Applications Development and Diagnostics Team FY10 Progress Report

Transcription

1 Applications Development and Diagnostics Team FY10 Progress Report Mark DeMaria NOAA/NESDIS/STAR Wallace Hogsett NOAA/NCEP/NHC HFIP Annual Review Meeting November 9-10, 2010 Miami, FL 1

2 ADD Team Members Co-Leads Ed Rappaport (NHC) and Mark DeMaria (NESDIS) NCEP/NHC W. Hogsett, R. Pasch, C. Sisko, J. Franklin NCEP/EMC V. Tallapragada, J. O Connor, Bob Tuleya, S. Trahan NWS/OST N. Surgi NESDIS and CIRA J. Knaff, K. Musgrave, B. McNoldy, L. Grasso OAR/HRD S. Gopal, R. Rogers, S. Goldenberg, T. Quirno OAR/ESRL M. Fiorino, P. McCaslin DTC and TCMT L. Nance, B. Brown NRL Y. Jin, B. Sampson, A. Schrader FSU T. Krishnamurti, M. Biswas 2

3 ADD FY10 Milestones Develop Stream 1.5 concept of operations NHC, TCMT, NESDIS Synthetic satellite imagery NESDIS, EMC, HRD and NRL Develop first HFIP model output products for NHC NHC Develop capability to use aircraft data in model diagnostic routines HRD Diagnostic Studies NESDIS, EMC, NHC, HRD, ESRL, NRL, NCAR, SUNYA ATCF Upgrades for HFIP NRL Establish first generation HFIP data service TCMT Run SHIPS/LGEM off other models besides GFS, predictability study NESDIS, ESRL and NRL Develop statistical analysis system for assessing source of model errors FSU, EMC 3

4 6.2.1 Synthetic Satellite Imagery EMC adapted community radiative transfer model (CRTM) to HWRF output for GOES IR channels Images displayed on HWRF web page HRD developing microwave capabilities NRL developing COAMPS-TC imagery NESDIS/CIRA collecting real GOES data for comparison Automated model-goes comparison system under development Input: Model GRIB, McIDAS Area input Output: Statistical analysis 5

5 HWRF Synthetic Real GOES Comparison GOES Channel 3 (Water Vapor) 6 hr Images 0 to 126 hr at 6 hr for Hurricane Earl starting 29 Aug 2010 at 18 UTC 6

6 Brightness Temperature Error (K) MAE - Alex Bias - Alex MAE - Earl Bias - Earl Water Vapor Band (6.7 µm) Errors Forecast Hour 7

7 6.4.1 Aircraft Data for Model Diagnostics Two primary uses for airborne datasets: 1. Quick-look assessments of model forecasts for a specific case 2. Comprehensive analyses in a composite framework for assessment of systematic biases Forward model for HWRF-X Microwave Equivalent (37H, 37V, 85H) derived from 3km resolution real-time HWRFX runs 8

8 Quick-look assessments Evolution of axisymmetric tangential wind (shaded, m/s) during RI 21:44Z Aug 28 21:00Z Aug 29 13:40Z Aug 30 23:03Z Aug 28 22:13Z Aug 29 21:23Z Aug 30 09:39Z Aug 29 00:44Z Aug 30 22:31Z Aug 30 11:58Z Aug 29 11:17Z Aug 30 23:33Z Aug 30 9

9 Dropsonde Composites 10

10 6.4.2 Model Diagnostics NHC Product Tools Input: GRIB, NetCDF Software: HPLOT, ATCF, N-AWIPS, grads, Fortran NESDIS/CIRA Input: GRIB, McIDAS Software: Fortran, IDL HRD Diapost Input: Binary Software: Fortran, grads, java, GWT ESRL Input: GRIB Software: python and grads EMC Input: GRIB Software: HPLOT, Fortran, grads NRL Input: IEEE Binary Software: Fortran NCAR/SUNYA Input: NetCDF Software: Fortran, NCL, Matlab 11

