The Mid-Atlantic Upper Low and Heavy Rainfall of 13-15 August 2011

by Richard H. Grumm

National Weather Service State College, PA

Abstract:

An unseasonably deep 500 hPa low moved over the eastern United States from 13-15 August

2011. This system triggered showers and thunderstorms which produced heavy rainfall along the East

Coast. As it moved across the Midwest it triggered convection which led to the deadly tragedy at the

Indiana State Fair.

The deep low and the surge of moisture into the Mid-Atlantic region produced some locally

heavy rainfall. Locally heavy rain from several bands of showers produced some impressive 24 hour

rainfall totals. The highest totals were 10.64 and 9.20 inches in Seabrook and Bridgeton, New Jersey.

This paper documents the event and shows some of the limitations of numerical models and

ensembles in predicting warm season heavy rainfall events. The results shown herein imply some

issues related to both initial conditions and internal model physics which caused the NCEP NAM,

NCEP GF, and EC model to pull the plume of moisture too far west of the observed location and thus

incorrectly put the initial heavy rain threat too far west of where heavy rain was observed.

1. INTRODUCTION

A deep 500 hPa trough moved across the Midwest and over the Mid-Atlantic region on 13-15

August 2011 (Fig. 1). This deep upper-level system triggered convection and produced severe weather as

it moved eastward on 13 August (Fig. 2). The severe weather in the Midwest included a squall line which

produced strong winds. Winds over 70 mph toppled the stage at the Indian State Fair killing 5 people and

injuring 45 people (SPC and AP2011). As the system moved over the Mid-Atlantic region it produced

heavy rainfall (Fig. 3).

The pattern for heavy rainfall was relatively well predicted by both the National Centers for

Environmental Predictions (NCEP) deterministic models and ensemble forecast systems. It will be shown

despite the generally well predicted pattern favoring heavy rainfall; these forecast systems had difficulty

placing where the heavy rain would fall and the areas which would receive the heavy rainfall. The

convective nature of the rainfall likely related to these quantitative precipitation forecast issues. As noted

by Fritsch and Carbone (2004), convectively driven warm season rainfall continues to be a challenging

and difficult forecast problem for our forecast systems. This event shows that little progress with

operational models has been made to correct this situation over the past 7 years. The best approach to this

problem is likely very high resolution models and ensembles of these models run at or near the scale of

larger convective systems. Barthold et al. (2011) has shown that ensembles of high resolution models,

that is, models run at 3-5km in grid spacing, offer the best hope to improve the forecasting of these

convectively driven events. Clearly, relying on coarser models and ensemble forecast systems with

convective parameterization schemes is of limited value in events of this type.

This paper will document the heavy rainfall event 13-15 August 2011. The focus is on the pattern

and how it related to the heavy rainfall. Short-term forecasts from NCEP models and ensembles are

presented focusing on the limitations of these systems in warm season heavy rainfall events involving

convection.

2. METHODS AND DATA

The 500 hPa heights, 850 hPa temperatures and winds, and other standard level fields were

derived from the NCEP Global Forecast System (GFS), North American Mesoscale Model (NAM),

Global Ensemble Forecast System (GEFS), Short Range Ensemble Forecast (SREF) and the

NCEP/NCAR (Kalnay et al. 1996) reanalysis data. The means and standard deviations used to compute

the standardized anomalies were from the NCEP/NCAR data as described by Hart and Grumm (2001).

Anomalies were displayed in standard deviations from normal, as standardized anomalies are used to

show the anomalies, the probability of heavy rainfall and the Quantitative Precipitation Forecasts (QPF)

associated with this event. All data were displayed using GrADS (Doty and Kinter 1995).

The standardized anomalies computed as:

SD = (F – M)/σ (1)

Where F is the value from the reanalysis data at each grid point, M is the mean for the specified

date and time at each grid point and σ is the value of 1 standard deviation at each grid point.

Model and ensemble data shown here were primarily limited to the GFS and GEFS. The NAM

and SREF data were also available for use in this study. Displays will focus on the observed pattern and

some forecast issues associated with the pattern.

