DynamicalTropicalCyc.. - DYNAMICAL TROPICAL CYCLONE 96-H AND 120-H by hcj

VIEWS: 7 PAGES: 53

									  DYNAMICAL TROPICAL CYCLONE 96-H AND 120-H TRACK
   FORECAST ERRORS IN THE WESTERN NORTH PACIFIC




                                      Ryan M. Kehoe

                                      Mark A. Boothe

                                    Russell L. Elsberry



                    Graduate School of Engineering and Applied Sciences

                                Naval Postgraduate School

                                Monterey, California 93943




                           Submitted to Weather and Forecasting

                                       January 2006
                                     (Revised 7-24-06)
                                     (Figures 7-21-06)




Corresponding author: R. L. Elsberry, Department of Meteorology, MR/Es, 589 Dyer Rd.,
Room 254, Monterey, California 93943-5114. E- mail: elsberry@nps.edu
                                             Abstract

       The Joint Typhoon Warning Center has been issuing 96- h and 120-h track forecasts since

May 2003. It uses four dynamical models that provide guidance at these forecast intervals and

rely heavily on a consensus of these four models in producing the official forecast. Whereas

each of the models have skill, each occasionally has large errors. The objective of this study is to

provide a characterization of these errors in the western North Pacific during 2004 for two of the

four models: Navy Operational Global Atmospheric Prediction System (NOGAPS) and the

Navy version of the Geophysical Fluid Dynamics Laboratory model (GFDN). All 96- h and 120-

h track errors greater than 400 n mi and 500 n mi, respectively, are examined following the

approach of Carr and Elsberry (2000a, b). All of these large-error cases can be attributed to the

models not properly representing the physical processes known to control tropical cyclone

motion, which were classified in a series of conceptual models by Carr and Elsberry for either

tropical- related or midlatitude-related mechanisms. For those large-error cases that an error

mechanism could be established, midlatitude influences caused 83% (85%) of the NOGAPS

(GFDN) errors. The most common tropical influence is an excessive Direct Cyclone Interaction

in which the tropical cyclone track is erroneously affected by an adjacent cyclone. The most

common midlatitude-related errors in the NOGAPS tracks arise from an erroneous prediction of

the environmental flow dominated by a ridge in the midlatitudes. Errors in the GFDN tracks are

caused by both ridge-dominated and trough-dominated environmental flows in the midlatitudes.

Case studies illustrating the key error mechanisms are provided. An ability to confidently

identify these error mechanisms and thereby eliminate likely erroneous tracks from the

consensus would improve the accuracy of 96-h and 120-h track forecasts.




                                                 2
1.     Introduction

       Carr and Elsberry (2000a, b) examined all of the highly erroneous (> 300 n mi or 555 km

at 72 h) Navy Operational Global Atmospheric Prediction System (NOGAPS) and U. S. Navy

version of the Geophysical Fluid Dynamics Laboratory model (GFDN) tropical cyclone track

forecast errors in the western North Pacific during 1997. They described the responsible error

mechanisms in terms of conceptual models that are related to known tropical cyclone motion

processes that are being misrepresented in the dynamical models. The motivation for the Carr

and Elsberry study was to help the forecaster detect when the dynamical model guidance is likely

to be erroneous and thus should be rejected during preparation of the warning.

       A change in paradigm to consensus tropical cyclone track forecasting was occurring at

the same time. Goerss (2000) had developed a three-global- model or two-regional model

consensus technique at the synoptic and off-synoptic times, respectively. As part of the

Systematic Approach Forecasting Aid (SAFA; Carr et al. 2001), the above models and the Japan

Global Spectral Model (JGSM) and Typhoon Model (JTYM) and the UK Meteorological Office

(UKMO) global model were interpolated in time so that five tracks would be available each six

hours (Carr et al. 2001). Elsberry and Carr (2000) docume nted that a small spread (< 300 n mi)

among these five model 72- h tracks often implied a small consensus error, but that a large spread

did not necessarily imply a large consensus track error because the errors of two (or more) of the

models may be compensating. One of the objectives in SAFA is to examine the model guidance

and eliminate the model tracks that are likely to have large errors to form a selective consensus

of the remaining model tracks that should be more accurate than a non-selective consensus of all

five dynamical models.




                                                 3
       The Joint Typhoon Warning Center (JTWC), with the assistance of the Naval Research

Laboratory-Monterey, has expanded the consensus forecasting concept so that during the 2004

typhoon season its consensus forecast (CONW) included ten global and regional model tracks.

In addition to the five models mentioned above, the CONW also includes the National Centers

for Environmental Prediction Global Forecast System (GFS), the Australian Bureau of

Meteorology Tropical Cyclone Local Area Prediction System, the Weber Barotropic Model, Air

Force MM5 Model, and the Navy Coupled Ocean/Atmospheric Mesoscale Prediction System

(COAMPS). Each of these models has been improved over the years, and with the application of

the consensus forecasting paradigm, the JTWC has been able to markedly improve its 72-h

forecast accuracy (Jeffries and Fukada 2002).

       Based on these improvements and internal tests of 96- h and 120-h track forecasts during

the 2001 and 2002 seasons, the JTWC implemented official 96-h and 120- h forecasts beginning

in May 2003. Only four models provide guidance for these longer forecast intervals: NOGAPS,

GFS, GFDN, and UKMO. During the 2004 season, the first two models provided guidance each

6 h, but the GFDN (UKMO) tracks were only available for 0600 (0000) UTC and 1800 (1200)

UTC. Because the dynamical model forecasts are not available until 4-5 h after the synoptic

time, the track forecasts are interpolated from the prior 6-h integration (or 12-h for the GFDN

and UKMO models) to the position at the warning time. As demonstrated by Kehoe (2005, Fig.

2), the JTWC relies heavily on the consensus of these four models for its 120-h forecasts since

the correlation coefficient between the JTWC and consensus errors was 0.943 during 2004.

       The approach in this study follows that of Carr and Elsberry (2000a, b) with the objective

to determine the 96-h and 120- h track error characteristics and thus provide guidance to the

forecaster as to likely erroneous forecasts. Jeffries and Fukada (2002) had demonstrated that




                                                 4
more than three track forecasts were highly desirable for 72-h consensus forecasting to achieve

the canceling of random errors, which would suggest that all four of the 120- h track forecasts are

needed. A dilemma then exists when one of the four models has a highly erroneous 120-h track

that may cause the consensus to be seriously degraded. Nevertheless, the hypothesis is that

elimination of the erroneous model track and reduction of the consensus to only three model

tracks would still lead to a more accurate forecast. Since this is a retrospective study of cases in

which it is known that the dynamical model had a large error, it is a separate issue whether these

errors can be detected in real-time.

2.     Methodology

       The approach has been to analyze all cases in which large 120-h track errors occurred in

the NOGAPS and GFDN forecasts of western North Pacific tropical cyclones during 2004. Only

these two models are examined because a complete archive of analyses and forecast fields that

are necessary for error mechanism determination was not available for the GFS and UKMO

models.

