Using Reconnaissance Data in Weather Models
Jason Sippel NOAA AOML/HRD 2021 SECART series Error trends
Official TC Track Forecast Errors: • Hurricane track forecasts 1990-2020 have improved markedly
300 • The average Day-3 forecast location error is 200 now about what Day-1 error was in 1990 100 • These improvements are 1990 2020 largely tied to improvements in large- scale forecasts Error trends
• Hurricane track forecasts have improved markedly
• The average Day-3 forecast location error is now about what Day-1 error was in 1990
• These improvements are largely tied to improvements in large- scale forecasts Error trends
Official TC Intensity Forecast Errors: 1990-2020 • Hurricane intensity 30 forecasts have only recently improved 20 • Improvement in intensity 10 forecast largely corresponds with commencement of 0 1990 2020 Hurricane Forecast Improvement Project HFIP era Error trends
HWRF Intensity Skill 40 • Significant focus of HFIP has been the 20 development of the HWRF better 0 Climo better HWRF model -20
-40 • As a result, HWRF intensity has improved
Day 1 Day 3 Day 5 significantly over the past decade HWRF skill has improved up to 60%! Michael
Talk focus: How better use of data, particularly from recon, has helped improve forecasts Michael
Talk focus: How better use of data, particularly from recon, has helped improve forecasts Using TC Observations
• US has used dropsondes Dropsonde impact on GFS TC track for TC model forecast 20 improvement since 1997
0 With drops better • Significant track With drops worse improvement globally -20 • Consistent across many Day 1 Day 3 Day 5 studies Impact of dropsondes in September 2008 Using TC Observations
Observations Analysis • Starting in 2008, it became apparent that assimilating 88D Doppler velocity could improve coastal TC forecasts Observations Analysis • Assimilating radar data significantly improved analyses and forecasts of Hurricane Humberto Using TC Observations
Fcst. & Obs. Maximum winds 40 Observation Forecasts • Starting in 2008, it became apparent that 20 assimilating 88D Doppler velocity could improve NO Doppler coastal TC forecasts 0 40 • Assimilating radar data significantly improved 20 analyses and forecasts of Hurricane Humberto WITH Doppler 0 Date Using TC Observations
Experimental & Operational Operational Wind Errors (No TDR) • Subsequent work 20 showed forecast
15 improvements from assimilating tail Doppler 10 radar (TDR) velocity from NOAA recon 5 Experimental (with TDR)
Day 1 Day 3 Day 5 • These results led to a Maximum wind errors from operational dedicated effort to forecasts (no TDR) and an experimental system that assimilated TDR data. assimilate TDR operationally Using TC Observations
Fcst. & Obs. Maximum 80windsNO TDR DATA HWRF forecast • TDR data began being assimilated in HWRF in 2013 40 * * Observed • For weak storms like 80 WITH TDR DATA Karen (left), there was substantial improvement of a positive intensity 40 * bias in HWRF *
Day 1 Day 3 Day 5 HWRF DA improvements HWRF improvements
CURRENT OBSERVATIONS ASSIMILATED BY HWRF INCLUDE: • Conventional observations (radiosondes, dropwindsondes, aircraft, ships, buoys, surface observations over land, scatterometer, etc) • NEXRAD 88-D Doppler velocity • ALL reconnaissance (HDOB, TDR) • Atmospheric motion vectors • Clear-sky satellite radiance observations HWRF improvements
• Recon benefit assessed in 2016-2018 Intensity error in 2019 HWRF high impact storms No recon 20 Recon • Many major hurricanes in this 10 sample
0 • Recon has a clear Day1 Day2 Day3 Day4 Day5 positive impact on intensity, 10-15% improvement through 72h Recent Performance
Intensity skill: Near-CONUS • Model intensity skill varies HWRF greatly by region 60 CTCX IVCN • Highest skill is where we have 40
the most data (esp. HWRF) 20
0 Where the P3 flies (circles) Intensity skill: MDR HWRF 60 CTCX Near- IVCN CONUS 40 MDR 20
0 Day 1 Day 3 Day 5 Recent Changes
Example of end-point drop positions “End-point” dropsondes from USAF C-130 missions • Dropsondes at end-points of “alpha” pattern from C-130 missions tested in 2017
• Data denial tests suggested a Impact on intensity skill 10% impact on intensity skill 30
15 With drops better • Based on these results, this 0 practice was implemented -15 With drops worse operationally in 2018 -30 Day 1 Day 3 Day 5 Recent Changes
• Upgrade to GFSV16 in March included better Additional reconHB20 impact (basin-scale on GFS H220) track use of dropsondes and HB20 – no dropsondes flight-level data 12 8 • Added data improves 4 MesonetAdded datatest: better Intensity Error (kt) entire NATL sample track 0 Added data worse by ~5% -4 Day 1 Day 3 Day 5 Day 7 • Higher impact in cycles with data & strong storms Ongoing developments
• Ongoing work assessing how best to deploy dropsondes Dropsonde Test: Intensity Error using basin-scale HWRF ALL DROPSONDES NO DROPSONDES
• Dropsondes directly benefit track by 5-10% and intensity by 10-15% Mesonet test: Intensity Error (kt)
• Removing dropsondes anywhere (e.g., inner core vs. environment, etc.) has Day 1 Day 3 Day 5 negative consequences Conclusions
• NOAA TC prediction is undergoing dramatic advancements, lead by improvements in global models and HWRF
• We are using more of the available data in DA
• DA results are guiding us on how to approach reconnaissance, which should further improve forecasts Future direction: HAFS (Hurricane Analysis and Forecast System)