Both the unified precipitation dataset (UPD: Seo 1998) and the stage-IV precipitation data (Lin

and Mitchell 2005; Nelson et. al. 2010) were used to evaluate the rainfall associated with this event. The

stage-IV data are presented in this document. The first estimates of the rainfall were derived from the

National Mosaic and Multi-sensor Quantitative precipitation estimate site.

Point data from rain gages was obtained in real-time from National Weather Service Cooperative

observers and spotters.

In addition to computing standardized anomalies for distinct times associated during the event,

multi-period standardized anomaly values over 0.5 to 2 day periods were computed. This facilitated

identifying the signal and the persistence in the long-term signal of the pattern. This concept was then

used to obtain multi-day pattern forecasts from the NCEP GFS and GEFS.

Verification was computed for the 24-hour period where the heaviest rainfall fell along the East

Coast (1200 UTC 14 through 1200 UTC 15 August 2011). The Stage-IV data was used as ground truth.

For each system, the NAM, GFS, and SREF, the mean QPF was computed and the Stage-IV data was

remapped to each systems grid. The QPF (mm) for each model cycles is displayed. Beneath each forecast

the model QPF and the remapped Stage-IV data was shown as forecast minus observed. Positive

(negative) values indicated over (under) forecasting of QPF. As will be shown, the regions of heaviest

rainfall routinely had under forecast errors of over 50 mm (2 inches).

For brevity, times will be displayed in day and hour format such a 14/0000 UTC signifies 14

August 2011 at 0000 UTC.

3. METEOROLOGICAL OVERVIEW

i. Synoptic scale pattern

The 500 hPa pattern over the eastern United States (Fig. 1) shows the relatively deep 500 hPa low

which moved over the eastern United States from 13-15 August 2011. As this system moved eastward a

surge of deep moisture (Fig. 4) and above normal precipitable water (PW) moved up the East Coast of the

United States. Though not shown there was a surge of strong southerly winds associated with this plume

of deep moisture and modest instability.

The persistence of the pattern likely contributed to the heavy rainfall over the Mid-Atlantic

region. Over the 48 hour period from 13/1200 UTC through 15/1200 UTC (Fig. 5) a deep trough was

present over the Ohio Valley (Fig. 5a) with a surge of strong winds at 850 hPa (Fig. 5b) and a surge of

deep moisture up the coastal plain (Fig 5c). During the 24-hour period of heavy rainfall over New Jersey

and New York, the pattern over Mid-Atlantic region (Fig. 6a) showed a deep trough over Virginia. The

plume of high PW air was focused along the coastal plain and into southern New England (Fig. 6c) with

strong southerly flow along the coastal plain (Fig. 6b). A deep surface cyclone was present in the mean

along the coastal plain (Fig. 6d).

The 250 hPa winds and 850 hPa u-winds during the peak of the rainfall (Fig. 7) showed a strong

jet at 250 hPa and easterly winds north of the surface cyclone. The 850 hPa temperatures (Fig. 7c)

indicated a surge of below normal temperatures at 850 hPa and implied a cold front from the Gulf States

into the Mid-Atlantic region.

ii. Precipitation

The 48-hour accumulated rainfall from the NMQ data (Fig.3) showed the focus of the heavy

rainfall over coastal regions from Virginia into New England. Farther inland these data implied a cyclone

nature to the bands of rain west of the surge of strong southerly flow and the surge of high PW air (see

Figs. 5-7). Similar to the NMQ data, the Stage-IV data (Fig. 8) showed the focus of heavy rainfall from

Virginia northeastward over Long Island. The Stage-IV data showed the heaviest rainfall for the period

over Long Island and showed only 75-100 mm over southern New Jersey. The region of southern New

Jersey is where the 9.20 and 10.64 inch values from gauge points were observed.

The Stage-IV data clearly shows the same overall pattern as the NMQ data. However, there are

differences as to where the heaviest rainfall was estimated by these two datasets. The reality of rainfall

and what rain fell where is not an exact science. Twenty 26 individual locations received over 5.50

inches of rainfall (Table 1) and over 68 locations received 4 or more inches of rainfall (not shown).