       The definition of a ―large‖ 96-h and 120- h track error as being equal to 400 n mi and 500

n mi was based on the histogram of errors during 2003 and 2004 (Kehoe 2005, Figs. 3 and 4).

Whereas this definition is somewhat arbitrary, it is consistent with the approach of Carr and

Elsberry (2000a, b) in selecting a value that is twice as large as a reasonable goal for the 96-h

and 120-h track forecast accuracy. The error needs to be large enough that an error mechanism

can be confidently established, and the elimination of > 500 n mi errors at 120 h would greatly

improve the seasonal error statistics and the confidence of the customer in the warning system.

       To maximize the number of 120-h forecast verifications for both NOGAPS and GFDN,

the best-track positions were manually extended beyond the point that JTWC declares the




                                                  5
tropical cyclone to be extratropical. Continuing the positions using mean sea- level pressure

(MSLP) analyses is considered valid since the hazards associated with the wind, precipitatio n,

and waves accompanying an extratropical cyclone do not suddenly diminish at the time it is

declared extratropical. When the MSLP center was predicted by the GFDN model to have left

the model domain, the last location inside the domain was used as the predicted position for

calculating the errors. This is a conservative estimate because the actual error would be larger

than this calculated error. Consequently, the error summaries in this study will not match the

JTWC summaries and will in fact be larger. By following these procedures to maximize

verifications of 120- h forecasts, the number of cases with large errors increased nearly 28%

(24%) for NOGAPS (GFDN) (Table 1). One reason that the GFDN increase was lower than for

NOGAPS was because 26 of the 134 large-error cases in GFDN (described later in Table 2) did

not have forecast fields archived to 120 h that are necessary for extending the tracks.

       In the identification of these large errors, the GFDN had track errors at 0600 UTC and

1800 UTC for which no archived fields were available from the Navy Master Environmental

Laboratory. To still incorporate these forecasts in the summary, if the 0600 UTC (1800 UTC)

error was between 0000 UTC and 1200 UTC (1200 UTC and 0000 UTC) forecasts that also had

large errors, and both of these adjacent forecasts had the same error mechanism, then the 0600

UTC or 1800 UTC error was assigned the same error mechanism. If the 0600 UTC or 1800

UTC error had a large error only on one side, and the error values and the track forecasts of the

two times were similar, it too was assigned the same error mechanism. By applying this

procedure, the GFDN sample included 65 of 72 of the 0600 UTC or 1800 UTC large forecast

errors. A similar procedure for evaluating the off- time GFDN error mechanisms retained 65 of

the 72 large forecast errors at 0000 UTC and 1200 UTC. If a large 0600 UTC or 1800 UTC




                                                 6
error did not have a large error at either adjacent time, it was listed as ―no fields available‖ but it

was included in the error summary.

        In 2004, a total of 354 NOGAPS forecasts to 120 h were made for the western North

Pacific tropical cyclones. For GFDN in the same year, 262 forecasts to 120 h were available for

the same basin (Table 1). Of those forecasts, 162 (134) cases for NOGAPS (GFDN) had a large

forecast error at 96 h and/or 120 h (to be described in Table 2. To identify the causes of these

large errors, both the predicted and verifying analysis fields of the winds and geopotential

heights at 200, 500, 700, and 850 mb and the mean sea- level pressures were utilized. The

geopotential heights from 850 through 500 mb were found to be most beneficial in diagnosing

the cause of the large track forecast errors when midlatitude synoptic features were affecting the

steering current for the tropical cyclone. If vertical wind shear effects were suspected of causing

the error, the vector difference in winds between 850 mb and 500 mb, as well as the 200 mb

level, was vital. When the large size of a tropical cyclone was contributing to a beta-effect

propagation error, the mean sea- level pressure fields were found to be most effective in detecting

the cause of the error.

        As noted above, the dynamical model 72-h track guidance has been improved since the

1997 season that was studied by Carr and Elsberry (2000a, b). However, Kehoe (2005)

documented that an excellent (< 150 n mi), good (< 200 n mi), or fair (<300 n mi) forecast at

72 h does not guarantee even a fair (< 500 n mi) forecast at 120 h. Of the 162 NOGAPS large

96-h or 120-h errors during 2004, 13% occurred when the 72-h error was excellent, 25% when

the 72-h error was good, and 39% occurred when the 72-h error was fair. For the 134 GFDN

large 120-h errors, the corresponding percentages of excellent, good, and fair 72-h errors were

5%, 21%, and 45%. However, it will be shown in the next section that the error mechanisms




                                                   7
causing large 120-h errors are similar to those at 72 h, and that these mechanisms just come into

play later in the forecasts now with these improved models.

3.     Summary of large track error cases

       Following Carr and Elsberry (2000a, b), the hypothesis in this analysis is that the large

track error is due to an improper representation by the model of the tropical cyclone interaction

with the environmental flow. The intent of the analysis was therefore to diagnose when and

where the model was not properly predicting the correct interaction that ultimately would steer

the tropical cyclone. If the intensity or horizontal scale of a synoptic feature was found to be

incorrectly forecast, or if the timing of a transition between synoptic features was found to be

incorrect, the track errors were directly related to that incorrect prediction and assigned a specific

error mechanism.

       For the 146 (110) large-error cases in which an error mechanism can be defined for the

NOGAPS (GFDN) model, tropical influences caused 25 or 17% (14 or 13%) of the large errors,

and midlatitude influences caused 121 or 83% (96 or 87%) (Table 2). The predominance of

midlatitude influences for 96- h and 120-h track errors during the 2003 and 2004 seasons is a

marked deviation from the Carr and Elsberry (2000a, b) studies based on the 1997 season. Carr

and Elsberry found about one large 72- h errors were associated with tropical influences. Part of

the explanation is the improvement in horizontal resolution and physical processes in the

numerical models since 1997, and perhaps a better representa tion of the tropical cyclone

environment from assimilation of the satellite observations.

       A second reason for the predominance of midlatitude influences is that the 4-day and 5-

day forecasts are more likely to involve an interaction with the midlatitude circulations,

especially because the forecasts in this study (but not in the Carr and Elsberry studies) were




                                                  8
verified against extended best tracks. Thus, the percentages in Table 2 indicate that the proper

prediction of amplitude, scales, and translation of midlatitude synoptic features is a critical

component to 96-h and 120-h tropical cyclone track forecasting.

       Not all large track errors could be assigned a conceptual error mechanism. In 8 (4) cases

for NOGAPS (GFDN), the model field accuracy was acceptable, but an undiagnosed problem

with the tracking algorithm caused the large error. In 6 (4) cases for NOGAPS (GFDN), the

tropical cyclone had decayed to the point where it was no longer discernable in the verifying

analysis fields but the model still predicted a circulation (false alarm). Since the best track stops

when no circulation is present, there was no way to calculate a 120-h error. The last group that

could not be assigned an error mechanism was 2 (16) cases when no fields were available for

NOGAPS (GFDN) (Table 2).

