Presentation on theme: "Bill Lawrence – DOH. “Hydrologic Model Output Statistics” Current short term ensembles have proven unreliable, mainly because no hydrologic uncertainty."— Presentation transcript:
Bill Lawrence – DOH
“Hydrologic Model Output Statistics” Current short term ensembles have proven unreliable, mainly because no hydrologic uncertainty included. Mary Mullusky suggested using previous forecasts at XEFS workshop in 12/06 Includes both meteorological and hydrologic uncertainty. Temporary gap-filler till better hydro uncertainty process developed
Basic premise is to compare historical forecasts with observed, then infer how a operational forecast might vary considering current observation and deterministic forecast. Requires long archive of forecasts and observations. Likely only usable at daily forecast points. ABRFC provided data (forecasts and observed) for 12 points for 10 year period. DJ Seo and Satish Regonda primary players.
HMOS Ensemble Processor Reflect meteorological & hydrological uncertainties Deterministic forecasts of precipitation, temperature Deterministic forecasts of streamflow Ensemble traces of streamflow Weather forecasts Correct bias, account for all uncertainties Deterministic Preprocessor Hydrologic Models HMOS State Updating/ Data Assimilation
Observation Single value ensemble mean forecast Oper. Forecast Ensem. mean Number of ensembles 2000, Regime moves from low to high Line numbers Percentiles 1 – 0 th (minimum), 2 – 1 th, 3 – 5 th, th, 5 – 50 th 6 – 75 th, 7 – 95 th, 8 – 99 th, 9 – 100 th ( maximum)
Fully functional new system in place across Southern Region. Pros: Much faster and capable than previous system Allows for creation and distribution to web page of graphics. Allows for >90% functionality (No FOP, no ESRI, etc)
Cons: Region/NOAA required security patches/updates are killing the effectiveness of the system. Requires tremendous resources. Future of the system is very much up in the air with AWIPS2 Where does SR system fit in with “new” national project???
ABRFC started using P1 in Oct 1996 Previous years/months were created using Stage 3 DMIP1 showed different biases pre/post Oct 1996 ABRFC wants consistent bias in gridded data sets ABRFC also wants to leverage current knowledge of precipitation processing for early years; eliminate obvious bad data Hope is to rerun bigger events (pre Oct 10\1996) using P3
Initial team meetings showed project to be complex and time consuming Needed new software and data restoration In early 2000s, ABRFC deleted all but 6 hourly PC and HG data from older archive db due to space limitations, wrote to tape Need to restore hourly data into new archive db on ax Restore data off old tapes Write shef encoding software Process thru shefdecoder – “clean entry”
James Paul has already written application to access ax and create gage files needed for RetroP3 Next step is to write software to create mosaic of old DPAs Likely pick an event or two to test, using human qc and intelligence If testing is successful, final step is human processing of any/all hours Very labor intensive, but outcome worth it
James Paul has also written a PostP3 application…needs additional testing and coding ABRFC eventually plans to develop a gridded version of local bias for WFOS Latest hour Average of Last 6 hours Average of Last 24 hours
ABRFC has been collaborating with NSSL for several years now regarding Q2 Goal of Q2 is to create an automated QPE for the entire lower 48 every 5 minutes; main goal is to aid in the flash flood arena. NSSL is thinking beyond the box, ie different z-r relationships for different areas depending on soundings, etc. ABRFC is downloading xmrgs of data to compare to our QPEs
ABRFC will start sending “bad gage” list soon for automtic inclusion into analysis. NSSL in no way intends to compete with RFC produced QPE ABRFC’s initial impressions are that the QPE is very good, esp for an automated process; AP removal is excellent. Web site :
Website is useful operationally, as one can “review” echos from “yesterday”. Excellent archive of both qpe and radar echos. Since this method will have real-time gages/biases input, along with varying ZRs, and will be updated every 5 minutes, it shows real promise as a precipitation source for the Flash Flood program.