Seabrook, NJ received 10.64 inches on the 14th and came in with 11.43 inches for the duration of the

event.

iii. Forecasts

The overall pattern in the NCEP models showing a plume of deep moisture and the strong low-

level jet up and along the East Coast was relatively well predicted. Clearly, the devil was in the details

and in this case, the details were the quantitative precipitation forecasts (QPF) as shown in Figures 9-11.

All 3 systems, the GFS, NAM, and SREF failed to predict the focused area of heavy rainfall along the

East Coast from northern Maryland to southern New England. Initially, the NAM had the axis of heavy

QPF located over 100 km too far west (Fig. 9a) and grossly under-predicted the heavy rainfall potential

along the East Coast with over 50 mm errors extending from Maryland into Connecticut (Fig. 9d). The

GFS (Fig. 8a) and SREF (Fig. 10a) though to the west of the observed precipitation maximum, were well

east of the NAM.

The NAM shifted the axis of heavy QPF to the east by 13/1200 UTC (Fig. 9b) and thus reduced

some its errors along the coastal plain form New Jersey to Maryland, with little improvement over Long

Island and southern New England (Fig. 9e). The 13/1200 UTC GFS shifted the heavy rainfall axis too far

east but produced sufficient QPF along the coastal plain to reduce its overall error. The 14/0000 UTC

GFS (Fig. 8c) was wetter still and pulled the axis of heavy rainfall to the west1. The NAM axis of heavy

rainfall was just west of that produced by the GFS but had lower amounts (Fig. 9c).

The 21-member SREF had similar issues displayed by the NAM and GFS. Due to averaging, it

produced lower amounts, though individual members produced over 2 inches of QPF over the Mid-

Atlantic region during this 24 hour period (Fig. 11) and over 4 inches in the 48 hour period ending at

15/1200 UTC. The SREF suffered less with the westward and later eastward bias relative to the NAM and

the GFS. Its strength may have been better depiction of the high threat area relative to the more erratic

GFS and NAM.

1 Only visible if you graphically subtract the forecasts from 13/1200 UTC and 14/0000 UTC.

The European Center model (ECM: Fig. 12) had difficulty with the QPF from forecasts issued at

12/1200 UTC (Fig. 12a & 12e). The ECM forecasts and errors in both the PW and MSLP fields (Fig. 14)

suggest that the ECM developed too deep a cyclone which it tracked into western New York (Fig. 14)

where -4 to -16 hPa pressure errors were evident along with significant PW anomalies (Fig. 14e).

Forecasts from the 13th and 14

th lacked this cyclone and the large errors implying why later forecasts had

the rainfall close to the coast.

Comparable NCEP GFS and NAM verification (Figs. 15 & 16) show large PW errors in the

12/1200 UTC NAM (Fig. 16d) over Pennsylvania implying similar to the 12/1200 UTC ECM, the

forecast PW ridge was west of the verifying position. The GFS had a similar negative pressure error over

western New York (Fig. 15a) but lacked the large PW error over Pennsylvania and New York relative to

the other two models (Fig. 15d). The errors decreased markedly as the forecast length decreased.

4. CONCLUSIONS

A slow moving upper level low (Fig. 1) and a surge of moisture up the East Coast brought

localized heavy rainfall to portions of the Mid-Atlantic region and southern New England (Figs. 2 & 7).

The heaviest rainfall was over portions of New Jersey and Long Island. The heaviest rainfall was

observed over southern New Jersey where over 11 inches fell in Seabrook, New Jersey.