       For convenience, error mechanisms will henceforth be referred to by their three- letter

acronym given in Table 2 with a prefix of E (excessively) or I (insufficiently), e.g., excessive-

direct cyclone interaction is abbreviated E-DCI. Each of these large-error conceptual models in

Table 2 is defined and discussed in Carr and Elsberry (2000a, b) or Carr and Elsberry (1999). It

is emphasized that except for a slight modification of one error mechanism, the same error

mechanisms as in Carr and Elsberry (2000a, b) lead to large track errors at 96 h and 120 h. That

is, no error mechanisms were discovered or required. What has changed is the decrease in the

number of tropical-related mechanisms and a proportional increase in midlatitude- influenced

errors at the longer forecast intervals. When the numerical models are upgraded, the relative

frequency of occurrence of these error mechanisms may change. Thus, a detailed examination of

the sources of trace errors needs to be conducted preferably with the beta-test sample, or as a

minimum after each season.




                                                  9
       In the following sections, conceptual models of the most important error mechanisms

leading to the large track errors will be presented and described along with the frequency of their

occurrence based on the 2004 typhoon season.

4.     Key tropical-related error mechanis m

       Each of the tropical interactions generally occurred when the tropical cyclone was south

of the subtropical ridge axis and the environmental flow had either an easterly or southerly

component or a combination of the two. Because the tropical cyclone is south of the subtropical

ridge axis, its motion is not directly affected by midlatitude synoptic features. Therefore, the

poorly predicted interaction of the tropical cyclone was typically with another warm-core

circulation (e.g., monsoon depression), or with the subtropical ridge. Of the tropical-related

errors in Table 2, only the Direct Cyclone Interaction—tropical (DCI-t) had a large number of

cases, and thus will be discussed here. Discussion of the Reverse Trough Formation and Beta-

Effect Propagation error mechanisms in Table 2 is given in Kehoe (2005).

       The conceptual model of DCI (see Fig. 2, Carr and Elsberry 2000a) involves the mutual

cyclonic rotation of two cyclones and a potential merger into one circulation that is usually larger

in size than the analyzed tropical cyclone. In the analysis of the 2004 forecasts, the track of the

larger circulation was found to be less affected than the track of the smaller circulation in E-DCI-

t. The smaller circulation usually accelerated rapidly as it rotated counterclockwise (CCW)

around the larger circulation. Such differences from the 1997 cases studied by Carr and Elsberry

are likely due to numerical model and observation improvements.

       Direct cyclone interactions in the tropics during the 2004 season were found to always be

excessively predicted by both NOGAPS or GFDN, i.e., no insufficient cases are shown in Table

2. As Carr and Elsberry (2000a) summarized, E-DCI errors (tropical or midlatitude) occurred




                                                 10
when the tropical cyclone circulation was forecast to directly interact with an adjacent cyclonic

circulation such that the predicted interaction is either false or is significantly more vigorous than

in reality. The adjacent cyclonic circulation in the E-DCI-t would often then become aligned in

such a manner that their peripheral anticyclones formed a reverse-oriented monsoon trough

(description given in Chap. IIIB.2, Kehoe 2005). The flow in the reverse-oriented monsoon

trough would then become dominant and overpower the influence the two cyclones had on each

other during DCI. Thus, both cyclones would then track to the northeast. Since it was the E-

DCI that lead to the reverse-oriented monsoon trough error mechanism, E-DCI was the

conceptual error mechanism assigned.

       The reasons for E-DCI with another real cyclonic circulation in the tropics include: (i) too

large a horizontal extent and associated outer wind strength of the tropical cyclone and/or other

cyclone in the initial analysis or forecast; (ii) misplaced tropical cyclone and/or other cyclone in

the initial analysis or forecast, such that the separation of the two cyclones is smaller than in

reality; and (iii) the 2004 analysis of NOGAPS and GFDN forecasts indicated that an improper

intensification of a tropical cyclone (weaker than reality) caused the larger circulation to

dominate the steering flow of the smaller tropical cyclone as they came in close proximity.

a.     Frequency and characteristics

       In the 2004 sample, 20 NOGAPS and 11 GFDN track forecasts with large errors involved

E-DCI-t (Table 3). The range of consecutive integrations affected ranged from as few as one by

NOGAPS in Talas (31W) to as many as nine by GFDN in Chaba (19W). In NOGAPS, other

periods of consecutive model-predicted E-DCI in the tropics included two occurrences in

Chanthu (08W), two in Mindulle (10W), seven in Chaba (19W), and eight in Aere (20W).

Consecutive E-DCI predictions by GFDN in addition to Chaba included two consecutive




                                                  11
occurrences in Aere. Especially for the longer sequences of consecutive erroneous forecasts, the

forecaster will have a better opportunity to detect the error.

       Although the environmental structure of all tropical cyclones during the period of E-DCI-

t in the tropics was classified as being in the synoptic pattern/region called Standard/ Tropical

Easterlies (S/TE), the incidence of E-DCI-t then resulted in a shift from S/TE to

Standard/Poleward Flow (S/PF). In 17 (55%) of the 31 cases, the tropical cyclone was less than

typhoon strength (64 kt) during the period of E-DCI-t.

       As in the Carr and Elsberry (2000a) study of the 1997 typhoon season in the western

North Pacific, every case of E-DCI-t in the tropics was deemed to have falsely occurred. That is,

there were no occurrences of a model exaggeration of an actual DCI in the tropics. If the

analyses of the 2004 and 1997 seasons are taken to be representative, the models are biased

towards E-DCI-t rather than real DCI. The 2004 analysis reinforces the asse rtion made by Carr

and Elsberry (2000a) that if the forecaster in real time can discern the occurrence of DCI in a

dynamic model, the probability is high that the event is excessive. The forecaster would

therefore be justified in removing the model track from consensus and forming a selective

consensus forecast track that would be more accurate. A case study will outline how the

forecaster can easily identify the key features that lead to the occurrence of E-DCI-t in the

models.

b.     Case study

       Both the NOGAPS and GFDN forecasts of Typhoon Chaba (19W) experienced E-DCI-t

problems for a total of 16 model integrations. The E-DCI-t first emerged in the 1800 UTC 19

August 2004 model integration of NOGAPS and in the 1200 UTC 19 August 2004 integration of

GFDN (Table 3, row 3). The forecast tracks for 19W (Fig. 1) by both NOGAPS and GFDN are




                                                  12
fast, and they are eastern outliers of all the models and, in hindsight, of the tropical cyclone best

track. Because the GFDN is a regional model embedded in the NOGAPS, the lateral boundary

conditions for the GFDN are interpolated from the previous NOGAPS integration. When

significant interactions of the tropical cyclone with the midlatitude flows are predicted by

