Computing difference fields

Met.3D can compute the difference between two data fields on-the-fly via the difference variable, a special actor variable type available in any actor that supports multiple variables (horizontal cross-sections, vertical cross-sections, isosurfaces, volume rendering, etc.).

Typical use cases include:

Forecast error: forecast field minus analysis field.
Tendency fields: variable at time T minus the same variable at T − N hours.
Inter-model comparison: variable from dataset A minus the same variable from dataset B.
Ensemble spread: one ensemble member minus another.
Mean flux over a time interval: de-accumulated from running-mean-since-start fields (e.g. ICON surface flux variables).

Adding a difference variable

In the actor properties panel, locate the Difference computation group under the variable list and click Add. A new difference variable is added to the actor.

Configure it using the two dropdown properties that appear:

First variable: The minuend: select any existing variable already added to the same actor.
Second variable: The subtrahend: select any existing variable already added to the same actor.

The computed value at each grid point is first − second. The output grid takes the shape and level type of the first variable. For 3D fields, the second variable is interpolated onto the grid of the first, so the two inputs do not need to share the same resolution or level type. Be aware of interpolation errors and artifacts, though.

The difference variable does not have its own time or ensemble controls. Its data is purely driven by the available data of the two input variables.

Note

Circular dependencies are detected automatically. If a difference variable references itself (directly or via nested difference variables), Met.3D resets the inputs.

De-accumulation (means since forecast start)

Some NWP models store surface variables as running means since forecast start rather than instantaneous values. ICON radiation and surface flux fields follow this convention: a field at forecast hour 12 stores the mean flux averaged over the entire 0-12 h period, not the instantaneous value at hour 12.

To extract the mean over a specific sub-interval (e.g. forecast hours 6-12), enable Variables are means since forecast start.

With this option active, the difference computation de-accumulates the two inputs:

\[\text{result} = t_1 \cdot V_1 - t_2 \cdot V_2\]

where \(t_1\) and \(t_2\) are the seconds elapsed since forecast start for each input, and \(V\) is the stored mean value. The result is the total accumulated quantity over the interval \([t_2, t_1]\).

Leave this option disabled for instantaneous fields or for fields that already store interval differences.

Time normalization

The Time normalization property divides the difference result by the length of the time interval expressed in a chosen unit. This converts a total accumulated quantity into a rate.

Available options:

Disabled: no normalization; the raw difference (or de-accumulated total) is used.
Per second: divide by the interval length in seconds.
Per minute: divide by the interval length in minutes.
Per hour: divide by the interval length in hours.
Per day: divide by the interval length in days.

The two options combine independently. For example, to obtain mean flux in W m⁻² over a sub-interval from ICON flux data:

Enable Variables are means since forecast start to de-accumulate.
Leave Time normalization disabled: The result is already in W m⁻² · s, which after de-accumulation gives the total energy in J m⁻². To get W m⁻², set time normalization to Per second.

For accumulated precipitation fields (kg m⁻²) converted to a rate (mm h⁻¹):

Leave de-accumulation disabled (precipitation is typically not a running mean).
Set Time normalization to Per hour.