MET

MET is the compute core of the ecosystem. It is a set of C++ command-line tools, developed by the Developmental Testbed Center (DTC) at NCAR, that match a forecast to what actually happened and turn that comparison into numbers and diagnostics. You feed it gridded forecasts plus observations; it emits tabular ASCII statistics and NetCDF fields. Everything else in the ecosystem either drives MET (METplus) or reads what MET produces (METviewer, METexpress, METplotpy, METcalcpy, METdataio).

What MET is — and what it is not

Forecast verification answers a deceptively simple question: how good was the forecast? To answer it you have to pair each forecast value with the matching observation, then summarize the mismatch with well-defined statistics. MET is the engine that does exactly this, at scale, in a reproducible way.

MET began as the verification package for the Weather Research and Forecasting (WRF) model and has since grown to support a range of modeling systems — including the Unified Forecast System (UFS) and SIMA — as long as inputs follow supported file formats. It is deliberately modular: small single-purpose tools that you chain together, each reading standard formats and writing standard formats.

The one-sentence mental model

MET is a collection of command-line tools. Reformatting tools get your data into a common shape, statistics tools do the forecast-vs-observation math, and analysis tools roll the results up across many cases. MET computes; it does not orchestrate runs or draw final plots — that is what the rest of the ecosystem is for.

Forecast field — A gridded model prediction of some variable (temperature, precipitation, wind, …) valid at a given time.

Observation — Ground truth to compare against — either point measurements (stations, soundings, satellites) or a gridded analysis treated as truth.

Matched pair — A single forecast value paired with the observation at the same place and time. Statistics are computed over collections of these pairs.

Line type — A named row format in MET’s ASCII output. Each line type carries a particular family of statistics; the header columns stay the same while the data columns change per line type. The payoff is that one output file holds several row shapes side by side: an MPR line gives the individual forecast/observation pairs, a CNT line gives continuous error scores, and a CTC line gives yes/no event counts — same file, different row shapes, so you read off whichever you need.

Where MET sits in the workflow

Inputs arrive on the left, MET’s tools do the work in the middle, and standardized output fans out to the right — first to MET’s own aggregation step, then downstream into the database, viewer, and plotting components of the ecosystem.

Figure 1. Inputs are reformatted to a common shape, the statistics tools compute matched-pair statistics, and standardized .stat / NetCDF output feeds the rest of the ecosystem. METplus orchestrates the MET runs.

Ecosystem link

MET is the engine; the other components surround it. METplus wraps and automates MET runs, and MET’s output feeds METcalcpy, METplotpy, METdataio, METviewer, and METexpress.

Verification paradigms — how MET’s tools group up

The MET User’s Guide lists tools alphabetically, but the more useful way to think about them is by the kind of comparison they make. Pick the paradigm that matches your question, and the right tool follows.

Point (grid-to-point)

Compare a gridded forecast against scattered point observations — stations, soundings, buoys. The forecast is interpolated to each observation location, then paired. Point-Stat is the workhorse; Pair-Stat works from pre-paired data.

Point-Stat · Pair-Stat

Grid (grid-to-grid)

Compare a forecast grid against an observation/analysis grid on the same grid. Covers continuous, categorical (yes/no threshold events), and neighborhood scores that relax exact location matching. Grid-Stat leads.

Grid-Stat · Series-Analysis · Grid-Diag

Ensemble

Evaluate a forecast ensemble: how well its spread captures uncertainty and how reliable its probabilities are. Ensemble-Stat computes rank histograms, the CRPS (Continuous Ranked Probability Score — a probabilistic forecast’s error in the variable’s own units), and spread-skill; Gen-Ens-Prod derives the ensemble products first.

Ensemble-Stat · Gen-Ens-Prod

Object-based (MODE)

Instead of grading point-by-point, identify coherent features (e.g. rain areas), match forecast objects to observed objects, and attribute the differences. This sidesteps the double penalty — where a forecast feature in nearly the right place is punished twice, once as a false alarm where it was forecast and again as a miss where it was observed. The MODE family leads.

MODE · MODE-Analysis · MODE-TD

Scale-based

Decompose error by spatial scale using a wavelet (intensity-scale) decomposition, revealing at which scales a forecast is skillful. Wavelet-Stat does this.