NOGAPS as in Fig. 1, this may contribute to GFDN track errors via the lateral boundary

contributions

       The NOGAPS and GFDN mean sea- level pressure fields for Chaba in the 0600 UTC 21

August 2004 forecast reveals a classic case of E-DCI. Comparing the forecast mean sea- level

pressure fields of GFDN (Fig. 2, row 1, column 2) to that of the verifying NOGAPS analysis

(Fig. 2, row 3, column 2), the strength (estimated from the number of contours) of Chaba (the

eastern tropical cyclone) in the two panels is similar, but Chaba has a much smaller horizontal

scale in the GFDN fields. At the same time, GFDN portrays Aere (the western tropical cyclone)

as a much broader and less organized circulation. Despite the similar intensity forecasts for

Chaba by GFDN, the falsely forecast large horizontal scale of Aere leads to the direct interaction

between the two. The corresponding NOGAPS forecast (Fig. 2, row 2, column 2) has the

eastern tropical cyclone (C haba) much weaker and it is predicted to rotate CCW around the

western tropical cyclone (Aere). The verifying analysis fields (Fig. 2, row 3, column 2)

illustrate that Chaba has instead intensified and both Chaba and Aere have remained on a west-

northwest track. The slowing and southward deflection in the forecast track for Aere combined

with the CCW rotation of Chaba around Aere are evidence that both GFDN and NOGAPS are

predicting a mutual interaction of tropical cyclones Chaba and Aere.

       While the track forecast of Chaba in Fig. 1 may not conclusively indicate E-DCI is

occurring, the inspection of the forecast fields reveals an adjacent circulation to the tropical




                                                  13
cyclone. The motion of this adjacent circulation viewed in conjunction with the forecast motion

of the tropical cyclone should be a clue for the forecaster that mutual interaction between the two

circulations is occurring. An additional case of E-DCI-t is discussed by Kehoe (2005).

5.     Key midlatitude-related error mechanis ms

       After a tropical cyclone begins moving poleward, the transition from the tropics to the

midlatitudes is usually complete within two to three days. Because the tropical cyclone is near

or poleward of the subtropical ridge, its motion is directly impacted by midlatitude circulations

(cyclones, troughs, anticyclones, or ridges). Poorly predicted development, dissipation, and/or

movement of these midlatitude circulations, which occur independently of the tropical cyclone,

can have a negative impact on predicted tropical cyclone tracks (Carr and Elsberry 2000b).

Therefore, it is no surprise that 83% (87%) of all large errors at 96 h and 120 h in NOGAPS

(GFDN) during 2004 were due to midlatitude influences (Table 2). Large track error

mechanisms due to midlatitude influences (Table 2, bottom) were Response to Vertical Wind

Shear (RVS), Baroclinic Cyclone Interaction (BCI), Midlatitude Cyclogenes is (MCG),

Midlatitude Cyclolysis (MCL), Midlatitude Anticyclogenesis (MAG), Midlatitude Anticyclolysis

(MAL), and, added in this study, Direct Cyclone Interaction-midlatitude (DCI-m). Only those

error mechanisms that occurred frequently will be described here. More discussion of the

midlatitude-related error mechanisms is found in Kehoe (2005).

a.     Direct Cyclone Interaction – midlatitude (DCI-m)

       The conceptual model of E-DCI- m is the same as that of E-DCI-t. However, the cause of

E-DCI-m is overly deep penetration of an upper- level midlatitude cyclonic circulation into the

lower troposphere where it can affect the steering of the tropical cyclone. The E-DCI- m errors

normally occur as the tropical cyclone is moving into the midlatitude westerlies, and instead of




                                                14
moving to the east in these westerlies, the model incorrectly predicts the tropical cyclone to

rotate CCW around a large midlatitude cyclone. When this occurred during the 2004 season, the

largest 120-h errors were over 2000 n mi, which is twice as large as the largest E-DCI-t errors.

Thus, correctly identifying and removing models displaying E-DCI- m would reduce the

consensus track error.

       In the 2004 sample, the six occurrences of E-DCI- m in NOGAPS involved only Nida

(04W) and the five in GFDN involved three tropical cyclones: two occurrences in Nida, two in

Dianmu (09W), and one in Tingting (11W)(see Table 4, column 3). The environmental structure

change during each occurrence was from Standard/Poleward Flow to Midlatitude/Poleward Flow

as the tropical cyclone was moving through the axis of the subtropical ridge and interacting with

the midlatitude westerlies (Table 4, column 4). In each occurrence, the tropical cyclone was a

moderate to strong typhoon ranging from 80-135 knots (Table 4, column 5), which is significant

because a moderate vertical structure of the tropical cyclone is required for it to interact with the

overly deep penetration of the midlatitude cyclone into the lower troposphere. As in E-DCI-t, no

instances of an exaggeration of an actual DCI event by the model were observed. Rather, every

occurrence was falsely predicted to occur. Therefore, the forecaster is once again justified in

omitting the model track displaying E-DCI- m from the consensus.

       In the 0600 UTC 18 May 2004 forecast of Nida (04W), three of the four 120-h model

tracks indicated the tropical cyclone would turn to the northwest once it was north of 40N (Fig.

3). Notice that the best track has been extended to maximize the 96-h and 120- h model

verifications beyond the time JTWC declared the storm to be extratropical. To ensure continuity

with the post-tropical cyclone positions in Fig. 3, the last best-track position 0600 UTC 21 May

2004 was discarded as it was considered to be an un-representative extrapolated position




                                                 15
associated with the rapidly accelerating motion of the upper- level remnants of Nida.. In addition

to NOGAPS and GFDN, the GFS forecast also predicts a northwest track in the midlatitudes.

Whereas the model tracks form a tight cluster up to 72 h, a large spread in the models occurs

beyond 72 h.

       Both the GFDN and NOGAPS models appeared to have a good initialization (not shown)

of the tropical cyclone and the deep midlatitude 500 mb trough over eastern Russia and China.

During the second and third days, the tropical cyclone is interacting with the midlatitude trough

and has accelerated to a position near 38146E. However, the forecast positions are slow and

west of the verifying position, with some errors greater than 300 n mi (Fig. 3). It becomes

apparent in the 90-h forecasts of GFDN and NOGAPS (Fig. 4, rows 1 and 2, column 1) that the

tropical cyclone and midlatitude low are interacting as both models predict a CCW rotation with

the tropical cyclone track turning to the northwest. The 500 mb circulation of the tropical

cyclone is then absorbed into the midlatitude cyclone in both models (Fig. 4, rows 1 and 2,

column 2). However, the verifying analyses (Fig. 4, row 3, columns 1 and 2) indicate that the

tropical cyclone remnants remained in the westerlies and tracked to the southeast (Fig. 3).

