1. Impact of US Polar Orbital Satellite Data Loss on Nation’s High Impact Weather Forecast Capability: Summary of Recent Data Denial Experiments Environmental Modeling Center National Centers for Environmental Prediction Camp Springs, Maryland, USA

2. Overview Performance Metrics Satellite Data Impact on Forecast Error Global NWP Impact of Actual Satellite Data Loss Global NWP Impact of Satellite Denial Experiments – Data Denial Experiment Background – Data Denial Experiment Design – Case 1: “Snowmageddon” Storm – Feb 5-6, 2010 – Case 2: NYC and New England Blizzard – Dec 26-27, 2010 – Case 3: Northern Pacific Alaskan Coastal Storm – Case 4: Tornado Outbreak – Case 5: Tornado Outbreak Summary

3. Performance Metrics Satellite data is required to help meet key performance metrics : – Numerical Weather Prediction Global Anomaly Correlation Score – “Internal” metric Related to ability to meet service-based metrics (below) – National Weather Service GPRA* Metrics (* Government Performance & Results Act) Hurricane Track and Intensity Forecast Accuracy Winter Storm Warning Lead Time and Accuracy Precipitation Threat Accuracy Flood Warning Lead Time and Accuracy Marine Windspeed and Wave Height Forecast Accuracy

4. Satellite Data Impact on Forecast Error Statistical estimation of relative improvement made to operational NWP system by various observing systems Based on adjoint model sensitivity technique

5. Operational ECMWF system September to December 2008. Averaged over all model layers and entire global atmosphere. % contribution of different observations to reduction in forecast error. Courtesy: Carla Cardinali and Sean Healy, ECMWF Forecast error contribution (%) Note: 1) Sounders on Polar Satellites reduce forecast error most 2) Results are relevant for other NWP Centers, including NWS/NCEP

6. Global NWP Impact of Actual Satellite Data Loss January 1, 2009 telecommunications outage Substantial fractions of polar satellite sensor data not available for model assimilation – Illustrated on following slide Numerical Weather Prediction centers of all nations suffered degraded skill in their global forecasting systems as result of lost data – See slide after next

7. 18% loss 36% loss 33% loss 100% loss 32% loss AMSUA HIRS4 GPSRO AIRS MHS QSCAT New Year’s Day 2009 Satellite Data Loss 7

8. Global Model Skill Scores After 5-6 Days Southern Hemisphere (SH) Dropouts (score below 0.7) for the GFS, UKMET, CAN, and FNMOC models; European Center skill degraded 8

9. Global NWP Impact of Satellite Data Denial Experiments

10. Data Denial Experiment Background Data denial experiments designed to assess impact of having no afternoon polar orbiter data on the accuracy of Global NWP – Intended to contribute to discussions concerning the impact of a potential gap in p.m. polar orbit coverage expected to result from JPSS launch delay Data denial case studies selected to include significant weather events that affected lives, property, and commerce for US citizens

11. 11 Data Denial Experiment Design Control: NCEP operational 27 Km Global Forecast Model (GFS) Experiment: Model T574L64. All U.S. Polar Orbiting Satellite data from afternoon orbits was removed from the experimental runs Satellite data removed: AMSUA_NOAA15, AMSUB_NOAA15, HIRS3_NOAA17, AMSUA_NOAA18, MHS_NOAA18, AIRS_AQUA, AMSUA_AQUA,HIRS4_NOAA19, & AMSUA_NOAA19 Satellite data kept: HIRS4_METOP-A, AMSUA_METOP-A, MHS_METOP-A, IASI_METOP-A, & ALL GOES SOUNDERS.

12. Case 1: “Snowmageddon” Storm – Feb 5-6, 2010

13. 13 24 Hr Accumulated Snowfall Totals (inches) for 6 Feb (am) Case Study 1: “Snowmageddon” February 6, 2010 Forecast comparison using NOAA’s Polar-orbiting Operational Environmental Satellite Data Observed Snowfall 5 day forecast with all NOAA orbiting satellite data 5 day forecast without NOAA afternoon orbiting satellite data Result: In DC and Mid-Atlantic coast, models without NOAA orbiting satellite data did not forecast this paralyzing event and under-forecast snow by at least 10 inches Impacts: Aircraft and airline passengers would have been stranded, ground commerce would have been halted with no mitigation plans, population would have been unprepared for paralyzing snow-depth 0.1 1 2.5 5 7.5 10 12.5 15 17.5 20 25 30 40 50 60 70 80 90 15-18” Forecast 7-10” Forecast 15-22” Actual