Wavelet-Stat

Tropical cyclone (MET-TC)

Track-and-intensity verification for tropical cyclones, working in ATCF deck format. A dedicated family pairs forecast and best tracks and summarizes the errors.

TC-Pairs · TC-Stat · TC-Gen · TC-Diag

Aggregation & analysis

Take the per-run statistics and roll them up: filter by case, aggregate across many runs, compute confidence intervals. Stat-Analysis works the .stat output; Series-Analysis accumulates statistics per grid point.

Stat-Analysis · Series-Analysis

Plotting utilities

Quick visual sanity checks of the data and objects, not publication graphics: plot_point_obs, plot_data_plane, and plot_mode_field.

plot_point_obs · plot_data_plane · plot_mode_field

Inputs and outputs

MET reads standard meteorological formats and writes standard verification formats. Getting your data into a shape MET accepts is half the work — which is why a whole category of tools exists just for reformatting.

What goes in

• Gridded data (forecasts and gridded observations/analyses) in GRIB version 1, GRIB version 2, and NetCDF following the Climate and Forecast (CF) conventions, plus raw WRF output and MET-generated NetCDF. For a comparison, forecast and observation grids must be on the same grid — MET can regrid for you, or you can use external utilities.
• Point observations in PrepBUFR, ASCII, or MADIS format. Point obs must be converted into an intermediate MET NetCDF file before the statistics tools can use them.
• Tropical cyclone data in ATCF ASCII format — A-deck, B-deck, and E-deck files.

Reformat before you verify

The statistics tools do not read PrepBUFR or raw point ASCII directly. Run the appropriate reformatting tool first (for example PB2NC for PrepBUFR) to produce a MET NetCDF file of observations. CF-compliant NetCDF must carry the expected Conventions, standard_name, units, and grid_mapping attributes; MET supports the latitude-longitude, Lambert conformal, polar stereographic, and geostationary projections.

What comes out

MET produces STAT files, ASCII files, NetCDF files, and PostScript / PNG plots.

• STAT (.stat) files are tabular ASCII with one record per line. A fixed set of header columns is shared across rows, while the trailing data columns change depending on the line type. Grid-Stat, Point-Stat, Ensemble-Stat, Wavelet-Stat, and TC-Gen all write STAT output.
• Per-line-type ASCII files contain the same content as the STAT file, but each line type is split into its own file with a column-header row — easier to read by eye or in a spreadsheet.
• NetCDF output holds gridded results — for example regridded fields, accumulated precipitation, or matched-pair grids.
• PostScript / PNG plots come from MODE, Wavelet-Stat, and the plotting utilities.

A few line types to recognize

Line type	What it carries	Produced by
MPR	Matched pair — the individual forecast/observation values that statistics are built from	Point-Stat, Grid-Stat
ORANK	Observation rank — where each observation falls within the ranked ensemble members	Ensemble-Stat
ISC	Intensity-scale — skill broken down by spatial scale	Wavelet-Stat

These three are illustrative, not exhaustive — MET defines many line types spanning continuous, categorical, probabilistic, and ensemble statistics. See the source guide for the full set.

The tool catalog

Every MET command-line tool, grouped by role. Reformatting and regridding tools prepare data; the statistics tools do the verification math; the analysis tools aggregate; the plotting utilities give quick visual checks.