       In summary, every occurrence of E-DCI-m during the 2004 season resulted from a false

interaction of a tropical cyclone with a strong midlatitude cyclone to the northwest. While the

key indicators of E-DCI- m are the same as those of E-DCI-t, an additional feature to cue the

forecaster is a departure from a northeastward track to a northwestward track in

Midlatitude/Poleward flow. Moreover, there were no occurrences of an exaggeration of an

actual DCI event between a tropical cyclone and deep midlatitude cyclone. The forecaster is

therefore justified in omitting the model displaying E-DCI- m from the consensus forecast.




                                                16
b.     Midlatitude System Evolutions (MSE)

       The MSE-related large 120-h errors by NOGAPS (GFDN) contributed 52% (65%) of all

large errors, and thus should be a major focus for the forecaster. As described in Carr and

Elsberry (2000b), the fundamental idea of the MSE error mechanisms is one of changes to the

tropical cyclone steering flow due to development, dissipation, and/or movement of midlatitude

circulations (cyclones, troughs, anticyclones, ridges) that occur essentially independent of the

tropical cyclone.

       A generalized conceptual model of MSEs is presented below (Fig. 5). Before

Midlatitude Cyclogenesis (MCG) takes place, the tropical cyclone labeled D in Fig. 5a is south

of the subtropical ridge axis and is in Standard/Tropical Easterlies pattern/region, but then

tropical cyclone D undergoes a transition to the Standard/Poleward Flow pattern/region as the

developing midlatitude trough or cyclone breaks the ridge and creates an environment of

poleward flow in the vicinity of the tropical cyclone (Fig. 5b). A vigorous MCG event could

change the direction of the environmental steering flow and result in a more poleward rather than

westward track as depicted by the transition from panels (a) to (b) in Fig. 5. Similarly, tropical

cyclone B north of the subtropical anticyclone in Fig. 5a could also have a track change to a

more poleward position. The process of Midlatitude Cyclolysis (MCL) is simply the reversed

order of MCG [i.e., (b) to (a)] depicted in Fig. 5 (Carr and Elsberry 2000b).

       When Midlatitude Anticyclogenesis (MAG) [Fig. 5, (c) to (d)] takes place, a tropical

cyclone labeled F in Fig. 5c that has been moving northward in the Standard/Poleward flow

southeast of the col in the subtropical ridge may be turned more westward as the developing

midlatitude ridge or anticyclone increases the strength of the subtropical ridge poleward of

tropical cyclone F. If MAG builds the ridge sufficiently, then the tropical cyclone will undergo a




                                                 17
transition to a Standard/Tropical Easterlies or even a Standard/Equatorward Flow pattern/region.

For example, the change in direction of the environmental steering flow may result in a more

westward rather than poleward track for tropical cyclone F as depicted in the transition from

panels (c) to (d) in Fig. 5. The MAG also may cause a track deflection of tropical cyclone E to

the north of the subtropical anticyclone in Fig. 5c. The process of Midlatitude Anticyclolysis

(MAL) is simply the reversed order of MAG [i.e., (d) to (c)] depicted in Fig. 5 (Carr and

Elsberry 2000b).

        If any of the MSEs are predicted to occur to a greater (lesser) extent by the model than in

reality such that a significant track error results, then the prefix of excessive (E) (insufficient (I))

is assigned to the event. It is stressed that the four MSE depictions in Fig. 5 are simply an

idealized representation of such events. These should be considered flexible templates in that

they can be manipulated to fit all of the complex shapes and amplitudes of the midlatitude

synoptic circulations. In addition, these MSE depictions of amplitude changes are generalized to

also allow for changes in the tropical cyclone steering flow associated with translation of the

midlatitude circulations, rather than with amplification as in Fig. 5.

        1)      FREQUENCY AND CHARACTERISTICS

                A more simple combination of the MSE-related errors of NOGAPS and GFDN

can be made by classifying the midlatitude-related errors in Table 2 into two groups. The first

group is comprised of I-MCG, E-MCL, E-MAG, and I-MAL events, which are all representative

of erroneously predicted environmental flows that are dominated by a ridge. The second group

is comprised of E-MCG, I-MCL, I-MAG, and E-MAL events, which are all representative of

erroneously predicted environmental flows that are dominated by a trough. With this

classification, the MSE events in NOGAPS were one-sided in that while NOGAPS was




                                                   18
indicating the environmental flow of the tropical cyclone would be dominated by the ridge

(Table 5), in reality the dominant feature was a midlatitude trough. Unfortunately, a similar

conclusion cannot be drawn for the GFDN errors as the MSE events in GFDN were two-sided.

That is, there were a large number of excessive and insufficient MSE events in GFDN (Table 2).

       Because of the two-sided errors in GFDN, no further conclusions could be drawn as to

systematic biases of the GFDN model always over- or under-predicting the amplitude or

translation of the midlatitude features. Thus, a more in-depth study is needed to uncover the

underlying causes of the combined total of 174 falsely predicted MSE events by NOGAPS and

GFDN (Table 5) as they accounted for many of the large errors at 96 h and 120 h. One factor to

be explored further is the extent to which the GFDN midlatitude-realted errors arise because of

improper lateral boundary conditions provided by an inaccurate NOGAPS position. Given the

sparcity of upstream observations over Russia, NOGAPS errors may be introduced in the GFDN

even though the outer domain is 75 lat. by 75 long.

       2)      INSUFFICIENT-MIDLATITUDE CYCLOGENESIS (I-MCG)

               This example will demonstrate for the forecast of Typhoon Tokage beginning

1800 UTC 17 October 2004 how the most frequently occurring error mechanism, I-MCG, caused

the GFDN track forecast error (Fig. 6). Notice that the GFDN track is very slow.

               In the 42-h GFDN forecast (Fig. 7, row 1, column 1), trough G is predicted to be

too weak relative to the verifying trough V (Fig. 7, row 3, column 1). Instead, the GFDN

forecast suggests the ridge to the northeast of Tokage will build westward. By 66 h, the isotach

maximum to the northwest of Tokage in the GFDN model indicates that the ridge to the

northwest of Tokage is the dominant influence on the environmental steering flow, and trough G

has moved eastward without ―catching‖ Tokage (Fig. 7, row 1, column 2). The verifying




                                                19
analysis for the same time illustrates that Tokage has been caught in the flow of trough V (Fig. 7,

row 3, column 2). By 66 h, the interaction of Tokage with the trough V (Fig. 7, row 3, column

2) has led to a rapid acceleration to the northeast (Fig. 6), so the 120-h GFDN track error is very

large. Thus, the I-MCG error mechanism is assigned to the GFDN forecast because it had under-

predicted the amplification of the trough G and thus did not move Tokage into the midlatitude

westerlies. Although the forecast fields for the UKMO model are not available, the similar

position as in the GFDN in Fig. 6 would suggest a similar error for the UKMO model.