11 Model Large-Scale Diagnostics File * HWRF * * AL11 IGOR * STORM DATA NTIME 022 TIME (HR) LAT (DEG) LON (DEG) MAXWIND (KT) RMW (KM) MIN_SLP (MB) SHR_MAG (KT) SHR_DIR (DEG) STM_SPD (KT) STM_HDG (DEG) SST (10C) OHC (KJ/CM2) TPW (MM) LAND (KM) TANG (10M/S) VORT (/S) DVRG (/S) SOUNDING DATA NLEV 022 SURF TIME (HR) T_SURF (10C) R_SURF (%) P_SURF (MB) U_SURF (10KT) V_SURF (10KT) T_1013 (10C) R_1013 (%) Z_1013 (DM) U_1013 (10KT) V_1013 (10KT) Currently available in real-time for HWRF, GFDL, COAMPS-TC

12 Model Inter-comparison of Large-Scale Parameter Errors HWRF, GFDL, COAMPS-TC Track Max Wind Land/Ocean Success Vertical Shear SST 250 hpa T HWRF GFDL COAMPS-TC Ground truth : Working best track, GFS analyses, Reynolds SST analyses 13

13 Pressure COAMPS-TC 72-h forecast for Celia: DTG = Control microphysics Schmidt microphysics Thompson microphysics Outer grid mean condensate mixing ratio Outer grid mean condensate mixing ratio Outer grid mean condensate mixing ratio

14 16 Back end: diapost system - transformed HWRF/ HWRFX outputs in varios coordinate systems & atcf tracks Front end: web portal 19 main products; e.g., Synthetic microwave imagery SHIPS predictors Shear and Steering Time-height cross-sections Swaths (wind and rainfall) 16

15 Intensity Error (kt) Statistical Intensity Models Run on Multiple Global and Regional Models SHIPS GFDL GFDL-SHIPS 10 HWRF HWRF-SHIPS COAMPS-TC Forecast Time (hr) COAMPS-TC-SHIPS 2010 Atlantic Cases, Bonnie-Tomas as of 4 Nov

16 Percent Improvement Percent Improvement LGEM Predictability Study How much improvement is possible from statistical intensity models? All Forecasts Perfect tracks Perfect tracks and large scale forecasts Over Water Forecasts Perfect tracks Perfect tracks and large scale forecasts Forecast Interval (hr) Forecast Interval (hr) 19

17 Bias (kt) Improvement (%) Deterministic Rapid Intensification Aid Issues with IVCN during Rapid Intensification (RI) Large negative biases Large errors RAPID Aid and Consensus RI probabilities from existing operational text product were used to make deterministic forecasts (RAPID) Consensus formed (IVCN+RAPID= IVCN members + RAPID) and compared with IVCN Results using >40% probabilities on independent 2009 cases shown Significant improvement in mean forecast error and bias reduction Improvement over IVCN IVCN+RAPID OFCL Significant Improvement Forecast Period (h) Forecast Intensity Bias IVCN+RAPID OFCL IVCN Publication Sampson, C. R., J. Kaplan, J. A. Knaff, M. DeMaria and C. A. Sisko: A deterministic rapid intensification aid. Wea. and Forecasting, submitted Forecast Period (h) Bias Reduction n=

18 6.6.1 Establish first generation HFIP Data Service Model Data Sources Real-time runs from annual demo Retrospective runs for Stream 1.5 evaluation Types Tier 1: ATCF files Diagnostic: ASCII files containing basic large-scale diagnostics Input to statistical intensity models Compare basic fields from regional and global model forecasts Tier 2: Gridded data sets containing select sub-set of full 3D model fields Processing Sanity check Files meet format specifications Files contain required fields & fields fall within realistic range Organized by project Individual Tier 1 files merged & sorted by storm Initial access Tier 1 & diagnostic files via TCMT website

