from ECMWF's global medium-range ensemble prediction system, which has 51 members and a resolution of roughly 18 km. The dataset includes data for the May 2020 - December 2022 time period and for forecast lead-times 6 to 48 hours.
The dataset contains both deterministic and probabilistic forecasts, which will allow you to test many features of Verif.
The map below shows what locations are included in the dataset.