       3)      EXCESSIVE-MIDLATITUDE CYCLOLYSIS (E-MCL)

               This same example will demonstrate how the E-MCL error mechanism can cause

a similar forecast track tendency as for I-MCG. In this case, the forecast track error is due to a

translation speed of the midlatitude trough that is too fast, rather than an amplitude change as

depicted in the conceptual model in Fig. 5. As described in section 5b(1), the MCL is part of the

first grouping in Table 2 and although other assignments may be appropriate at times, an MCL

assignment in this case was considered to best represent the cause of the track error. By 66 h,

trough N in the NOGAPS has moved east without ―catching‖ Tokage (Fig. 7, row 2, column 2),

while the verifying analysis illustrates that Tokage has been caught in the flow of V (Fig. 7, row

3, column 2). Later in the NOGAPS forecast (not shown), the model representation of Tokage

does interact with a second midlatitude trough and does accelerate to the northeast (Fig. 6).

               In this NOGAPS forecast, the trough was somewhat too deep, but the track error

was due to the too- fast eastward translation of the trough, which caused the tropical cyclone to

encounter a ridge instead of a trough during the middle stage of the forecast integration. The E-

MCL error mechanism is therefore assigned to the NOGAPS model. Although NOGAPS does




                                                 20
predict recurvature and poleward acceleration, it is due to the incorrect interaction with the

second midlatitude trough (not shown).

                A key result for the forecaster is that both I-MCG and E-MCL contributed to

numerous forecast degradations in both GFDN and NOGAPS. They occurred so frequently that

it would behoove the forecaster to investigate how the model in question is representing the

midlatitude trough compared to the other models when the NOGAPS and/or GFDN track

forecast is an outlier. It is also important for the forecaster to recall that once the error appears, it

will likely afflict the model in question for several successive integrations. Thus, the forecaster

should monitor the track trend display.

        4)      EXCESSIVE-MIDLATITUDE CYCLOGENESIS (E-MCG)

                The E-MCG error mechanism was the second most frequently occurring error

mechanism during the 2004 season and affected 28 (6) forecasts (Table 2). Since the GFDN has

many more E-MCG cases, these GFDN errors are not likely due to erroneous lateral boundary

conditions provided by the NOGAPS model in these cases. Because the GFDN model was more

susceptible to E-MCG, a case study of E-MCG in GFDN will be illustrated. The forecast tracks

from 1800 UTC 15 May 2004 for Nida (04W) indicate that Nida is predicted to translate faster in

both GFDN and UKMO than in the NOGAPS and GFS (Fig. 8).

                By 54 h, the trough in the GFDN forecast extends farther south and is

approaching the coast of China (Fig. 9, row 1, column 1), whereas the trough in the verifying

analysis does not have such a southern extent (Fig. 9, row 3, column 1). By contrast, the

NOGAPS forecast that had an excellent track forecast suggests that Nida (04W) is only

beginning to interact with a major midlatitude trough to the northwest.




                                                   21
                By 90 h, the GFDN forecast has merged the tropical cyclone with the deeper,

faster-moving trough (Fig. 9, row 1, column 2), while the verifying analysis indicates the trough

is farther to the west, and the tropical cyclone and trough are still separate entities (Fig. 9, row 3,

column 2). The NOGAPS forecast (Fig. 9, row 2, column 2) has a much better prediction of the

interaction between Nida (04W) and the midlatitude trough, so it is not surprising that the

NOGAPS 120-h track forecast is excellent (Fig. 8).

                While it is clear in this retrospective study that the GFDN error is due to E-MCG,

in real-time the forecaster may have suspected that the NOGAPS had an insufficient MCG error,

which happens frequently (Table 2). The fact that UKMO also predicted a similar 120- h position

as the GFDN might give support to that scenario, although the GFS 120- h forecast position in

Fig. 8 similarly might support the NOGAPS forecast. In the SAFA framework, the

recommendation would be to accept the consensus of the four model tracks if the forecaster

could not confidently establish that the GFDN (and not the NOGAPS) was likely to be in error.

        5)      EXCESSIVE-MIDLATITUDE ANTICYCLOGENESIS (E-MAG)

                The E-MAG error mechanism was the third most frequently occurring midlatitude

error (Table 2). It affected nine forecasts by the GFDN for five tropical cyclones during the

2004 season, while six forecasts in one tropical cyclone had this error in the NOGAPS. Because

the GFDN is more susceptible to E-MAG, a case study of E-MAG occurring in GFDN will be

illustrated. The track forecasts for 0600 UTC 2 September 2004 for tropical cyclone Songda

(22W) indicate that the GFDN and UKMO tracks are outliers to the far left of the other 120- h

tracks (Fig. 10). Whereas both the NOGAPS and GFS correctly predict a recurvature-type track,

both fail to predict the rapid acceleration to the northeast of Songda (22W).




                                                  22
               By 42 h (not shown), the GFDN has predicted a midlatitude anticyclone to

develop over the Yellow Sea, but this is not substantiated in the verifying analysis. The

NOGAPS forecast (not shown) has a much weaker ridge directly to the north of Songda, but the

ridge to the northeast has been predicted to extend westward too far. Consequently, the

NOGAPS track is also too far westward (Fig. 10).

               By 54 h, the GFDN predicts this midlatitude anticyclone will translate to the

northeast and merge with the subtropical ridge to form a substantial ridge to the north of Songda

(Fig. 11, row 1, column 1), which is consistent with the predicted westward track. The verifying

analysis (Fig. 11, row 3, column 1) reveals instead that this midlatitude anticyclone is a weak

ridge at 700 mb and does not add appreciably to the strength of the subtropical ridge. Thus,

Songda actually undergoes a transition to a Standard/Poleward Flow pattern/region because the

dominant steering current is the subtropical ridge to the east-northeast of the tropical cyclone

(Fig. 11, row 3, column 1). By 114 h (Fig. 11, row 1. column 2), the earlier westward and

equatorward error in the track forecast then leaves Songda too far south and west so that no

interaction with the passing midlatitude trough is predicted by GFDN. Instead, the dominant

influence on the GFDN forecast track is another strong midlatitude anticyclone that is not

substantiated by the verifying analysis (Fig. 12, row 3, column 2). W hereas the NOGAPS

forecast has a somewhat better depiction of the midlatitude circulation to the north of Songda, its

prior westward and southward track error does not put Songda in the proper position to interact

strongly with the midlatitude trough.

               A forecaster examining the model track forecasts of Fig. 10 in real- time would see

two clusters. The GFDN and UKMO models form a cluster that implies Songda will track

westward. However, the NOGAPS and GFS predict that Songda will turn poleward after 72 h.




                                                 23
Because 57 GFDN forecasts during the 2004 season were degraded by an erroneous prediction of

environmental flow dominated by a ridge (Table 5), the expectation might be that the GFDN

model is in error. By contrast, NOGAPS forecasts during 2004 were degraded by overly weak

troughs (53 occurrences of I-MCG compared to six occurrences of E-MCG in Table 2), which

results in erroneous track forecasts that remain in the tropics too long. Since NOGAPS rarely

predicts a track that is too far poleward when an MSE scenario is involved, the forecaster should

expect that the NOGAPS forecast is probably a more reliable forecast in this case.

