Beyond the PROFET Juergen Schulze MET Norway
Who's MET Norway?
3 PROFET ?
4 PROFF Field Editor
5
6 Prior to PROFET
7 Text forecast ·50000 Pages of Text forecasts a year ·5-20 people 24/7
8 TV forecast
9 Edited Meteogram ·Offshore forecast ·Subjective Point forecasts
10 Auto Meteogram
11 ICAO Products
12 And then came yr.no
13 yr.no ·Launched in September 2007 ·Free data policy ·5th biggest weather website on the planet ·Interpolation on demand ·9 mio places registered ·Top=6.9 mio unique users/week ·Point forecasts obsolete
14 YR Heatmap
15 PROFET … the intentions Bringing the forecaster into the production chain
16 More intentions Creating products from a single source
17 Even more intentions Forecast consistency by using a single source
18 How PROFET works ·Choosing the “correct” model ·Fixing systematic errors in the model fields
19 Usage – Intended ·Intention to point out dangerous errors
20 Usage – Intended … and real ·Intention to point out dangerous errors ·Used to adapt the model output to fit the forecasters opinion
21 Usage - frequency PARAMETERSUMMERWINTER 1h Precipitation1.9 % 2.8 % Wind0.9 %2.4 %
22 VERIFICATION
23 Equitable Threat Score (ETS) YESNO YESab NOcd ·Threat Score: TS= a / ( a + b + c ) ·Hits expected by chance: E = (a+b) ( a+c ) / n ·ETS = ( a – E ) / ( a + b +c – E )
24 Verification: Precipitation 24h precipitation / summer1h precipitation / summer APPROVED DATASET / DEFAULT MODEL
25 Verification Systematic error Precipitation +30h forecast BIAS [mm / 24h ] SUMMERWINTER ALL CASES DEFAULT MODEL APPROVED DATASET 0.7 EDITED CASES DEFAULT MODEL APPROVED DATASET
26 Verification: comments ·Edited forecasts are more pessimistic than the model ·Forecasters are able to identify situations where the model has considerable errors, but editing the fields is not the optimal way to distribute this knowledge ·Situation has changed since the beginning. The default model has been improved by postprocessing the dataset
27 And the production chain? More and more statistical and parametrisation fixes
28 Temperature? ·High resolution height model (50m horizontal) ·Inhomogeneous ·Deep valleys
29 Temperature? ·High resolution height model (50m horizontal) ·Inhomogeneous ·Deep valleys ·Editing forbidden
30 Precipitation ·Use of prob forecasts by median ·Editing is not WYSIWYG ·Post processing is time cost
31 Wind ·Statistical corrections for mountains and coastlines
32 Chosing the model ? Just two times a day, night shift has not the capacity
33 Single point of production? ·Auto text forecasts – only at high sea. ·Norway is quite Inhomogeneous – auto text did not succeed over land
34 Consistency then? ·Still TV forecast produced by table ·ICAO Products made by hand ·Text forecasts still 50K pages
35 Consistency - yr.no/kart vs yr.no
36 Consistency - yr.no/kart vs yr.no
37 Consistency - yr.no/kart vs yr.no
38 Why stopping right now? ·New Supercomputer – New Model ·AROME 2.5 km High res model over Norway ·Updates 6 or 8 times a day ·High res postprocess chains – time intensive with editing
39 The final reason ·Moving PROFET to AROME would cost a recognizable amount of infrastructural change – due to the implementation of NetCDF as the new grid format
40 What next? The Forecaster is off the production chain again
41 Meteorologen live ·Making direct comments on the situation ·Model/run/grid independent ·Issuing geolocalised warnings, alerts, etc directly into the final product ·Maximum latency of 10 minutes
42 Implementation ·Phase 0: Soon ·Using Common Technology for communication (Twitter/Blog) ·Phase 1: September 2013 ·Adapting existing infrastructure to a set of the new requirements ·Phase 2: Not yet set ·Complete new infrastructure with a postGIS database engine for the geolocalisation