Clarify probabilistic/deterministic in prediction-mode

[yhe-wmo]

The two allowed values at Level 12 (prediction-mode) are "deterministic" and "probabilistic". see https://github.com/wmo-im/tt-nwpmd/blob/main/weather/prediction/type/prediction-system/prediction-mode/prediction-mode.csv.

The following sentence is extracted from the Manual on the GDPFS (WMO-No. 485). Noted that an ensemble system can produce deterministic products.

The Ensemble Prediction System provides a complete estimation of the forecast probability distribution, including a best-estimate deterministic forecast from the ensemble mean, as well as measures of forecast uncertainty and probabilities.

It will be confusing, as for example, if the EPS outputs end up both in the "deterministic" and "probabilistic" trees.

From an end user perspective, I guess an user would like to navigate through the topics and select outputs generated by either the deterministic NWP system or the ensemble NWP system. In this case, is it better to change "probabilistic" to "ensemble"? And in the description we may need to clearly specify that this Level refers to the model rather than the "mode of product".

[yhonda21]

According to the discussion at the meetig on 17 April, it was proposed to chage the description of those control vocaburaries from 'the prediction system' to 'data set'. Levels 9 and 10 are defined from the 'data set' perspective. Levels below 11 were defined from the 'system' perspective. But by this change, level 12 is now defined from 'data set' perspective while levels 11 and 13 are still defined from 'system' perspective. The perspectives to define levels don't look consistent...

[yhe-wmo]

TT-NWPMD meeting on 17.04.2023:

• Some TT members opined that "probabilistic" would be more appropriate than "ensemble", as
  • some centres prefer "probabilistic" rather than "ensemble" in describing the data, as the latter is not widely understood or adopted by data consumers.
  • some centres might terminate the high-resolution deterministic run soon.
• Some suggested the descriptions of controlled vocabulary items should refer to the datasets rather than the prediction system; as
  • it would be more useful to refer to the datasets, irrespective from which prediction system the datasets are produced.
  • Data coming from machine learning systems can also be accommodated.
• The meeting recommended to keep "probabilistic".

[yhe-wmo]

I am thinking about how the proposed core data will be mapped/categorized to the topic hierarchy. This is one example that we may need to consider. The tropical cyclone tracks from global ensemble NWP will become core data. Although the details of TC tracks are yet to be defined (a task team under ET-OWFS is working on it), it is likely that tracks from each individual ensemble member will be provided as core data (e.g. the ensemble plumes of TC tracks). Do you think “probabilistic” is applicable to this particular dataset? Or should we reconsider the description of “probabilistic”?

[ThomasColemanBoM]

I am curious what other task teams have done in this space. I imagine the ocean area would have a similar topic of discussion?

[sebvi]

I am in favor to keep deterministic and probabilistic as is.

[yhe-wmo]

TT-NWPMD Meeting 11.05.2023 decided to keep "deterministic" and "probabilistic", and asked for better descriptions of the two controlled vocabulary items.

Proposed descriptions for TT-NWPMD's discussion and decision:

• deterministic: outputs from a prediction system which provides one single scenario
• probabilistic: outputs from a prediction system which provides multiple scenarios, including probability distribution and uncertainty information

[yhe-wmo]

TT-NWPMD Meeting 2023.06.13 supported the proposed descriptions for deterministic and probabilistic. These descriptions will be validated during the wis2 pilot phase, and to be revisited if necessary.