Ensemble variability in rainfall forecasts of Hurricane Irene

Ensemble variability in rainfall forecasts of Hurricane Irene (2011) Molly Smith, Ryan Torn, Kristen Corbosiero, and Philip Pegion NWS Focal Points: Steve Di. Rienzo and Mike Jurewicz Fall 2016 CSTAR Meeting 2 November, 2016

Motivation • Landfalling TCs are often associated with widespread heavy precipitation, which can lead to devastating flood events.

Motivation • Landfalling TCs are often associated with widespread heavy precipitation, which can lead to devastating flood events. • Relatively few studies have explored the sensitivity of precipitation forecasts for such events to model initial conditions.

Motivation • Landfalling TCs are often associated with widespread heavy precipitation, which can lead to devastating flood events. • Relatively few studies have explored the sensitivity of precipitation forecasts for such events to model initial conditions. • What modulates precipitation variability in ensemble modeling of heavy rainfall events?

Motivation • Landfalling TCs are often associated with widespread heavy precipitation, which can lead to devastating flood events. • Relatively few studies have explored the sensitivity of precipitation forecasts for such events to model initial conditions. • What modulates precipitation variability in ensemble forecasts of heavy rainfall events? • This work uses Hurricane Irene (2011) as a case study. • Irene featured catastrophic inland flooding caused by widespread rainfall totals of 4– 7 inches. • The Catskill region of New York received up to 12 inches.

Methods • An 80 -member GFS ensemble was initialized at 0000 UTC 27 August 2011 and run for 48 hours, until 0000 UTC 29 August. • These GFS ensemble forecasts were created with the operational version of the GFS in use from 2013– 2015.

GFS Ensemble Mean Precipitation GFS Predicted Accumulations (mm) Observation source: EOL

Methods • An 80 -member GFS ensemble was initialized at 0000 UTC 27 August 2011 and run for 48 hours, until 0000 UTC 29 August. • These GFS ensemble forecasts were created with the operational version of the GFS in use from 2013– 2015. • Members were ranked by the amount of precipitation they brought to the Catskill region of New York during this time period.

Methods • An 80 -member GFS ensemble was initialized at 0000 UTC 27 August 2011 and run for 48 hours, until 0000 UTC 29 August. • These GFS ensemble forecasts were created with the operational version of the GFS in use from 2013– 2015. • Members were ranked by the amount of precipitation they brought to the Catskill region of New York during this time period. • The synoptic characteristics of the ten wettest ensemble members were then compared to those of the ten driest, to see if any large-scale patterns were behind the differences in forecasted precipitation.

Why did some storms track father west?

Hypothesis: Differences in storm track are associated with differences in steering flow.

Composite Difference Plots 250 -850 h. Pa Zonal Steering Flow at Initialization easterly anomalies westerly anomalies

Composite Difference Plots 250 -850 h. Pa Zonal Steering Flow at Initialization Comparison of 10 wettest ensemble members and 10 driest easterly anomalies westerly anomalies

Composite Difference Plots 250 -850 h. Pa Zonal Steering Flow at Initialization Comparison of 10 wettest ensemble members and 10 driest Vectors: Ensemble mean easterly anomalies westerly anomalies

Composite Difference Plots 250 -850 h. Pa Zonal Steering Flow at Initialization Colors: Standardized difference between wet and dry members Comparison of 10 wettest ensemble members and 10 driest Warm colors: Wet members have greater value Vectors: Ensemble mean Cold colors: Wet members have lesser value easterly anomalies westerly anomalies

Composite Difference Plots 250 -850 h. Pa Zonal Steering Flow at Initialization Colors: Standardized difference between wet and dry members Comparison of 10 wettest ensemble members and 10 driest Vectors: Ensemble mean Warm colors: Wet members have greater value Stippling: Significant at 95% level Cold colors: Wet members have lesser value easterly anomalies westerly anomalies

250 -850 h. Pa Zonal Steering – 00 hr Easterly flow anomalies arise early in the wetter members. easterly anomalies westerly anomalies

Potential vorticity (PV) anomalies are associated with wind flow anomalies. Is that a factor here?