19 6.6.1 Establish first generation HFIP Data Service Proto-type for all data types Builds on current processing system to provide access to all types of data collected by TCMT Uses RAMADDA (Unidata content repository) to expose data files to HFIP collaborators Organized by project & file type Browse & download on file-by-file basis Archival of directory tree to download entire set of data Updated weekly Future additions Real-time updates Tagging data with metadata Thumbnails Visualization of file content using IDV username & password: hfipteam

20 Regional Models Establish first generation HFIP Data Service Inventory of files that have passed sanity check for Demo 2010 as of 2 Nov 2010 % expected: BT classification of TD,TS,HU,SD & SS and planned runs per day Global Models

21 6.8.1 FSU Advanced System for Model Error Source Assessment Start with closed system of full model equations Regress each term to tendency errors Observed tendency from sequence of verifying analyses Can identify systematic errors in dynamics and physics Fig 1: Deep convective heating (10-4 K/sec) for Hurricane Celia, 2010 (9mn forecast 18z 22 June) from the HWRF model output (upper panel) and the corresponding error term (10-5 K/sec). 24

22 Bonus Material: Business Case (Norfolk, VA KNGU) Greatest chance for 34 kt wind: (72hr Incremental period) 12z Thu to 12z Fri (Norfolk, VA KNGU) Greatest chance for 50 kt wind: (72 hr Incremental period) 12z Thu to 12z Fri 72 hr Cumulative Probability 50 kt winds 24 hrs ago Norfolk, VA (KNGU) 7% 6% This depiction is for OFFICIAL USE ONLY! Not for public release 72 hr Cumulative Probability for 50 kt winds 34 kt winds 24 hrs ago Norfolk, VA (KNGU) 28% 19% 5-9% 10-19% 20-29% *Experimental Threshold Probability: Norfolk, VA (KNGU) 50 kt Forecast = NO 34 kt Forecast = YES Based on 15Z 31 Aug statistical and dynamical model data and Monte Carlo Wind Speed Probabilities

23 Bonus Material: Business Case COMSECONDFLT SORTIE RECOMMENDATION/DECISION TIMELINE 1500 UTC 30 Aug 10 - Set Sortie "Charlie" SOPA (Senior Officer Present Afloat) Hampton Roads 1229 UTC 31 Aug 10 - Set Sortie "Bravo" SOPA Hampton Roads 2030 UTC 31 Aug 10 - SOPA/COMSECONDFLT recommendation to Commander, US Fleet Forces Command to remain in-port 2204 UTC 01 Sep 10 - Set Sortie "Charlie" and COR (shore installation Condition of Readiness) III for Hampton Roads 1430 UTC 03 Sep 10 - Secure from Sortie "Charlie" and shore COR for Hampton Roads Notes: Sortie "Charlie" - All units make preparations to be able to get underway within 48 hours. Sortie "Bravo" - All units make preparations to be able to get underway within 24 hours. Sortie "Alpha - All units get underway as specified in Sortie plan. COR III - Condition of Readiness; Destructive winds (50 kt or greater) are possible within 48 hours. *In addition to the WSP Product, Sortie time-lines are based on numerous factors that may impact the Sortie Evasion Track, (35 kt winds, 12 ft seas, etc...).

24 Bonus Material: Business Case Conclusions and Probability Validation 36 Ships were expected to get underway had SORTIE "Alpha" been attained It would have cost the Navy ~ 6 million dollars in fuel alone to successfully proceed with the complex and expensive evolution Highest winds recorded at KNGU: 340/18G28 kt at 1059 UTC 03 Sep 10 Highest wind recorded in the Hampton Roads area: - CBBV2 (Chesapeake Bay Bridge) 34G40 kt at 1100 UTC 03 Sep 10 - Weatherflow sensor X316 (Little Creek) 34.8G40 kt at 1108 UTC 03 Sep 10 Yes/No 34 and 50 kt Wind Threshold prediction: Verified successfully* * No 34 or 50 kt winds observed at KNGU, however, 34 kt winds were recorded in the vicinity as given in the interpretation of the 72 hr cumulative value documented just above (YES) the Threshold Probability. Predicted winds at Norfolk based on the WSP interpretation were sustained minimal TS winds with TS gusts.