               Although this forecast scenario is difficult in that the actual tropical cyclone track

is to the east of all four model predictions, this bifurcation scenario between a recurvature track

cluster and a westward track cluster is one in which the forecaster can add value over a non-

selective consensus track forecast by correctly rejecting the erroneous cluster containing GFDN

to create an improved selective consensus. The forecaster can then potentially improve the

selective consensus by realizing that the remaining models have a one-sided tendency to falsely

predict an environment that is dominated by a ridge (e.g., NOGAPS does not move poleward fast

enough). This requires examination of the forecast fields with thorough knowledge of the

conceptual model error mechanisms and their frequency.

c.     Excessive response to vertical wind shear (E-RVS)

       In the RVS error conceptual model introduced by Carr and Elsberry (2000b, Fig. 3), the

basic assumption is that a significant difference exists in the vertical depth between the actual

and model-predicted tropical cyclone in the presence of a vertically sheared enviro nmental flow.

A deeper (less deep) vertical extent of the tropical cyclone causes the model-predicted tropical

cyclone to have a faster (slower) translation speed (especially in the midlatitude westerlies) than

that of the actual tropical cyclone. Typically, the difference in vertical structure between the




                                                 24
model-depicted and actual tropical cyclone tends to grow with increasing forecast interval, which

accentuates the differences in translation speeds, and thus the track errors.

       Excessive-RVS (E-RVS) is said to be occurring when the model-depicted tropical

cyclone is too shallow and excessively tilted downstream. When this occurs, the upper (and

possibly middle) vortex is sheared from the lower vortex that is then advected by the low- level

environmental flow. Steering by only the low- level flow will then cause a slow bias compared to

steering by stronger upper-level flow.

       Carr and Elsberry (2000b) suggested the use of sea- level pressure forecasts to identify E-

RVS. It was found in this study that geopotential heights (not available to Carr and Elsberry in

SAFA) at 850, 700, and 500 mb were also a useful tool. Those tropical cyclones that had less

deep (deeper) vertical structures were found to have fewer (more) concentric geopotential

isopleths at higher isobaric levels. That is, fewer (more) closed geopotential isopleths existed at

500 mb and 700 mb for a less deep (deeper) tropical cyclone.

       The RVS events during the 2004 season were all excessive and occurred only in

NOGAPS (Table 2). These E-RVS events were responsible for 26 degraded forecasts in five

tropical cyclones during the 2004 season. In all cases, the tropical cyclone was in the

Standard/Poleward Flow or Midlatitude/Poleward Flow with an approaching upper-level

midlatitude trough that imposed vertical wind shear over the tropical cyclone. A key indication

(not included in the Carr and Elsberry (2000b) description) that E-RVS was occurring was when

the tropical cyclone track suddenly switched from a poleward to equatorward track, which

suggested that the upper vortex was being decoupled from the lower vortex. Since the lower

vortex was only being steered by the low- level environmental flow, the model predicted a slower

and more equatorward track.




                                                 25
               The track forecasts for Meari (25W) from 0600 UTC 26 September 2004 in Fig.

12 have a large spread about the consensus mean (not shown) with longer-range forecast tracks

that are highly diverse. The NOGAPS track forecast has a sudden reversal fro m poleward to

equatorward flow near 31131E, which may indicate E-RVS is a possible error. Whereas the

GFDN and GFS models also have a reversal of track directions, they have a more arcing track. It

is suspected that the UKMO forecast track was also affected by E-RVS, but no forecast fields

were available to investigate further. Comparison with the verifying analysis (Fig. 13, row 2,

column 1) confirms that the NOGAPS forecast (Fig. 13, row 1, column 1) has an even weaker

700 mb circulation and smaller horizontal scale. By 90 h, only one closed 700 mb isopleth is

predicted, and it lags behind the midlatitude trough (Fig. 13, row 1, column 2). In addition, the

500-mb forecast (not shown) indicates the upper and lower vortices have separated with the 500

mb vortex farther to the northeast, which is another indicator of vertical wind shear (Carr and

Elsberry 2000b). By contrast, the verifying 700 mb analysis (Fig. 13, row 2, column 2) has

Meari embedded in the trough and with a much stronger circulation.

               Several clues that a tropical cyclone moving poleward is experiencing vertical

wind shear are described above. The NOGAPS (and likely other global models) are susceptible

to vertical wind shear because the resiliency of the actual vortex cannot be simulated on a coarse

grid. If the NOGAPS predicts a slowing of the motion, or a turn equatorward as in this example,

the E-RVS mechanism should be expected. The proper identification and removal of the

NOGAPS track forecast displaying E-RVS would normally provide a selective consensus that is

more accurate than the average of all four model tracks. However, in the 0600 UTC 26

September 2004 forecast for tropical cyclone Meari, all four forecast tracks (Fig. 12) are outliers

and therefore eliminating just the NOGAPS track would go against the rules set forth in SAFA.




                                                26
6.     Summary and conclusions

       This study has extended the large track error conceptual models of Carr and Elsberry

(2000a, b) to 96- and 120-h forecasts by the NOGAPS and GFDN models during the 2004

western North Pacific season. Large track errors are defined to be 400 n mi at 96 h and 500 n mi

at 120 h. This study maximized the number of 120-h forecast verifications for both NOGAPS

and GFDN by manually extending the best-track positions beyond the point of extratropical

declaration using mean sea- level pressure analyses. By following these procedures, an increase

of nearly 28% (24%) in the number of cases with large 120- h forecast errors was realized for

NOGAPS (GFDN).

       An important conclusion is that all of these large-error cases not associated with false

alarms, tracker errors, or field non-availability could be attributed to the models not properly

representing the physical processes known to control tropical cyclone motion, which were

classified in a series of conceptual models by Carr and Elsberry for either tropica l-related or

midlatitude-related mechanisms. In contrast to the large 72-h errors during the 1997 season

studied by Carr and Elsberry, a smaller fraction of the 96-h and 120- h errors are tropical-related,

with only 17% (13%) for the NOGAPS (GFDN) forecasts. However, the Direct Cyclone

Interaction error mechanism continues to be the most frequent contributor to large errors, and

always in the sense of being excessively predicted in the forecasts.

       For those large-error cases that an error mechanism could be established, midlatitude

influences caused 83% (87%) of the NOGAPS (GFDN) errors. The most common midlatitude-

related errors in the NOGAPS tracks arise from an erroneous prediction of the environmental

flow dominated by a ridge in the midlatitudes. Errors in the GFDN tracks are caused by both

ridge-dominated and trough-dominated environmental flows in the midlatitudes. Case studies




                                                 27
illustrating the key error mechanisms indicate that the proper prediction of the amplitude, scales,

and transition of midlatitude synoptic features is a critical component to 120-h tropical cyclone

track forecasting. The NOGAPS and GFDN models had significant problems both in the

development and movement of midlatitude troughs. Now that some of the four longer-range

forecasts are available each 6 h, the tendency for repeated error mechanisms of the same type is

more obvious, and the forecaster may be able to detect the error by examining the track trends

over 36-48 h.

