Hybrid variational-ensemble data assimilation at NCEP

NCEP GFS Forecasts From ECMWF Analysis Physics, Resolution, Assimilation? Jordan Alpert1 DaNa Carlis1, Krishna Kumar2,3, Bradley Ballish2 EMC 2 NCO 3 DELL Federal 1 Aksel Wiin-Nielsen Symposium 24 January 2012, New Orleans, Louisiana 1 AMS 92nd Annual Meeting Outline

What are forecast busts or Dropouts? GFS versus ECMWF forecast divergence NCEP dropout prediction Use of ECMWF analyses to create initial conditions (ICs) to study GFS dropouts (ECM runs) Observing System Experiments (OSEs) and Quality Control Low Skill Event Initial Condition Source Location Comparison of ECMWF and GFS forecasts a case study for the GFS Northern Hemisphere big dropout on 00Z November 25 th, 2011 Summary 2

GFS Forecast Skill (all cycles) versus ECMWF (current) On approximately a monthly basis, poor forecasts or Skill Score Dropouts plague GFS performance (see right). Big Dropout European Centre for Mediumrange Weather Forecasts (ECMWF), often does not exhibit this loss in skill. Our goal is to find differences that can lead to algorithms to detect and correct a data quality control (QC) problem, bias correction, and analysis issues in initial conditions before the forecast begins

Dropout Forecast Divergence (GFS/ECMWF) implemented to predict dropouts next slide 3 The Forecast-Forecast (F-F) Correlations may be computed between any two models, eg., GFS and the ECMWF, using the ECMWF as verifying data. NCOs GFS vs. ECMWF Forecast Divergence 2011120100 Dropout Prediction 5-days in advance

Predicted 5 days in advance The F-F Correlations can now be used to predict the occurrence of a dropout 5 days in advance as for this dropout case. Predicted 5 days in advance Confirmation of Low Skill (Dropout) Event Source Location

Forecast Divergence of GFS and ECMWF operations gives a heads up for model problems and extreme code analysis of observations shows difference diagnostics with the GFS GDAS analysis and with ECMWF analyses Use ECMWF analysis to generate Pseudo Obs for input to the Grid Statistical Interpolation (GSI) to make GFS forecasts with ECMWF model information content called ECM experiments. The result is an analysis that is a grand interpolation for GFS ICs from the original ECMWF information content. These analyses are used in conjunction with GFS production surface and fix field archives as ICs for 5-day forecast experiments. Experiments can be run for an overlay patch area (box) from either ECMWF over GFS or GFS over ECMWF to check the potential source location for the dropout. If the area patch or box of data is the determining factor causing the low skill in the initial condition, and when the ECMWF patch is laid over GFS global fields, then ECMWF skill is retrieved. And if the GFS (same) area patch is laid over ECMWF global IC fields then GFS skill should be obtained.

5 Is the dropout due to model problems or analysis problems? The GSI as a Grand Interpolator Treating the ECMWF 1x1 degree grid analysis as pseudo observations (eg., RAOBS), and using them as sole input into the GFS (GSI) analysis, which then acts as a grand interpolator, generating new initial conditions that inherit ECMWF analysis (and to some extent ECMWF model) system characteristics. This analysis/forecast is labelled as ECM runs. To analyze forecast skill dropouts, we compare the operational GFS and ECMWF analyses and provide a way to study the impact of observations with controlled experiments. PSEUDO ECM OBS GFS Production GUESS ------RUN GSI-----ECM ANLYSIS (00Z) First Time

ECMWF INITIAL CONDITIONS FOR GFS FORECASTS ECM cycled Runs Make 6-h GFS forecast 00Z Pseudo OBS 6-h GUESS ----RUN GSI---ECM ANALYSIS 06Z Pseudo OBS 6-h GUESS ----RUN GSI---ECM ANALYSIS Make 6-h GFS forecast

Representation of an ECM Make 6-h GFS forecast run using the GSI/GFS and ECMWF 14 level pressure and 18Z Pseudo OBS 6-h GUESS 1x1 degree analysis files. ----RUN GSI---ECM ANALYSIS Make 6-h GFS forecast 12Z Pseudo OBS 6-h GUESS ----RUN GSI---ECM ANALYSIS INPUT ---Run GSI--OUTPUT ECM pgbanl IC in Green,

ECMWF Production in Red, GFS Production in Black 7 5 Day Anomaly Correlation Scores at 500 hPa for Dropout Cases ECM Performs Better than GFS (NH) 2007-2008 Lat/Lon Box 20N 80N 150E 230E IC Date GFS

ECMWF ECM OVRLY 2007102212 0.61 0.87 0.89 0.90 1 500 mb Anomaly Correlation

0.9 0.8 GFS ECMWF ECM OVRLY 0.7 0.6 0.5 0.4 2007102112 2007102200 2007102212 2007102312

2008021712 2008030112 2008030400 2008030412 Initialization Date ECM runs (blue) are a good representation for ECMWF analysis OVRLY runs (green) with ECM psuedo-obs over the Central Pacific drastically improve8 two 8 October 2007 dropout cases (102200 & 102212). Observing System Experiments (OSEs) OSE SH 5-day AC Scores for 2010092100 Dropout 0.9 5-day Anomaly Correlation 0.85 0.8

0.73 0.75 0.69 0.7 0.65 0.64 0.6 0.55 0.5 0.45 0.4 OSE results show that the top 2 observation types within the target area were partially responsible for the poor skill of the GFS 5-day forecast.

