New England Deluge of May 2006 Ensemble Forecast Evaluation
Shared by: MHairston
New England Deluge of May 2006: Ensemble Forecast Evaluation By Richard H. Grumm National Weather Service Office State College, PA 16803 and Walt Drag National Weather Service Office Taunton, MA 02780 1. INTRODUCTION New England exceeded 5 inches over large portions of three aforementioned Heavy rains impacted portions of New States. Several localized areas of 10-15 England on 13-15 May 2006. The inches of rain were observed. The storm persistent rain produced locally heavy delivered heavy rain, intermittently, over rain and flooding over portions of a period that spanned about 100 hours. Massachusetts, New Hampshire, and The initial heavy rains affected the southeast Maine. As shown in Figure 1, Connecticut River valley with flooding the 5-day total rainfall in east-central on 12 May. Areas of Connecticut Figure 1 Spotter report rainfall collected by the National Weather service office in Taunton, Massachusetts. Contour interval is 2 inches. received 5 to 8 inches of rainfall with ensemble forecast system (SREF) did lesser amounts outside the flood area. indicate a significant rain event. Overall, The rain continued over the 13 and 14th in this event, recognition of a long adding to the overall rainfall duration event and the pattern conducive accumulation. Heavy rains were for heavy rains may have been as observed in portions of Massachusetts important as ensemble prediction system and New Hampshire over this two day (EPS) precipitation forecasts. period. A final slug of 1.5 inches of rain brought more flooding late on the 15th This type of event, though rare is not and early on the 16th in northeastern uncommon, a heavy rain event affect Massachusetts. portions of New England on 7-8 October 2005. This event brought heavy rains The National Centers for Environmental and flooding to portions of New Predictions (NCEP) short-term ensemble Hampshire (Grumm 2005) key findings forecast system (SREF) will be used to associated with this event included: evaluate forecasts of this event. It is difficult to conceive a model predicting a) A strong, slow moving frontal zone, typically oriented in a north-south the observed rainfall amounts. However, configuration. it will be shown that the short-range Figure 3 As in Figure 1 except 00-hour forecast valid at 2100 Figure 2 SREF21 forecasts of 500 hPa heights initialized on UTC 13 May 2006. 2100 UTC 11 May 2006 valid at 1200 UTC 13 May 2006 showing a) spaghetti plots of 5460 and 5760 m contours by member and the ensemble mean (black) with the spread about the mean and b) the ensemble mean its departure from normal in standard deviations from normal. The May 2006 event was not directly b) Heavy rainfall occurring in the warm associated with a tropical system. moist air along the axis of the strong low-level jet. However, satellite imagery in this event and moisture forecasts in the SREF c) A large anticyclone was present over imply a tropical moisture connection eastern Canada. during the event. A similar connection with tropical moisture was observed in d) The event is observed with precipitable water anomalies which are the southeast Maine heavy rain event of well above normal. These highly 20-22 October 1996 (Konrad 2001). In anomalous values were due to the fact, the 1996 event ingested tropical interaction with a frontal system with the moistures associated with hurricane remnants of a tropical storm. Lilli. A key feature, not noted as e) This was a common transition season important in the October 2005, event heavy rain event type with the added was an anomalous blocking ridge and punch of tropical moisture. strong low-level anticyclone to the north and east. The ingredients found to be important to forecasting heavy rain Figure 4 As in Figure 2 except MSLP (hPa) forecasts valid at Figure 5 As in Figure 3 except precipitable water (mm). 