1 / 11

Advances in Rfdbk based Verification at DWD

Advances in Rfdbk based Verification at DWD. Felix Fundel Deutscher Wetterdienst FE 15 – Predictability & Verification Tel.:+49 (69) 8062 2422 Email: Felix.Fundel@dwd.de. Content. Rfdbk package news Verification news Significance Tests Conditional Verification I & II

olivers
Download Presentation

Advances in Rfdbk based Verification at DWD

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Advances in Rfdbk based Verification at DWD Felix Fundel DeutscherWetterdienst FE 15 – Predictability & Verification Tel.:+49 (69) 8062 2422 Email: Felix.Fundel@dwd.de

  2. Content • Rfdbk package news • Verification news • Significance Tests • Conditional Verification I & II • Probabilistic Verification • Smaller Changes III. Plans

  3. Rfdbk Package • Bug occurring when extracting only one observation is fixed • New function “bitcheck” allowing to filter bit encoded integers bitcheck Some observation attributes in feedback files are given as bit string, converted to a integer (e.g. flags) By this any combination of attribute values can be assigned to a observation Bitcheck allow to filter all observations for a set of bit values e.g. bitcheck(flags,c(10,18)) tells if the FG and redundancy flag is set Feedback File Definition 2014

  4. Rfdbk Verification Optional namelist options NAME VALUE DESCRIPTION IdentList ‘/path/to/your/identlist’ # Use only station(s) given in list file (integer) statidList ‘/path/to/your/statidlist’ # Use only station(s) given in list file () dateList ‘/path/to/your/datelist’ # Verify dates in list separately (YYYYMMDD) lonlims/latlims ‘0,30’ # Restrict verification domain (faster) iniTimes ‘0,12’ # Use only runs given in argument inclEnsMean ‘TRUE’ # Include EPS mean in det. verification mimicVersus `TRUE` # Uses VERSUS quality check only sigTest ‘TRUE’ # Perform sign. test on differences in score mean conditionN `R code defining condition` # Perform conditional verification alignObs `TRUE`| `FALSE`|`REDUCED` # full/no/reduced observation alignment insType `1,2,3..` # Select txpe of instrument If not given, DWD standard settings are used! Essentially any observation/forecast characteristic contained in feedback files (~50) can be used to refine the verification via namelist.

  5. Significance Test Included in TEMP, SYNOP, det. and ensemble verification • t-test for significant difference from 0 of difference of scores from 2 experiments • Implemented for area mean scores (not station based) • 24 hours between score validity time needed (e.g. test for each initial time separately) • t-tests requirements (normally dist. measurements, iid) is approximately valid for daily scores • Point colors indicate significance • Only visible if runs are not aggregates (e.g. Ini Time 00 or 12) • Visible also when score difference is plotted • Works also in hindcast mode As for now, not possible with conditional verification

  6. Conditional Verifikation I Implemented for SYNOP deterministic verification • Needs a list (ascii file) of dates (,YYYYMMDD) given in the namelist • Model name is extended by +/- (date in/not in list) • Stratification by dates happend during score aggregation, i.e. already produced score files can be reused • Meant to do a conditional verification for e.g. weather types

  7. Conditional Verifikation II Implemented for SYNOP deterministic verification • Using observation properties to define conditions • Several properties can be combined • Arbitrary number of conditions is possible • Conditions are set in namelist • Model name is extended by number of class • Stations that do not report an observation used in a condition statement are not used Example namelist condition1 "list(N='obs==0',N='abs(veri_data-obs)<1') " condition2 "list(N='obs==0',N='abs(veri_data-obs)>=1') " condition3 "list(N='obs%between%c(1,4)',N='abs(veri_data-obs)<1') " condition4 "list(N='obs%between%c(1,4)',N='abs(veri_data-obs)>=1') " condition5 "list(N='obs%between%c(5,7)',N='abs(veri_data-obs)<1') " condition6 "list(N='obs%between%c(5,7)',N='abs(veri_data-obs)>=1') " condition7 "list(N='obs==8',N='abs(veri_data-obs)<1') " condition8 "list(N='obs==8',N='abs(veri_data-obs)>=1')" With this implementation conditions need to relate to the observation (i.e. not possible is lon%between%c(0,20))

  8. Probabilistic Verification • Based on „probability files“ that need to be produced from feedback files. • „probability files“ hold information on probability of an EPS forecast to exceed a threshold • Arbitrary probability files can be aggregated to calculate e.g. Brier Scores (and decomposition), ROC-Area, ROC curve, reliability diagram • This approach is time consuming and not very flexible • Working on verification script to combine ensemble and probabilistic verification

  9. Smaller Changes • Verification • added scores for dew-point, calculated from RH and T (TEMP) • Use of RH observation >300hPa only from Vaisalasondes (before no observations used) (TEMP) • More concise formulation of score aggregation, also consistent across observation & forecast types (groupingsets function coming with newer data.table version) • Including obs. and forecast mean in scores (SYNOP & TEMP) • Parallelization (partly) of station based verification (less run time with same memory requirements) • Visualization • Show station ID with station based scores • Speed improvements for station based scores app • Summary plots (score cards) allow to modify shown variables and sub-domains • Summary plots allow to adjust score range (to increase visibility of small or very large effects)

  10. Plans • Rfdbk Verification • Conditional verification for TEMP • Unify probabilistic and ensemble verification • Add flexibility concerning bias correction (allow to use/discard correction) • Significance test for categorical scores (not good idea yet) • Allow for single member verification • Maybe separate verification functionality from Rfdbk and create a extra R package for that • Spatial Verification • See other presentation

  11. Thank you!

More Related