25 Future Directions for HFIP Diagnostics Organize effort into Stream 1 and 2 Provide diagnostic system for HWRF pre-season testing Includes comparison with in situ, satellite data Add microwave forward models to HWRF post-processing Continue advanced diagnostics Continued diapost development Storm and cloud scale diagnostics Balanced model analysis Statistical error source analysis Adjoint based diagnostics Community diagnostic effort Stream 1.5 test of SHIPS/LGEM run off other models 28

26 Overview of Monday Diagnostics Meeting at HRD Purpose: Identify ways that HFIP can help with stream 1 diagnostics Organizers: V. Tallapragada (EMC), M. DeMaria (NESDIS), N. Surgi (NWS), E. Rappaport (NHC) Summary of application development and diagnostics (ADD) team stream 2 activities Three discussion topics Rapid Response Team for real time support Assistance with 2011 HWRF implementation Longer plans for stream 1 and 2

27 Meeting Action Items Establish Rapid Response Team Provide support to EMC for 2011 HWRF pre-implementation testing Monitor models is real time and document problems for later diagnosis Provide transition of stream 2 research into stream 1 operational support Reps from HRD, GFDL, NHC, NESDIS, NRL, DTC, NRL, SUNYA Guidance from ADD team Longer stream 2 effort Possible gaps include housing and maintaining common software, data bases Interact with DTC to provide additional support Meeting web page will be set up

28 HFIP FY2010: NHC Progress Presented by Wallace Hogsett In collaboration with: Ed Rappaport, Chris Sisko, Ann Schrader, Buck Sampson, and Mike Brennan 31

29 FY2010 NHC ADD Deliverables Prototype shear analysis for user-specified layers Ability to record model and forecast probabilities in ATCF Prototype point-and-click vertical cross sections Model intercomparisons Model diagnostic studies 32

30 HFIP FY10 Deliverable User-Selectable Shear for Global, Hurricane and Regional Models Example of new user-selectable shear options in NHC s Nmap2 system which includes level differencing options based on the available levels in the gridded datasets. This example depicts the HWRF shear calculation with Hurricane Igor and Hurricane Julia official forecast overlays. 33

31 HFIP FY10 Deliverable User-Selectable Shear for Global, Hurricane and Regional Models Model Type Shear Selections CMC Global , , ECMWF Global , , GFS Global , , , , , NOGAPS Global , UKMET Global , , , , JMA Global , , , , HWRF Hurricane , , , , , GFDL Hurricane , , , , , NAM Regional , , , , , , RUC Regional , , , , Table outlines the different shear level options available in NHC s Nmap2 display system for the Global, Hurricane and Regional models received operationally at NHC. 34

32 6.5.1 HFIP funded ATCF Upgrades Web-ATCF (next slide) TC genesis probabilities Forecast dialog Database Expanded objective aid format Allows more HFIP-related information Documentation expanded Probability data format Allows many parameters, formats (e.g., RI) Read/write infrastructure in ATCF 60-h Forecast capability Display capability expanded to 1300 aids (ensembles) GPCE version of NHC Wind Speed Probabilities 35

33 Web-ATCF Server Live data feed Windows or Linux Password available on request Objective Aids Fixes Forecasts Satellite imagery Raw data Gridded fields Statistics 36

34 Bonus HFIP Deliverable FY2010 #1: Capability for NHC to generate Day 6 and 7 forecasts for internal testing (Joint HFIP and JHT work by Schrader and Sampson) Added capability to create Day 6 and Day 7 forecast positions. Also, ATCF now has capacity to store more genesis probability information. 37