2100 UTC 13 May 2006. (Doswell et al 1996) were present with All MREF and SREF data were archived this event including abundant, moisture, in real-time along with all the NAM and lift, and a boundary. GFS model data. All images were recreated after the event using GrADS This paper will document the heavy rain (Doty and Kinter 1992). Standard event of 14-16 May 2006. The focus is displays of “spaghetti” and “spread” are on the performance on the NCEP SREF shown. The spread is normally shown in predicting the overall pattern with gray shading. The spaghetti plots associated with the event; the ingredients show distinct contours for each conducive to producing heavy rains; and parameter (Sivillo et al 1997). Lower the quantitative precipitation forecast panels for some fields show the produced by the EPS. This document ensemble mean with color shading will not provide details about the showing the ensemble mean in standard catastrophic flooding, including the deviations from normal, referred to as historic floods along the Merrimac standardized anomalies (Grumm and River. Hart 2001;Hart and Grumm 2001). Precipitation displays show the critical 2. METHODS thresholds over a time period, normally 24 hours and 0.50 inches of quantitative precipitation forecast (QPF). Upper Figure 6. SREF forecasts initialized at 0900 UTC 10 May 2006 showing a) precipitable water anomalies and b) 850 hPa winds. Precipitable water shows spaghetti plots and dispersion about the mean and the ensemble mean with the departures from normal. Winds show wind speed (kts) and the U and V wind anomalies in standard deviations from normal. panels normally show the probabilities Figures 2 & 3 show forecasts and and the consensus of the specified “ensemble verification” on 13 May thresholds. Lower panels show the 2006. Key features with this event position of the specified threshold for included an anomalous high latitude each EPS member and shading shows ridge over eastern North America and an the ensemble mean QPF. anomalous upper-level trough over the east-central United States. The forecasts The spaghetti plots follow the standard of positive height anomalies from ensemble concepts outlined by Sivillio et 11/2100 UTC under forecast the al (1997). This paper is a good reference intensity of the ridge over eastern point for those learning to use Canada. The anomalies were on the ensembles. order of 2-3 standard deviations (SDs) above normal but were observed to be Forecast data shown here will primarily on the order of 3-5SDs above normal. from the NCEP SREF. All dates and The trough and the anomalies in the times will be in the format of day and trough were closer to the observed hours. Such that 0900 UTC 14 May values. 2006 would be displayed at 14/0900. The blocking pattern, with the high- 3. RESULTS latitude ridge and the trough formed the western end of an omega block over the i. Overview of the salient features western Atlantic and eastern North America. This block produced a strong surface anticyclone (Fig. 4) which, as will be shown, supported a persistent frontal boundary which was the focus of a multi-day rain event over New England. This boundary had normal to slightly above normal precipitable water (PWAT: Fig. 5) which persisted through out the event. ii. SREF Forecasts 10 May SREF forecasts of precipitation and the features associated with the event will be the focus of this section. It would be exhaustive to show all forecasts so a limited suite of data is presented. Figure 6 shows the SREF forecasts from 10/0900 UTC valid at 13/0900 UTC of Figure 7. SREF forecasts initialized at 0900 UTC PWAT and 850 hPa winds. The winds 10 May 2006 showing accumulated precipitation show the U (east-west) and V (north- for the 36-hour period ending at 2100 UTC 13 May 2006. Upper panels shows the probability of south) wind anomalies respectively. In exceeding 1 inch and the lower panels shows the addition to the slightly above normal ensemble mean (shaded) and each members 1.0 PWAT over south-central New England, inch contour. an anomalous southeasterly 850 hPa jet spaghetti plots showed a large spatial was forecast over the region. As will be variation in the area of heaviest rain shown, this anomalous jet verified and which led to the lower mean persisted for an extended period of precipitation forecast and the relatively time. low probability (60% or less) of exceedance. The precipitation forecasts for the period extending from 12/0900 UTC through Forecasts from 10/2100 UTC were 13/2100 UTC is shown in Figure 7. similar and for brevity they are not These data show only 50-60% of the shown. members forecasting 1 inch or more of rainfall over central New England along the boundary. The highest probability of iii. SREF forecasts from 11 May 1 inch or more was over portions of Connecticut and Massachusetts. However, the 1 inch contour in the