Group	Tool	Purpose
Reformat point obs	PB2NC	Convert PrepBUFR point observations to MET NetCDF.
Reformat point obs	ASCII2NC	Convert ASCII point observations to MET NetCDF.
Reformat point obs	MADIS2NC	Convert MADIS-format observations to MET NetCDF.
Reformat point obs	LIDAR2NC	Convert CALIPSO satellite lidar data to MET NetCDF.
Reformat point obs	IODA2NC	Convert IODA-format observations to MET NetCDF.
Reformat point obs	Point2Grid	Interpolate point observations onto a grid.
Reformat gridded	Pcp-Combine	Add, subtract, or accumulate gridded precipitation across files.
Reformat gridded	Regrid-Data-Plane	Regrid a gridded field onto a target grid.
Reformat gridded	Shift-Data-Plane	Shift a gridded field in space (e.g. correct a track offset).
Reformat gridded	Gen-Vx-Mask	Generate verification masking regions for subsetting domains.
Ensemble prep	Gen-Ens-Prod	Derive ensemble products — mean, spread, and probabilities.
Point verification	Point-Stat (deep dive →)	Verify a gridded forecast against point observations (grid-to-point).
Point verification	Pair-Stat	Compute statistics from pre-paired forecast/observation data.
Grid verification	Grid-Stat (deep dive →)	Grid-to-grid verification: continuous, categorical, and neighborhood scores.
Grid verification	Series-Analysis	Accumulate statistics independently at each grid point over a series.
Grid verification	Grid-Diag	Build joint and marginal distributions (multivariate PDFs) from gridded fields.
Ensemble verification	Ensemble-Stat	Ensemble statistics — rank histograms, CRPS, spread-skill, probabilities.
Scale-based	Wavelet-Stat	Intensity-scale (wavelet) decomposition of forecast error.
Object-based	MODE (deep dive →)	Identify, match, and attribute forecast vs observed objects (features).
Object-based	MODE-Analysis	Summarize MODE object attributes across many cases.
Object-based	MODE-TD	Extend object analysis into the time dimension (space-time objects).
Tropical cyclone	TC-DLand	Compute distance-to-land used by other TC tools.
Tropical cyclone	TC-Pairs	Pair forecast tracks with reference (best) tracks.
Tropical cyclone	TC-Diag	Compute diagnostics along tropical cyclone tracks.
Tropical cyclone	TC-Stat	Filter and summarize paired TC track-and-intensity statistics.
Tropical cyclone	TC-Gen	Verify tropical cyclone genesis forecasts.
Tropical cyclone	TC-RMW	Compute fields relative to the radius of maximum winds.
Tropical cyclone	RMW-Analysis	Aggregate TC-RMW output across cases.
Analysis & aggregation	Stat-Analysis	Filter and aggregate .stat output across cases; compute summaries.
Analysis & aggregation	GSID2MPR	Reformat GSI diagnostic files into MET matched-pair records.
Analysis & aggregation	GSIDENS2ORANK	Reformat GSI ensemble diagnostics into observation-rank records.
Plotting	plot_point_obs	Plot point observation locations and values.
Plotting	plot_data_plane	Plot a single gridded data plane for a quick look.
Plotting	plot_mode_field	Plot MODE object fields.
Plotting	WWMCA tools	World Wide Merged Cloud Analysis utilities (Appendix E).

Tool grouping reorganizes the MET User’s Guide table of contents for readability; purposes are summarized from the overview and data-I/O pages. Some tool names appear with minor spelling variants across the guide (e.g. plot_point_obs vs Plot-Point-Obs).

Deep dives

Three of the statistics tools get their own pages — the ones most scientists meet first.

Point-Stat → Grid-to-point verification: forecasts vs station/point observations.

Grid-Stat → Grid-to-grid verification: continuous, categorical, and neighborhood scores.

MODE → Object-based verification: identify, match, and compare features.

Running a tool — the shape of a MET job

Every statistics tool follows the same rhythm: get inputs into a supported format, point the tool at a forecast and an observation with a config file, then read the standardized output.

1. Reformat the inputs. Convert point observations to MET NetCDF (e.g. PB2NC for PrepBUFR), and regrid or accumulate gridded fields as needed (Regrid-Data-Plane, Pcp-Combine) so forecast and observation share a grid.
2. Choose the tool by paradigm. Point obs → Point-Stat; grid-to-grid → Grid-Stat; ensemble → Ensemble-Stat; features → MODE.
3. Configure and run. A MET config file declares the fields, thresholds, masking regions, interpolation, and which line types to write. Run the tool from the command line (or let METplus generate the config and drive the run).
4. Read the output. Inspect the .stat file (or the per-line-type ASCII files) and any NetCDF grids. Continuous scores like FBAR and ME live in continuous line types; thresholded events produce categorical line types.
5. Aggregate across cases. Feed many .stat files to Stat-Analysis to filter, aggregate, and compute summary statistics — then on to the database and viewers.

Ecosystem link

You rarely run these by hand for production work. METplus wraps every MET tool with reusable configuration, and the resulting .stat output is loaded by METdataio into a database for METviewer and METexpress, or processed by METcalcpy and plotted with METplotpy.

---

A derived, human-readable re-presentation — not official documentation. Sources: MET User’s Guide — Overview · MET Data I/O