Tools for Trainers Discussion

Hi Pat,

I think this is analogous to Vesa's point about people knowing satellite theory not necessarily being able to apply it to diagnosis and prediction. My view is that knowing the theory of NWP models in no way guarantees that this can be applied to forecasting decisions. I have known many modelers who were unable to apply their knowledge to advise on forecast decisions.

I totally agree that forecasters should know as much about the physics as possible, although not the mathematical details. However, they need to put that knowledge into practice. See below.

How would I write the competencies? Here's a VERY rough draft I wrote in an email to Heleen when she was preparing a course, particularly around ensembles. The original learning outcomes (not written by Heleen) were all in terms "describe", "explain", etc. (I would want to give a lot more consideration to the wording and content so don't hold me to the details. I was trying to give an idea of a possible direction for improvement.)

Competencies
1. Improve weather forecasts by applying and adapting operational NWP guidance, including ensemble predictions, in the forecast process. This is done through identification of model reliability, errors and biases and model characteristics due to:
o    analysis errors and biases due to the data assimilation scheme
o    specific model characteristics (resolution and grid point or spectral scheme, physical parameterisations, hydrostatic or non-hydrostatic)
o     etc.
 
and their response to
o    meteorological situations and systems (eg. How they perform in a heavy rain situation, a frontal situation, cut-off, etc)
o    forecast lead time

2. Identify situations and systems where operational models will be more or less predictable

3. Assess the current forecasts, including ensembles, to determine how well their evolution matches that of the real atmosphere.


4. Use ensemble predictions as guidance for probability weather forecasts and to assess the likelihood of extreme situations

One of the best Advanced Forecasters courses Jeff Wilson and I ran was way back in 1993 or 1994. We were asked to improve the use of nwp by the forecasters.

We had a new computer lab with brand new x486 computers with the screens set to 256 colours so the satelllite images looked OK. Very  exciting!And one of our colleagues in the research centre had a numerical model that could run on a PC. (He had adapted the real-time model for use in briefing sailors in the America's Cup ocean yacht racing competition.) With his assistance we installed the model on our PCs.

We asked the forecasters coming on the course to pair up and select a  situation that occurred in their region and was significant for them, eg a heavy rain situation. It was their choice.

We (our friend) ran the analysis and initialisation schemes for each case before the course and produced boundary conditions from the global model. On the first day we taught people how to run the model on the PCs and they produced a forecast with the standard model configuration. They used McIDAS to investigate the meteorological predictions and compare them with their experience from the situation.

During the course we had a theme each day - moisture and cloud schemes, boundary layer, radiation, topography, resolution, etc, etc. A researcher came and presented an aspect of the theory with an emphasis on its role in the model. The participants then ran the model (an hour and a half or so over lunch time) with changed configurations based on advice from the researcher. For example, add levels in the boundary layer but there is no point if it is in the constant flux layer, remove the mountains, remove the moisture, increase resolution, etc.

The afternoon was spent with the participants conducting a meteorological assessment of the changes brought about by the changed model. The researcher spent the afternoon with them, helping them to interpret the results in terms of model behaviour. For example, in one case the rainfall increased when the convection was turned off. This seemed paradoxical but the researcher was able to relate this to the fact that the moisture was still there and that the layer cloud precipitation processes in this model were more efficient at converting the available misture into precipitation. As there was more moisture available to the large scale processes, due to the lack of convection, the total rainfall increased.

The course was extremely successful. The forecasters loved it and the researchers were very happy, being able to make their passions concrete for forecasters and, in turn, learning a lot from the forecasters. The participants returned to their forecast offices with the model software and some of them installed it and continued to run it locally for several years.

I would love to see a numerical model that the training community could run (maybe online) in such a manner.

Does anyone else have experiences teaching NWP that you could share? Also any NWP competencies?

cheers

Ian

---------------------------------------------------------------------------------------
This is my third try to send this. I had a message saying, "Incorrect sesskey submitted, form not accepted!" Luckily I had copied it to a Wordpad doc.

Hi Pat

Forgive but I answer you without having reviewed Ian latest presentation.  However I can give some insite after my experience in the CMO aeronautical Continuing Professional development course in which one of the topics in Use if Mesoscale NWP products operatioanlly.

The approach we took is first introducing the participants to the model, its make up and dynamics, cautioning then that diffirent models especially on the mesoscale have various parametization factors and that the forecaster much be aware of these in the model.

NEXT remeber MODEL are guidance NOT GOSPLE!

In the use of the model we teach that the forecasters but first develop a 'conceptual model' of the particular weather situation based on the various observations and their own knownledge and theory.  Tehn thsy should compare the NWP initial output to the their Conceptual model, and decide if the model output matches their conceptual. If it does go with it if not use other parameters to make up the forecast.  They should also repeatedly monitor the performance of the model to determine its biases. 

But I agree with you that it is a continuous learning process and the forecasters have to made mindful of that fact, especially since NWPs are continuously being updated.