The overall pattern conducive for heavy rainfall was relatively well predicted by the NCEP

models and ensemble forecast systems (not shown). The 00-hour forecasts from these models suggested a

plume of deep moisture up the coastal plain, in close proximity to where the heavy rains were observed,

and strong low-level jet (Figs. 4,5, & 7). Despite capturing the pattern, the heavy rainfall and the area

where the heaviest rains would fall were not well predicted by the NCEP NAM, GFS, or SREF. The

GEFS though not shown was similar in its errors to the GFS. Clearly, 3-days out the NAM forecast the

heavy rainfall too far west. The SREF and GFS also predicted the rainfall too far west, but not as far as

the NAM. Perhaps due to model and initial conditions variability, the SREF appeared to have the smallest

(qualitatively) placement errors though the ensemble mean was too low on the maximum rainfall. Most

SREF members also produced lower total QPF values relative to the GFS. This may be due to the higher

resolution of the GFS (~27km) verse the SREF (most members ~32km). Despite higher resolution

(~12km) the NAM produced less QPF in most forecast cycles relative to the GFS.

In this case, where convection clearly played a critical role in the evolution of the rainfall, the

NAM was clearly outperformed by both the GFS and SREF in terms of the location of heavy rainfall. The

12/1200 UTC EC had shortcoming similar to those found in the NAM. Shorter range NAM forecasts

outperformed the GFS with regard to the region where the heaviest rains would fall. The SREF appears to

have one advantage over both systems in that averaging and computing probabilities indicated the

highest threat area for heavy closer to the region where the heaviest rainfall was observed. No system

was perfect but at longer ranges, the GFS clearly outperformed the NAM in terms of location and even at

shorter ranges, it had higher QPF amounts.

The larger errors in the NCEP NAM and ECM from forecasts issued at 12/1200 UTC and the

errors in the GFS imply some issues with model initializations. However, the large pressure error in the

ECM may imply resolution and convective feedback as well as initialization errors played a role in the

relatively poor forecasts at 15/0000 UTC from the 12/1200 UTC forecast cycle. The NAM had slightly

finer resolution than the ECM likely suffered from both initialization errors and internal feedback. GFS

had smaller errors in the PW and MSLP fields but was on better by degrees.

The current NCEP SREF is of too course a resolution to produce the QPF amounts that the finer

resolution GFS and NAM can produce. Though not known, the internal model QPF climatology of SREF

members is likely lower than the internal model climatologies of either the GFS or NAM. Higher

resolution regional ensembles may someday aid in better predicting the regions susceptible to heavy

rainfall.

From a forecast perspective, the NAM likely caused forecasters to be concerned for heavy rainfall

well west of where more widespread heavy rainfall was observed. Additionally, it and the westward

biased SREF and GFS on the 12th may have created the perception that most of New Jersey and southern

New England would be too far east of the heavy rainfall axis. This can relate back to decisions related to

anticipating heavy rainfall and potential flooding. The uncertainty between the two deterministic models

and the low probabilities of high QPF in the SREF likely were valuable uncertainty and thus confidence

information. General outlook areas need to be large and refining them should be a slow and incremental

process due to the inherit uncertainty and low skill in model and thus ensemble QPFs.

This case, similar to the East Coast storm of 26-27 December 2010, demonstrate that sometimes

models have difficulty getting the pattern correct and thus misplace the area where the heaviest QPF will

fall. Confidence in models should increase as forecast length decreases, however, we have limited means t

judge when a models forecast is more or less uncertain. No test of a Poorman’s ensemble was attempted

here, but clearly some means to gage uncertainty is required if forecasters are to effectively communicate

the threat, in this case for heavy rainfall, in a meaningful manner to planners and emergency responders.

This continual inability to predict warm season convective rainfall (Fritsch and Carbone 2004)

was identified years ago. For most coarse resolution models to include the NCEP SREF, relying on

convective parameterizations to predict rainfall is as problem. It is unclear if higher resolution models and

ensemble will solve the QPF issue. Storm scale ensemble forecast systems (SSEF: Barthold 2011) hold

some hope in improving our ability to forecast and identify events such as this. But these systems will

likely be of limited value at forecast lengths longer than 18-30 hours and they will suffer from

initialization limitations as do the larger scale models.