350 K PV – 00 hr Wetter members are initialized with cyclonic PV anomalies to the south and west of Irene. These anomalies help to steer Irene inland. Less cyclonic PV More cyclonic PV

350 K PV – 06 hr Wetter members are initialized with cyclonic PV anomalies to the south and west of Irene. These anomalies help to steer Irene inland. Less cyclonic PV More cyclonic PV

350 K PV – 18 hr Wetter members are initialized with cyclonic PV anomalies to the south and west of Irene. These anomalies help to steer Irene inland. Less cyclonic PV More cyclonic PV

350 K PV – 36 hr Wetter members are initialized with cyclonic PV anomalies to the south and west of Irene. These anomalies help to steer Irene inland. Less cyclonic PV More cyclonic PV

With what mechanism did Irene slow the approaching trough?

250 h. Pa Divergent U – 15 hr In the wetter members, Irene experienced greater outflow on its western side. This outflow may have contributed to blocking the approaching trough. easterly anomalies westerly anomalies

Downscaling • However, the GFS is less than ideal for modeling terrain and mesoscale processes. • Our 0. 5 degree GFS output was downscaled to 3 km using WRF, with physics comparable to those employed in HRRR.

WRF 3 km Ensemble Mean Precipitation WRF 3 km Predicted Accumulations (mm) Observation source: EOL

WRF 3 km Ensemble Precipitation Spread Observed Catskills Precipitation (170 mm)

Three hypotheses to explain variability: 1. Wetter members have stronger upslope forcing over the Catskills than drier members.

Three hypotheses to explain variability: 1. Wetter members have stronger upslope forcing over the Catskills than drier members. 2. Wetter members have greater moisture convergence over the Catskills than drier members.

Three hypotheses to explain variability: 1. Wetter members have stronger upslope forcing over the Catskills than drier members. 2. Wetter members have greater moisture convergence over the Catskills than drier members. 3. Wetter members have stronger Q-vector convergence over the Catskills than drier members.

Objective Clustering of Ensemble Members • Used k-means algorithm to sort the ensemble members into three clusters, based on the 39 -h horizontal distribution of precipitation over the domain 41. 5– 43. 5 N, 73– 76. 5 W. • 39 -h was the interval of maximum precipitation rates over the Catskills. • Three clusters is enough to accurately portray the variability between members.

Objective Clustering of Ensemble Members Cluster 3: Western Cluster 2: Eastern Cluster 1: Low Rainfall

Hypothesis 1: Upslope Wind Angle Wetter, Western Cluster Forecast Hour: 39 900 -h. Pa Winds 3 -hourly Precipitation Drier, Eastern Cluster

Hypothesis 2: Moisture Convergence Wetter, Western Cluster Forecast Hour: 39 1000– 700 -h. Pa Mean Winds Integrated Moisture Trans. Drier, Eastern Cluster

Hypothesis 3: Q-Vector Convergence Wetter, Western Cluster Drier, Eastern Cluster Forecast Hour: 39 700 -h. Pa Isotherms 700 -h. Pa Q-Vector Conv.

Wetter, Western Cluster Drier, Eastern Cluster V 10 • ▽Zs

Wetter, Western Cluster Drier, Eastern Cluster Rain initialed by upslope, but primarily forced by Q-vector conv. V 10 • ▽Zs

Wetter, Western Cluster Drier, Eastern Cluster There is strong upslope present initially, but very little moisture available for it to produce rain. V 10 • ▽Zs

Wetter, Western Cluster Drier, Eastern Cluster Although initial upslope is similar to the eastern cluster, it coincides with increased moisture, causing much greater rainfall. V 10 • ▽Zs

Wetter, Western Cluster Drier, Eastern Cluster Q-vector conv. replaces upslope forcing, maintaining high rain rates. V 10 • ▽Zs

GFS Conclusions • Precipitation differences in the GFS ensemble are largely due to differences in Irene’s position. • Analysis errors associated with a PV anomaly to the SW of Irene appear to be related to track differences. • Initial steering flow anomalies set up a feedback loop with an upstream trough, causing the pattern to amplify.

WRF Conclusions • When downscaled to 3 km with WRF, precipitation variability in the 80 member ensemble is driven by factors other than storm track (although storm track does play a role). • The WRF ensemble mean does a very good job forecasting precipitation amounts over the Catskills, due to its superior terrain resolution and ability to simulate mesoscale processes. • Precipitation in the wetter WRF members appears to be driven by a combination of terrain effects, synoptic forcing, and high available moisture, while precipitation in the drier WRF members appears to be driven primarily by synoptic forcing.

General Conclusions • A small east-west deviation in a member’s storm track can have a huge effect on the amount of rain received by a particular location. • Clustering ensemble members into specific forecast scenarios can reveal more information than just using the ensemble mean and standard deviation.

Questions?