Without GPSRO and AMSUA observations the 5-day AC score would be 10 AC points higher than the GDAS CONTROL. 0.74 Comparing GFS and ECMWF Forecasts: 2011112500 Cycle fh00 wf ? 10 Courtesy: Fanglin Yang Comparing GFS and ECMWF Forecasts: 2011112500 Cycle fh24 wf

Courtesy: Fanglin Yang 11 Comparing GFS and ECMWF Forecasts: 2011112500 Cycle fh120 wf wf 12 Courtesy: Fanglin Yang GFS and ECMWF Forecasts: NH 2011112500

13 GFS Rainfall Forecasts over CONUS from 2011112500 Cycle ECMWF Cut-off low GFS: the trough moved too fast east. Failed to predict the cut-off low pressure system. Courtesy: Fanglin Yang 84-108hr fcst missed the rainfall along the East Coast from Monday to Tuesday 14 GFS forecast troughs (blue) displaced compared to its analysis (white) ECMWF (right) and GFS (left) analyses on 30 Nov 2011 00Z look similar

15 36 hrs 60 hrs 2011112700_f036_5640 2011112812_f000_5640 Blue-ECMWF members (50) Tan GFS (20) Green CMC (20) White Verification, Solid red 00Z ECMWF ensemble mean Solid brown 00Z GEFS mean 2011112600_f060_564 0 2011112812_f000_564 0 84 hrs

2011112500_f084_5640 2011112812_f000_5640 108 hrs GFS has much bigger spread than normal 2011112400_f108_5640 2011112812_f000_5640 16 Difference plot shows the ECMWF winds are stronger in the select area 17 18 GFS Production

30NOV (IC 25NOV) 00Z 0.544 06Z 0.684 12Z 0.756 18Z 0.840 ECM Pseudo Obs 30NOV (IC 25NOV) 00Z 0.743 06Z 0.783 12Z 0.723 18Z 0.804

ECMWF Operations 30NOV (IC 25NOV) 00Z 0.841 06Z N/A 12Z 0.857 18Z N/A 19 wf WF 20

G 0.743 wf G 0.841 ECM WF 21 G 0.743 0.631 wf 0.606

wf 0.644 0.637 22 EnKF Q3FY12 Hybrid GDAS Package Parallel - Northern Hemisphere 23 Some Differences Between ECMWF and GFS Model Physics - GFS & ECMWF physics are comparable The higher resolution of ECMWF model and physics may make a big difference in forecasting precipitation amounts, biases etc. Data Assimilation, Quality Control and Data Monitoring ECMWF has more resources to develop and monitor a robust QC system than NCEPs ECMWFs bias corrections of observations are different than NCEPs ECMWF runs far less models than NCEP and does not have the time pressure to disseminate

model forecasts at an early time like NCEP ECMWF assimilates observations in a 12-hour time window with a later data cut-off time . This may have beneficial effects to their analysis system by using additional satellite data with a wider geographic domain NCEP has a 6-hour time window with an early data cut-off time ECMWFs assimilation system runs a 4DVAR rather than NCEPs 3DVAR 4DVAR system can assimilate data at off times better than 3DVAR

Recently Viewed Presentations

  • SHRM-CP Exam

    SHRM-CP Exam

    CBA Financial Support. Dr. Guo has obtained a commitment from the College of Business Administration to reimburse students who take the exam for half of the exam fee and cost of study materials [up to approximately $500] provided that they...
  • www.sausd.us


    End ofUnitPerformanceTask:The Digital Survival Challengewill ask students to take what theyhavelearned about in theLord oftheFlies unitand applyit to creatingtheirown society.This will challengetheirthinkingprocess, strengthen their collaboration skills, andwill provide an inquiry-based learningexperience.
  • Answers The Bible Questions - Word and Sword

    Answers The Bible Questions - Word and Sword

    SIN. Isa. 53. CHRIST. TOMB. Grace. Eph. 1:7 In Him we have redemption through His blood, the forgiveness of sins, according to the riches of His grace. Eph. 2:8 For by grace you have been saved through faith, and that...
  • The Breastplate of Righteousness - Woodburn Bible Church

    The Breastplate of Righteousness - Woodburn Bible Church

    The Breastplate Symbolizes Christ's Righteousness. When Jesus gave His life for us, 2 major things happened: He took our sin upon Himself. He gave us His righteousness. When God looks at us, he sees the righteousness of Christ - This...
  • UCL Centre for Medical Image Computing Poster Title

    UCL Centre for Medical Image Computing Poster Title

    Poster Title Author/s UCL Centre for Medical Image Computing. Title: PowerPoint Presentation Author: Media Resources Last modified by: ucacrpg Created Date: 7/18/2005 1:37:23 PM ... UCL Other titles: Times Arial Calibri Blank Presentation Slide 1 ...
  • Briefing on PathAguard Product Range - Focus Healthcare Products

    Briefing on PathAguard Product Range - Focus Healthcare Products

    To use as it is impregnated with silver-ion technology providing an antimicrobial barrier . Protection against cross infection as the antimicrobial effect is maintained for the lifetime of the product. Ion-silver is used in many medical applications to help the...
  • 「關懷失智.憶不容遲」 網路問卷調查活動報告

    「關懷失智.憶不容遲」 網路問卷調查活動報告

    失智症會不會找上我? 天主教失智老人基金會 台北市萬華老人服務中心
  • Remembering the Past Why do we still need

    Remembering the Past Why do we still need

    The poppy is a symbol of hope and Remembrance. It reminds us of all those people whose lives are affected by warsand asks us to think of these people. It makes us think about how important peace is in all...