METplotpy

METplotpy is the Python plotting library of the METplus ecosystem. It takes the numbers that verification tools produce — forecast-versus-observation statistics, and some diagnostic fields — and renders them as clean, reusable graphics: line and box plots, the classic verification diagrams, ensemble and probabilistic diagnostics, and a deep catalog of spatial, space-time, and specialized scientific plots. You drive it with a small YAML configuration file plus your data, and it writes a static image you can drop into a report or paper.

What METplotpy is

METplotpy is described by its authors as “a Python-based plotting library designed to generate a wide range of diagnostic and verification visualizations.” Put plainly: the verification tools in this ecosystem are very good at computing statistics, but a column of numbers is hard for a human to judge. METplotpy closes that gap — it exists to “help users interpret verification statistics through meaningful plots, making it easier to evaluate forecast performance and model behavior.”

A little vocabulary, defined on first use:

Verification statistic — A number that summarizes how well a forecast matched what was observed — a bias, a correlation, a hit rate, and so on. These are the primary input METplotpy plots.

Verification diagram — A standardized chart that encodes several statistics at once in a layout meteorologists recognize on sight — for example a performance diagram or a reliability diagram. METplotpy ships the classic set.

YAML configuration — A plain-text settings file (key/value pairs) that tells a plotting script what to read, which series to draw, and how the figure should look. It is how every METplotpy plot is parameterized.

METplotpy is written entirely in Python. Most plots are built with the plotly library; the performance diagram is built with matplotlib. It is developed and maintained by the Developmental Testbed Center (DTC) as one component of the larger METplus system.

Who uses it

Forecast developers and verification scientists who already have statistics in hand and need a figure for a report, a paper, or a dashboard. You can call METplotpy directly from a shell, let METplus orchestrate it as part of a larger run, or let METviewer generate plots through it behind a web interface.

Where it sits in a verification workflow

Think of the ecosystem as an assembly line. Statistics are computed first; METplotpy is the stage that draws them. It does not by itself decide what a good forecast is — it faithfully visualizes whatever statistics it is handed, so the numbers stay the source of truth and the plot is the readable face of them.

• Upstream of it sit the things that produce numbers: METcalcpy (the Python statistics engine METplotpy depends on) and MET itself, whose tools emit columnar ASCII statistics.
• METplotpy reads those statistics together with a YAML config and emits a static image.
• Around it sit the drivers: METplus can run it as a configured step, and METviewer uses it to render plots from a database of statistics. In fact the basic line plot is explicitly “used by METviewer for generating series plots.”

Standalone is first-class

You do not need METplus or METviewer to use METplotpy. Each plot is a Python script you can invoke on its own data and config file — which makes it convenient for one-off figures and reproducible scripting.

How a plot is built

Every METplotpy plot follows the same shape: statistics + configuration go in, a static image comes out. The statistics come from MET / METcalcpy as columnar text; the configuration is one or more YAML files; the result is, by default, a PNG image.

Figure 1. The pipeline: MET / METcalcpy statistics and diagnostic fields, paired with YAML configuration, flow into METplotpy, which renders a static image (PNG by default).

Configuration uses a layered scheme. Each plot ships a defaults file that is loaded automatically — for the line plot, line_defaults.yaml in metplotpy/plots/config — and you supply a small custom YAML file that overrides only the settings you care about. The output image is named by the plot_filename setting, and the default image type is PNG (the line plot’s plot_type defaults to png16m).

A configuration gotcha

METplotpy’s YAML files “do not support expanding environment variables.” Write real paths and values in the config rather than relying on $VARS being substituted inside the YAML.

Running a plot

The standalone path is short. Using the basic line plot as the worked example:

1. Prepare your data. Provide statistics as text output from MET in columnar format. The shipped sample lives at metplotpy/test/line/line.data — note the .data extension and tabular layout.
2. Write a custom YAML config. Start from the plot’s defaults and override what you need in a file such as custom_line.yaml (point it at your data, choose the series, set plot_filename). The defaults file is merged in automatically.
3. Run the plotting script. Each plot type is its own Python module — for the line plot: python metplotpy/plots/line/line.py custom_line.yaml.
4. Collect the image. METplotpy writes the static plot (a PNG by default) to the path you set, ready for a report, paper, or dashboard.