The near-term future of NOWCASTING and using remote sensing tools to anticipate flash

flooding and locally heavy precipitation events looks good for operational meteorologists. The biggest

threat over the next 2-10 years may be less from NWP and SSEFs but from artificial intelligence

applications. Examining the HRRR and 4km NAM during this event, resolution offered few useful clues

much beyond 12-18 hours and these systems too are subject to initial condition and model uncertainty

issues.

5. Acknowledgements

.

6. REFERENCES

Associated Press, 2011: Indiana State Fair Death Toll of 5 Could Go higher after ‘Fluke’ Storm

Fells Stage. (Daily wire reports with similar titles were published from 13 to 15 August 2011.)

Faye E. Barthold, F.E, M. J. Bodner, D. R. Novak, R. Bann, R. Oravec, B. Sullivan, A. R. Dean,

I. L. Jirak, C. J. Melick, R. A. Sobash, A. J. Clark, F. Kong, S. J. Weiss, and M. Xue;

2011: The quantitative precipitation forecasting component of the 2010 NOAA

Hazardous Weather Testbed Spring Experiment. AMS, Seattle WA.

Doty, B. E., and J. L. Kinter III, 1995: Geophysical data and visualization using GrADS.

Visualization Techniques Space and Atmospheric Sciences, E. P. Szuszczewicz

and Bredekamp, Eds., NASA, 209–219.

Fritsch, J. M., and R. E. Carbone, 2004: Improving quantitative precipitation forecasts in

the warm season. Bull. Amer. Meteor. Soc., 85, 955-965

Graham, R.A and R.H. Grumm, 2010: Utilizing normalized anomalies to assess synoptic-

scale weather events in the western United States. Wea. Forecasting, 25,428-445.

Grumm, R.H., and R. Hart, 2001a: Anticipating Heavy Rainfall: Forecast Aspects. Preprints,

Symposium on Precipitation Extremes, Albuquerque, NM, Amer. Meteor. Soc., 66-70.

Grumm, R.H., and R. Holmes, 2007: Patterns of heavy rainfall in the mid-Atlantic. Pre-prints,

Conference on Weather Analysis and Forecasting, Park City, UT, Amer. Meteor. Soc.,

5A.2.

Grumm, Richard H. 2000, "Forecasting the Precipitation Associated with a Mid-Atlantic States

Cold Frontal Rainband", National Wea. Digest, 24, 37-51.

Hart, R. E., and R. H. Grumm, 2001: Using normalized climatological anomalies to rank synoptic

scale events objectively. Mon. Wea. Rev., 129, 2426–2442.

--------, M.J. Brennan, F. Pereira, M.J. Bodner, and R.H. Grumm, 2009: Assessing the Potential for

Rare Precipitation Events with Standardized Anomalies and Ensemble Guidance at the

Hydrometeorological Prediction Center. Bull. Amer. Meteor. Soc., 90, 445–453.

Stensrud, D.J., M. Xue, L.J. Wicker, K.E. Kelleher, M.P. Foster, J.T. Schaefer, R.S. Schneider, S.

G. Benjamin, S. Weygandt, J.T. Ferree and J.P. Tuell (2009), CONVECTIVE-SCALE

WARN-ON-FORECAST SYSTEM A Vision for 2020, Bulletin of the American

Meteorological Society, 90,1487, 10.1175/2009BAMS2795.1.

Figure 1. GFS 00-hour forecasts of 500 hPa heights (m) and standardized height anomalies (standard deviations) for the 12-hour period from

1200 UTC 13 August through 0000 UTC 16 August 2011. The red dot is near Seabrook, NJ where over 11 inches of rain fell. Return to text.

Figure 2. Storm reports by type from the Storm Prediction Center (SPC) for the period ending at 1200

UTC 14 August 2011. Return to text.

Figure 3. NMQ Q2 site adjusted rainfall estimates. Values as indicated in the color key. Maximum rainfall is 8.89 inches.

This does not show the 9 to 10 inches in southern New Jersey detected by gage data. Return to text.

Figure 4. As in Figure 1 except for GFS precipitable water (mm) and precipitable water anomalies for the period from a) 0000 UTC 14 April

through f) 0600 UTC 15 April 2011. Return to text.