Figure 8 As in Figure 6 except SREF forecasts initialized at 2100 UTC 11 May 2006. Figure 8 shows the SREF PWAT and 850 hPa wind forecasts initialized at 11/2100 UTC valid at 13/0900 UTC. These data show the same concept from earlier forecasts, with a persistent boundary over south-central New England. Along this boundary, above normal PWAT was present along with above normal southeasterly flow. Most forecast periods covering the period of rain showed this persistent feature in the moisture fields as well as the persistent strong southeasterly flow. The 11/2100 UTC precipitation forecasts are shown in Figure 9. These data show a 30-50% chance of 1 inch or more QPF in the Connecticut Valley for the 24 hour period ending at 14/0000 UTC. The lower panel shows most SREF members focused the 1.00 inch contour over New England. Figure 9. As in Figure 6 except for SREF initialized iv. SREF forecasts from 12 May at 2100 UTC 11 May and forecast of 24-hour accumulations for the period ending at 0000 UTC 14 May 2006. Figure 10 SREF 850 hPa wind forecasts initialized at Figure 11. As in Figure 9 except valid at 0000 UTC 14 2100 UTC 12 May 2006 valid at 0900 UTC 13 May May 2006. 2006. Upper panel shows winds and U-wind anomalies and lower panel shows 850 hPa winds and the V-wind anomalies. SREF forecasts from 12 May 2006 continued to show the same general conditions as those produce in earlier forecasts. This included the persistent boundary in the PWAT field over south- central New England and the anomalous low-level jet. This strong low-level jet is shown in Figures 10 and 11. In these forecasts, some areas of negative U-wind anomalies in excess of -3SDs below normal were present at 13/0900 UTC and a broader area of -3SD easterly wind anomalies were present at 14/0000 UTC. Though not shown, this area of intense southeasterly flow maintained itself and Figure 12. As in Figure 8 except SREF forecasts intensified from 13/0900 UTC through initialized at 2100 UTC 12 May showing 1 inch 14/0000 UTC. The SREF’s forecast a or more in 24-hours for the 24 hour period persistent anomalous flow over the ending at 0300 UTC 14 May. boundary. These conditions may have favored training over and along the cool side of the boundary. Figure 13. SREF accumulated precipitation for Boston, MA from SREF forecasts initialized at 0900 UTC 10 May 2006. Gray lines show the instantaneous 3-hour amounts and the colored lines show the accumulated amounts. Sums by type are provided on the left hand side. Figure 14. As in Figure 12 except SREF forecasts initialized at 0900 UTC 11 May 2006. sufficient precipitation relative to what The SREF precipitation forecasts was observed. continued to show low precipitation amounts relative to what was observed v. SREF point forecasts (Fig. 12). To its credit, the SREF had the correct area but was unable to generate Figure 15. As in Figure 11 except initialized at 21Z 11 May 2006. Figure 16. As in Figure 11 except initialized at 0900 UTC 12 May 2006. A series of SREF point forecasts are Figure 13 shows the QPF from the SREF presented for the environs about Boston, for forecasts initialized at 10/0900 UTC. Massachusetts. These data put the These data show a prolonged period of forecast quantitative precipitation in a precipitation. The interval of the context not easily seen in plan view maximum accumulations varies widely maps. from member-to-member. The mean QPF is 1.51 inches and at least 1 member forecast over 3.60 inches of QPF. Nearly 33% of the members were Heavy rains impacted portions of New clustered with light amounts generally England on 13-15 May 2006. The less than 1 inch. persistent rain produced locally heavy rain and flooding over portions of Figure 14 shows the SREF forecasts Connecticut, Massachusetts, New initialized at 11/0900 UTC. Overall, Hampshire, and southeast Maine. Heavy there was an increase in the QPF rainfall, most of it falling over a 100 forecasts. The maximum was 5.40 hour period, caused widespread flooding inches and the mean was 3.14 inches. across much of eastern Massachusetts There still were some dry members and and southern New Hampshire of small the timing of when the accumulations streams and main stem rivers. Major would occur varied. This would impact flooding of a number of small streams the appearance of plan view displays. and main stem rivers was observed in portions of Massachusetts and New