Installation in brief

METplotpy installs into a Python (3.12) conda environment via pip install -e ., and relies on the METcalcpy package — install METcalcpy alongside it. Core dependencies include plotly, matplotlib, cartopy, scipy, xarray, pyyaml, and kaleido (for static image export).

The plot catalog

The library’s reach is its main feature. Its own overview promises plots “including, for example but not limited to, box plots, bar plots, performance diagrams, reliability diagrams, and more.” The full set, grouped here by family for readability, is below — every name is taken faithfully from the User’s Guide table of contents.

Family	Plot / diagram	What it is for
Basic statistic plots	Line Plot, Bar plot, Box plot, Scatter Plot, Histogram Plots, Histogram_2d plot	General-purpose charts of statistics over a series, category, or distribution — the everyday workhorses, including a line plot that METviewer uses for series plots.
Verification diagrams	Performance Diagram, Reliability Diagram, ROC diagram, Taylor Diagram, Economic Cost/Lost Value (ECLV) Plots	The classic, standardized verification charts that meteorologists read at a glance — skill, calibration, discrimination, multi-statistic comparison, and economic value.
Ensemble / probabilistic	Ensemble spread-skill plot	Diagnoses whether an ensemble’s spread (its stated uncertainty) matches its actual error — a core check of ensemble reliability.
Spatial & field plots	Contour Plots, Skew-T Log-P Diagram, Wind Rose Diagram, Polar Ice Plot	Plots of fields and soundings in space — filled/line contours, vertical thermodynamic profiles, wind direction/speed roses, and polar sea-ice maps.
Space-time & tropical	Hovmoeller Plot, Spacetime (Cross-spectra) Plot	Diagnostics that resolve structure jointly in space/longitude and time — Hovmöller diagrams and wavenumber-frequency cross-spectra.
S2S / climate	Weather Regime plot, S2S Blocking plot, MJO Plots, MaKE MaKI Plots, Stratosphere Diagnostics Plots, Stratosphere plots	Subseasonal-to-seasonal and climate diagnostics — recurrent weather regimes, atmospheric blocking, the Madden-Julian Oscillation, MaKE/MaKI indices, and stratosphere diagnostics.
Tropical cyclone	TCMPR Plots, TC-RMW Cross-section Plot	TC-specific output: matched-pair (TCMPR) statistics for track/intensity, and radius-of-maximum-wind cross-sections of the storm structure.
Specialized diagnostics	Difficulty Index Plot, FV3 physics tendencies	Targeted scientific plots — a forecast difficulty index map and FV3 model physics-tendency diagnostics.

On the MJO and a note of caution

The MJO Plots chapter covers Madden-Julian Oscillation diagnostics; the standard MJO indices in this space are RMM (the Real-time Multivariate MJO index) and OMI (the OLR-based MJO index), typically shown as phase diagrams and index time series. The brief also references revision and MPR plots — those names did not appear as standalone chapters in the table of contents consulted here, so they are not asserted in the catalog above. Treat the table as the verified set and the official guide as the authority for anything finer-grained.

Deep dives

Two families are worth a closer look. The verification diagrams are the ones you will reach for most in a forecast-evaluation report; the spatial and space-time plots are where the more specialized science lives.

The 4 classic diagrams → Performance, reliability, ROC, and Taylor — what each one shows, how to read it, and which statistics it encodes.

Spatial & space-time plots → Contour, Hovmöller, cross-spectra, Skew-T, wind rose, polar ice, and the S2S / TC diagnostics — the specialized scientific catalog.

Where it connects

Ecosystem links

METplotpy consumes statistics from METcalcpy and MET, is driven as a step within METplus, and provides the plots rendered inside METviewer. It is the visualization endpoint of the verification pipeline — numbers in, figures out.

---

A derived, human-readable re-presentation — not official documentation. Sources: Overview · Installation · Line Plot