14. 14 Observed 27 Km Operational NWP Data Denial NWP 6 Feb: Models without PM data under-forecasted snow totals: – Operational forecast shows paralyzing event – Data Denial Did not forecast paralyzing event in DC— at least 10” too low at Day 5 Low confidence in extreme snowfall at this point – Future errors of this scale could result in: Aircraft and airline passengers stranded Ground commerce halted with no mitigation plans Population unprepared for paralyzing snow-depth Wash DC 5-day Forecast: - With Data: Historical, paralyzing event - Data Denial: Significant; but not paralyzing 15-18” Forecast 7-10” Forecast 15-22” Actual Results 24 Hr Accumulated Precipitation Totals for 6 Feb (am) 5 Day Forecast Forecast Period: 5 Feb (am) – 6 Feb (am)

15. 15 Observed 27 Km Operational NWP Data Denial NWP 6 Feb: Models without PM data under- forecasted snow totals: – Operational forecast still superior Increasing confidence in the forecast for major snow event – Data Denial forecast Some improvement in snow forecast in DC area Still under-forecasting total precipitation maximum in VA and Carolinas by 50% 15-18” Forecast 12-15” Forecast Wash DC 4-day Forecast: - With Data: Historical, paralyzing event - Data Denial: Significant event 15-22” Actual Results 24 Hr Accumulated Precipitation Totals for 6 Feb (am) 4 Day Forecast Forecast Period: 5 Feb (am) – 6 Feb (am)

16. 16 Observed 27 Km Operational NWP Data Denial NWP 6 Feb: Models without PM data under- forecasted snow totals: – Operational forecast Shows lower snowfall totals in DC and surrounding area than previous model run but still superior to data denial Paralyzing event just south of DC Confidence in forecast for a major snow event still high – Data Denial Lower forecast snowfall totals – decreased confidence in extreme event affecting Delmarva area Still under-forecasting total precipitation maximum in VA and Carolinas by 50% Wash DC 3-day Forecast: -With Data: Significant Event DC; paralyzing event south - Data Denial: Significant Event 15-22” Actual Results 24 Hr Accumulated Precipitation Totals for 6 Feb (am) 3 Day Forecast Forecast Period: 5 Feb (am) – 6 Feb (am)

17. Case Study 1: Heavy Rainfall Event in Southern US February 5, 2010 Forecast comparison using NOAA’s Polar-orbiting Operational Environmental Satellite data 17 Observed Precipitation 1 day forecast with all NOAA orbiting satellite data 24 Hr Accumulated Precipitation Totals (inches) for 5 Feb (am) 1 day forecast without NOAA afternoon satellite data Result: Up to a 50% error increase in precipitation rates in southern US Impacts: Future errors of this scale could result in flood forecast error providing less time for population to react and increasing risk to life and property (hours vs days)

18. Results 24 Hr Accumulated Precipitation Totals for 5 Feb (am) One-day 18 Observed27 Km Operational NWP Data Denial NWP Up to a 50% error in precipitation rates and areal coverage in southern US – Future flood forecasts could be negatively impacted—less time for population to react increasing risk to life and property Forecast Period: 4 Feb (am) – 5 Feb (pm)

19. Case Study 1: Conclusion Accurate 3-, 4-, 5-day forecasts of the Feb 5-6 2010 East Coast Storms with PM satellite data mitigated effects on population and commerce – Had PM satellite data not been available less effective mitigation would have been possible NWS alerted, with enough advance warning to take action: – Communities and Individual Citizens – State Governments and Emergency Managers – FAA and Commercial Airlines; – Surface Transportation Industry; – Schools; – Power Industry 19

20. Case Study 2: NYC and New England Blizzard of Dec 26-27, 2010

21. 96-hr Forecasts of Sea-Level Pressure Valid for 12Z27Dec2010 operationalData Denial

22. 72-hr Forecasts of Sea-Level Pressure Valid for 12Z27Dec2010 operationalData Denial

23. 24-hr Accumulated Precipitation 12Z 26Dec2010 ~ 12Z 27Dec2010 3.5 - Day Forecast “Denial” missed

24. 24-hr Accumulated Precipitation 12Z 26Dec2010 ~ 12Z 27Dec2010 3 - Day Forecast Both captured the storm well Denial underpredicted