35 Bonus HFIP Deliverable FY2010 #2: Increased Maximum Number of Ensemble Members (Joint HFIP and JHT work by Schrader and Sampson) View model runs simultaneously in ATCF 38

36 Bonus HFIP FY10 Deliverable #3: Point-and-click vertical cross sections NAWIPS cross-section capability developed by Mike Brennan. Given user input (model, cycle, click-anddrag path of cross section), a script runs from a terminal window and generates a vertical cross section. This tool is impacting forecasts (see below). Dry layer Tilted vortex Excerpt from recent NHC forecast discussion: THERE CONTINUES TO BE LOW CONFIDENCE IN THE INTENSITY FORECAST. NONE OF THE INTENSITY MODELS SHOW MUCH STRENGTHENING BEYOND ABOUT 90 KT. THE 200-MB FLOW OVER THE CARIBBEAN SEA SHOULD BE FAVORABLE FOR STRENGTHENING...ESPECIALLY IF TOMAS MAINTAINS A MORE SOUTHERN COURSE. HOWEVER...CROSS-SECTION ANALYSES OF THE GUIDANCE DOES INDICATE THAT THE CYCLONE VORTEX COULD BECOME A LITTLE TILTED OVER THE NEXT 48 HOURS OR SO...POSSIBLY DUE TO SHEAR BELOW THE OUTFLOW LAYER. BECAUSE OF THESE MIXED SIGNALS...THE OFFICIAL INTENSITY FORECAST ONLY SHOWS MODEST STRENGTHENING THROUGH 24 HOURS AND ESSENTIALLY HOLDS THE INTENSITY FAIRLY STEADY THEREAFTER Above: West-East vertical cross section from a GFS forecast of Tomas, showing midlevel drying (shading) and eastward tilt of the vortex (contours). Courtesy Robbie Berg, HSU/NHC. BERG/FRANKLIN 30 OCT 2010, TOMAS 39

37 NHC Diagnostics NHC has developed a system to compare vortex structure among the models in real time. Output is posted on internal webpage and products are hard-coded. Results from the 2010 season have brought to light some important issues with the current models, e.g. the vortex initialization. GFS HWRF GFDL Above: South-north (top) and West-East (bottom) vertical cross sections from the GFS (left), HWRF (middle), and GFDL (right). Vorticity is contoured. Note the differences in the vortex structure among the models. 40

38 NHC Diagnostics, cont. HWRF vortex too broad By archiving and analyzing all of the output from the new real-time diagnostic system, we have been able to identify some systematic characteristics of the various models, e.g. the excessive breadth (compared to other models and observations) of the HWRF vortex in some cases 41

39 NHC Diagnostics, cont. and identify some special cases. For example, the GFS was unable to properly represent the recent Hurricane Paula. No storm in GFS Above: Near-surface initial wind fields from the GFS (left), HWRF (middle), and GFDL (right). Note that the GFS does not see a hurricane. 42

40 HPLOT EMC/NHC Collaboration Analysis and forecast display tool Grads-based GUI, Launches from a terminal window, Facilitates analysis of HWRF/GFDL models, Executes analysis commands very quickly 43

41 HPLOT Point-and Click Cross Sections Point and click cross sections 44

42 HPLOT User-selectable Skew-T User-selectable areaaveraged (or point) thermodynamic profiles. 45

43 Looking toward FY2011 HPLOT and future product training for NHC Obtain and address FY2011 specialists needs NHC-EMC collaborative activities New contractor hire 46

44 Proposed ATCF Improvements New NHC Requirements Other possibilities Pressure/Wind relationships Ocean Heat Content, SST display Mac capability for Web-ATCF GTK+ for graphics (vice Motif) Click to fix capability for imagery RI probability ingest and storage Live feeds of HFIP objective aids Moveable grid data ingest/display SHIPS/STIPS/LGEM diagnostic file computation Shear product computation and display *Improvements to ATCF rapidly transition to operations 47