Figure 14 shows the SREF plume from Hampshire. States of Emergency were forecasts initialized at 11/2100 UTC. declared for portions of Massachusetts The mean QPF is 2.76 inches with a and New Hampshire. Several locations maximum of 5.15 inches. The driest observed the worse flooding since the member still had 0.83 inches of QPF. 1938 Hurricane and the great rain and Similar to previous forecasts, the snowmelt floods in March 1936. This forecasts implied a long duration event. storm produced rainfall amounts similar But the time of accumulation varied to those observed in the 20-21 October from member-to-member. The varying 1996 event (Storm Series Report #96-7). interval of when the QPF would The flooding was comparable to and in accumulate made making probability some locations worse than that observed charts difficult. in the October 1996 event (NWS Figure 15 shows SREF forecasts Taunton). initialized at 12/0900 UTC. The overall maximum precipitation is lower; the As shown in Figure 1, the 7-day total mean was higher than previous SREF rainfall in east-central New England cycles. Similar to previous cycles, there exceeded 5 inches over large portions of was considerable variation in the time of three aforementioned States. The overall when the largest accumulations would pattern conducive to producing heavy occur. In this example, several members rains was well forecast; however, model show rapid accumulations on the 12th and ensemble QPF’s were too low for and early on the 13th while two members this event. show rapid accumulations after 14/0000 UTC. In this time period, there was a The relatively poor SREF QPF suggests clustering of members between 3.8 and there are still some limitations in using 4.6 inches of QPF. Note that 7 members ensemble prediction data to get the QPF arrived at such high numbers but the got “correct”. Though no single model there with the QPF falling at widely appeared to offer a better solution and varied times. some SREF members did forecast 2-3 inches of rainfall. All deterministic 4. CONCLUSIONS members should be considered part of an ensemble whenever practical to capture For this event, SREF precipitation the local and regional maximum. In this forecasts appeared to under forecast the case, no deterministic solutions were rainfall; when viewed in plan view over presented. Though the forecasts discrete time intervals; with this event contained the SREF RSM, NAM, and and thus under estimated the potential the two WRF set of members. for flooding. However, the features were well forecast. The persistent moisture Though the QPF’s were not particularly boundary with above normal PWAT accurate 1 , the SREF’s produced and the persistent strong southeasterly excellent forecasts of a pattern flow were well forecast by the SREF. conducive to heavy rain, and they Thus, the models captured a potential correctly forecast this pattern to persist heavy rain signal but were generally for a prolonged period of time. The unable to capture the timing and boundary outlined by the SREF PWAT intensity of the rainfall in the contours and the anomalies associated precipitation forecasts. with these forecasts proved quite accurate. This boundary played a critical The plume diagrams, shown only for role in the areas receiving heavy Boston, revealed that the individual precipitation. The ability to forecast this members of the SREF system did persistent pattern was a SREF strength forecasts a significant and long and provided useful guidance to duration rainfall event. Given the forecasters familiar with heavy rainfall resolution of the members, the fact that patterns. Admittedly, the PWAT several members forecast 4-5 inches of anomalies, though present were not QPF is quite remarkable. Clearly, timing significantly above normal. They issues led to the uncertainty, but taken remained about 1 SD above normal over the long course of this event, the throughout the event and in the QPF forecasts and the probability of forecasts. significant rainfall was quite well depicted by the plumes. This signal was Persistent southeasterly flow was a harder to see in the plan view maps. continual problem and threat. Due to Heavy rainfall, most of it falling over a southeasterly flow, there were both 100 hour period, caused widespread anomalous southerly and easterly winds. flooding across much of eastern Anomalies were on the order of 2-3SDs