Figure 5. GFS composite pattern for the period of 1200 UTC 13 through 1200 UTC 15 August 2011 showing a) 500 hPa heights (m)

and 500 hPa height anomalies, b) 850 hPa winds (ms-1) and 850 hPa v-wind anomalies, c) precipitable water (mm) and

precipitable water anomalies, and d) mean sea-level pressure and pressure anomalies. Return to text.

Figure 6. As in Figure 5 except for the 24 hour period of 1200 UTC 14-15 August 2011. Return to text.

Figure 7. As in Figure 6 except for a) 250 hPa winds and anomalies, b) 850 hPa winds and u-wind anomalies, c) 850 hPa

temperatures and anomalies, and d) precipitable water and anomalies. Return to text.

Town State Rainfall (in)

Seabrook NJ 11.43

Blackbird DE 8.88

Winslow Tw NJ 8.11

Middletown DE 7.80

Middlesex NJ 7.46

Newark AP NJ 6.74

LaGuardia NY 6.68

Caravel Fa DE 6.65

Belle Mead NJ 6.61

Harrison NJ 6.30

Philadelphia PA 6.11

Red Bank 1 NJ 6.00

Folsom GOE NJ 5.98

Elkton 3.8 MD 5.95

Blackwells NJ 5.90

Harrison 0 NJ 5.80

Philadelphia PA 5.80

N. Princeton NJ 5.80

Lake Hopatcong NJ 5.73

Freehold NJ 5.71

Brent Gap VA 5.66

Middlebush NJ 5.64

New Lisbon NJ 5.58

Lebanon 3. NJ 5.57

Newark 3.9 DE 5.56 Table 1. Listing of locations and State for rainfall in excess of 5.50 inches during the event. Data from MARFC. Return to text.

Figure 8. Stage-IV precipitation data for the 48 and 24 hour periods ending at 1200 UTC 15 August 2011.

Return to text.

Figure 9. NCEP GFS accumulated quantitative precipitation (mm) and forecast errors (mm) for the 24 hour period ending at 1200 UTC 15

August 2011 from forecast initialized at a & d) 1200 UTC 12 August 2011, b &d) 1200 UTC 14 August 2011 and c & e) 0000 UTC 14 August

2011. Upper panels show the model total precipitation and lower panels show the errors as described in the methods. Return to text.

Figure 10. As in Figure 9 except for NCEP NAM precipitation and precipitation verification. Return to text.

Figure 11. As in Figure 9 except for NCEP SREF mean QPF and verification from SREF initialized at a&d) 1500 UTC 12 August 2011, b&e)

0900 UTC 13 August 2011, and c & f) 0300 UTC 14 August 2011. Return to text.

Figure 12. NCEP SREF QPFs from SREF initialized at 0900 UTC 13 August 2011. Upper panels show the probability of exceeding (left)

2.00 inches in 24 hours and 2.00 inches in 36 hours. Lower panels show the SREF mean QPF (inches) and each member’s 2 inch contour

in (left) 24 hours and (right) 36 hours. Return to text.

Figure 13. As in Figure 10 except for European Center Model qpf (mm) and QPF errors. Return to text.

Figure 14. EC model forecasts (contours) and errors (shaded) valid at 0000 UTC 15 August 2011 showing MSLP forecasts and MSLP errors

from forecasts initialized at a) 1200 UTC 12 August 2011, b) 1200 UTC 13 August 2011 and c) 0000 UTC 14 August 2011 and lower panels

showing precipitable water (mm) from forecasts initialized at e) 1200 UTC 12 August 2011, f) 1200 UTC 13 August 2011 and g) 0000 UTC 14

August 2011. NCEP GFS 00-hour forecasts were used as verifying grid. Return to text.

Figure 15. As in Figure 14 except for NCEP GFS forecasts verified against the GFS 00-hour analysis. Return to text.

Figure 16. As in Figure 14 except for NCEP NAM forecasts verified against the GFS 00-hour analysis. Return to text.