25. Case Study 2: Conclusion For forecasts extending back to 3 and 4 days the “Data Denial” experiment under-predicted the storm that hit the NYC area. The “operational” forecast captured this storm in most of the cycles, though in a couple of cycles the storm was too weak and too far to the east. For forecasts extending back to 3 and 4 days, the coastal low pressure system in the “Data Denial” experiment was much weaker than that in the “operational” run. For forecasts shorter than 2.5 (I would say 3) days, both the “Operational” and “Data Denial” captured the storm quite well, both slightly over-predicting the snowfall along the East Coast. These good forecasts were likely the result of the good RAOB coverage over North America upstream of the region of rapid cyclogenesis 25

26. Case Study 3 : Northern Pacific Alaskan Coastal Storm, April 6-7, 2011 A significant storm occurred on April 6-7, 2011. It affected the Bering Sea, west coast of Alaska, south central Alaska, and portions of the interior. Damaging winds, blizzard condition, dangerous seas and sea ice conditions, and severe turbulence from hurricane force storm occurred in conjunction with this storm. Wind gust of 110 MPH (~50 m/s) was observed around 4 PM local time on 04/06/2011 ( 1 AM UTC on 04/07/2011).

27. 96-hr Forecasts Valid for 12Z 7Apr2011 operational Data Denial

28. 72-hr Forecasts Valid for 12Z 7Apr2011 operational Data Denial

29. Case Study 3: Conclusion Local Impact over the Bering Sea region: the data-denial experiment did not show any systematic difference from the operational model. Storm tracks, sea-level pressure, surface winds and precipitation from the data-denial experiment are similar to those of the operational model. Hemispheric Large-Scale Impact: the data-denial experiment is not significantly different from the operational model except that it has larger wind RMS errors. 29

30. Case Study 4 : Tornado Outbreak April 14-16, 2011

31. GSF 6-day Forecast for CAPE on April 14, 2011 GSF Forecasts of Convective Available Potential Energy (CAPE) for April 14, 2011 indicate similar 6 day outlook for severe storms over the United States whether p.m. polar satellite data is included (top) or not (bottom).

32. Case Study 4: Conclusion Briefly, no impact (refer to tornado two-pager to elaborate) 32

33. Case Study 5 : Tornado Outbreak April 25-28, 2011 An extremely large and violent tornado outbreak, the largest in United States history and popularly known as the 2011 Super Outbreak, occurred from April 25 to 28, 2011. The outbreak affected the Southern, Midwestern, and Northeastern United States, leaving catastrophic destruction in its wake, especially across the state of Alabama. It produced destructive tornadoes in Alabama, Arkansas, Georgia, Mississippi, North Carolina, Tennessee, and Virginia, and affected several other areas throughout the Southern and Eastern United States with a total of 332 tornadoes confirmed in 21 states from Texas to New York and even isolated tornadoes in Canada. Widespread and destructive tornadoes occurred on each day of the outbreak, with April 27 being among the most prolific and destructive tornado days in United States history.

34. 24-hr Accumulated Precip from 12Z 26Apr2011 to 12Z 27Apr2011 Day-7Day-6 Day-5 Day-4

35. 24-hr Accumulated Precip from 12Z 26Apr2011 to 12Z 27Apr2011 Day-3Day-2 Day-1 Not much difference between the control and data denial run. Both made good forecasts.

36. Forecasts from Different Cycles Valid at 00Z 28Apr2011 Day-7 Day-6 Day-5 Day-4

37. Forecasts from Different Cycles Valid at 00Z 27Apr2011 Day-3 Day-2 Day-1 Not much difference between the control and data denial run. Both made good forecasts starting from day-5 forecasts.

38. Case Study 5: Conclusion Little Impact from loss od p.m. satellite data: Both the control and experimental data-denial runs made good forecasts of precipitation and cyclone tracks over the central to eastern US five days back from the 27-28 April 2001 tornado outbreaks. Hemispheric mean verification statistics such as AC and RMSE of the data denial run are slightly worse than in the control run. The differences of CAPE at the lowest 180 hPa above ground and Storm Relative Helicity in the lowest 3 kilometers between the control run and data-denial run are small for all forecast time up to 168 hours. 38

39. Summary Five Data Denial Studies have been conducted to help evaluate the implications of a gap in afternoon polar satellite coverage Forecasts at 3 days & longer of significant weather episodes made w/o p.m. satellite data were – Significantly degraded in two cases – Largely unchanged in three cases Critical atmospheric signals detected by other observations These results are consistent with previous observation sensitivity studies Results underscore the need for continuity of both afternoon and morning polar satellite coverage

40. Temporary Backups