Massachusetts and southern New for several days. The strong deep Hampshire of small streams and main southeasterly flow is another signal stem rivers. Major flooding of a number associated with heavy rains in the of small streams and main stem rivers eastern United States. This feature and was observed in portions of significant positive PWAT anomalies Massachusetts and New Hampshire. were associated with heavy rains of States of Emergency were declared October 2005. portions of Massachusetts and New Hampshire. Several locations observed the worse flooding recorded since the 1 1938 Hurricane and the great rain and It is unrealistic to expect a 32km based forecast suite to predict rainfall in the 6-20 inch ranges. snowmelt floods in March 1936. This Forecasts of 3-5 inches are in and of themselves storm produced rainfall amounts similar quite respectable. to those observed in the 20-21 October amounts of moisture are 1996 event (Storm Series Report #96-7). present and strong The flooding was comparable to and in southeasterly flow is present, some locations worse than that 1996 the features and the anomalies event (NWS Taunton). may be signals to help anticipate a heavy precipitation The plume diagrams clearly pointed out event. several significant problems when c. When properly viewed, the displaying forecast data, most can be SREF system appears capable attributed to uncertainty in weather of forecasting large heavy rain forecasting in general; but the salient events. points are: a) the timing of when the 5. ACKNOWLEDGEMENTS accumulations would occur varied, NECP for continued access to the SREF b) this impacted the appearance of data. plan view displays over discrete intervals 6. REFERENCES c) creating a false impression of overall SREF skill. Doswell, C. A. III, H. E. Brooks, and R. A. Maddox, 1996: Flash flood Clearly, tools are needed to allow forecasting: An ingredients-based forecasters to adjust the time range and methodology. Wea. Forecasting, 11, critical threshold values to be evaluated. 560-581. In this case, using discrete times in the plan view maps masked the actual skill Doty, B. and J.L. Kinter III, 1992: The the SREF displayed in forecasting heavy Grid Analysis and Display System rainfall. It is prohibitive to make static (GrADS): A practical tool for research images in near real-time to fill this niche. science visualization. International In this case, the plumes revealed the real Conference on Interactive Information skill in the SREF and the overall and Procession Systems, Atlanta, advantages of an ensemble prediction Georgia, 5-10 January, 1992. system. The SREF clearly forecast the potential for heavy rains and the SREF Grumm, R.H., and RE Hart. 2001: clearly showed the long duration event Standardized Anomalies Applied to and the uncertainty about the time the Significant Cold Season Weather large accumulations would occur. Events: Preliminary Findings. Weather and Forecasting,16, 736–754. This case demonstrates that forecasters still need to apply conceptual models to Grumm, R. H., and Hart R., 2001: heavy rain events and clearly, Anticipating heavy rainfall events: a. There are still events were the Forecast aspects. Preprints, Symp. on human can add value to model Precipitation Extremes: Prediction, forecasts. Impacts, and Responses, Albuquerque, b. Whenever a persistent NM, Amer. Meteor. Soc., 66–70. boundary with copious Hart, R.E., and R H. Grumm. 2001: Using Normalized Climatological Anomalies to Rank Synoptic-Scale Events Objectively. Monthly Weather Review,129,2426–2442. Hart, R., and Grumm R. H., 2001a: Anticipating heavy rainfall events: Climatological aspects. Preprints, Symp. on Precipitation Extremes: Prediction, Impacts, and Responses, Albuquerque, NM, Amer. Meteor. Soc., 271–274. Hoffman, R.N., and E.Kalnay, 1983: Lagged average forecasting, an alternative to Monte Carolo forecasting. Tellus,35A,100-118. Konrad, C.E. II. 2001: The Most Extreme Precipitation Events over the Eastern United States from 1950 to 1996: Considerations of Scale. Jour. Hydrometeorolog,2,309–325. Maddox, R. A., C.F. Chappell and L. R. Hoxit, 1979: Synoptic and Meso-alpha Scale Aspects of Flash Flood Events. Bull. of Amer. Meteor. Soc., 60, pp 115- 123. Sivillo J.K, JE. Ahlquist and Z Toth. 1997: An Ensemble Forecasting Primer. Weather and Forecasting,12, 80 9– 818. Figure A1. NWS precipitation analysis old Figure 1.