       As expected, a global model such as NOGAPS is susceptible to vertical wind shear

causing excessive decay of the tropical cyclone vortex, and this is an important factor if the

cyclone moves toward the midlatitudes during the 120-h forecast. Both the forecast wind fields

and geopotential fields provide evidence that the vortex is being excessively decoupled during

these vertical wind shear scenarios.

       The hypothesis in this study is that the proper identification and removal of any model

track forecast displaying one of these conceptual error mechanisms could provide a selective

consensus that is more accurate than a non-selective consensus of all four longer-range forecasts.

As outlined in Carr and Elsberry (2000a, b), this would require the availability and the capability

to display the analysis and forecast fields from the model. Application of the conceptual models

presented here requires either the streamlines/isotachs, the sea- level pressures, or the

geopotential heights. The ability to determine conclusively the existence of these likely

erroneous tracks in real- time is not tested in this retrospective study of known large errors.

       An ability to identify confidently these error mechanisms and thereby eliminate likely

erroneous tracks from the consensus has been shown to improve the accuracy of 72-h track

forecasts (Elsberry and Carr 2000). However, removing even one of the four 120- h model tracks




                                                  28
that comprise the 96- h and 120-h consensus does raise concerns about the accuracy of the

remaining three- model consensus. Jeffries and Fukada (2002) noted that the consensus forecasts

were more accurate if more than three track forecasts were available. Thus, a dilemma occurs

when one of the four 120- h models has a highly erroneous track, since having only three model

tracks may cause the consensus to be degraded. Therefore, it remains to be demonstrated that an

improved consensus track forecast can be achieved by eliminating the positively identified

erroneous track through the conceptual error models outlined in this study. Adding more skillful

longer-range track forecast models to the consensus will help if their error characteristics are also

known.

Acknowledgements. This work is part of the M. S. thesis of R. M. Kehoe (Capt., USAF), who

was supported by the Air Force Institute of Technology. M. A. Boothe and R. L. Elsberry were

supported by the Office of Naval Research Marine Meteorology Division. The data sets were

provided by the Joint Typhoon Warning Center in Hawaii. The manuscript was expertly

prepared by Mrs. Penny Jones.




                                                 29
                                         REFERENCES

Carr, L. E., III and R. L. Elsberry, 1999: Systematic and Integrated Approach to Tropical

       Cyclone Track Forecasting. Part III: Traits Knowledge Base for JTWC Track Forecast

       Models in the Western North Pacific. Tech. Rep. NPS-MR-99-002, Naval Postgraduate

       School, Monterey, California 93943-5114, 227 pp.



Carr, L. E., III, and R. L. Elsberry, 2000a: Dynamical tropical cyclone track forecast errors. Part

       I: Tropical region errors. Wea. Forecasting, 15, 641-661.



Carr, L. E., III, and R. L. Elsberry, 2000b: Dynamical tropical cyclone track forecast errors.

       Part II: Midlatitude circulation influences. Wea. Forecasting, 15, 662-681.



Carr, L. E., III, and R. L. Elsberry, and J. E. Peak, 2001: Beta test of the systematic approach

       expert system prototype as a tropical cyclone track forecasting aid. Wea. Forecasting,

       16, 355-368.



Elsberry R. L., and L. E. Carr, III, 2000: Consensus of dynamical tropical cyclone track

       forecasts – error versus spread. Mon. Wea. Rev., 128, 4131-4138.



Goerss, J. 2000: Tropical cyclone track forecasts using an ensemble of dynamical models. Mon.

       Wea. Rev., 128, 1187-1193.




                                                30
Jeffries, R. A., and E. J. Fukada, 2002: Consensus approach to track forecasting. Paper TP3.2,

       Extended Abstracts, Fifth International Workshop on Tropical Cyclones, Cairns,

       Australia, World Meteorological Organization (Geneva).



Kehoe, R. M., 2005: Characteristic errors in 120-h tropical cyclone track forecasts in the

       western North Pacific. M. S. thesis, Naval Postgraduate School, 111 pp. [available at

       http://theses.nps.navy.mil/05Mar_Kehoe.pdf]




                                                31
                                       LIST OF FIGURES



Fig. 1. Interpolated forecast tracks for 19W by NOGAPS, GFDN, UKMO, and GFS for the 5-

day forecast (symbols each 24 h) beginning 0600 UTC 21 August 2004. The bold, solid line

with circles represents the tropical cyclone best track.



Fig. 2. Mean sea- level pressure (mb, contour interval 2 mb) fields for 19W at the initial time

(column 1) and the forecasts (column 2) by GFDN (row 1) and NOGAPS (row 2). The verifying

NOGAPS analysis (row 3) is at 1800 UTC 20 Aug 2004.



Fig. 3. As in Fig. 1, except interpolated forecast tracks for 04W by NOGAPS, GFDN, UKMO,

and GFS for the 5-day forecast beginning 0600 UTC 18 May 2004.              The extended track

verification is represented by the heavy dashed line and solid circles.



Fig. 4. Similar to Fig. 2, except for 500- mb streamline and isotach (shaded, beginning 20 kt,

contour interval 10 kt) forecast fields for 04W by GFDN (row 1) and NOGAPS (row 2) and

verifying NOGAPS analyses (row 3) at the times indicated. First column contains the 90-h

forecasts and the second column contains the 114-h forecast for NOGAPS and the 108-h GFDN

forecast because the later forecast was not available.       Note that the tropical cyclone was

approaching the eastern boundary of the GFDN domain.



Fig. 5. Schematics of the Midlatitude System Evolutions (MSEs) affecting the subtropical

anticuyclone cells (A) and thus the steering- level flow (arrows) that may lead to large tropical




                                                 32
cyclone (labeled B, D, E, and F) track changes with larger or smaller poleward motions. The

deepening of the midlatitude trough from (a) to (b) depicts MCG and the reverse order [(b) to

(a)] implies MCL.      Similarly, the midlatitude anticyclone change poleward of the tropical

cyclone from (c) to (d) depicts MAG and the reverse order [(d) to (c)] implies MAL (from Carr

and Elsberry 2000b).



Fig. 6. As in Fig. 1, except for interpolated forecast tracks for 27W by NOGAPS, GFDN,

UKMO, and GFS for the 5-day forecast beginning 1800 UTC 17 October 2004. The extended

track verification is represented by the heavy dashed line.



Fig. 7. Similar to Fig. 4, except for 700- mb streamline and isotach forecast fields for 27W by

GFDN (row 1) and NOGAPS (row 2) and verifying NOGAPS analyses (row 3) for the times

indicated. First column contains the 42-h forecasts and the second column contains the 66-h

forecasts. Troughs in the GFDN, NOGAPS, and verifying analyses are labeled G, N and V.



Fig. 8. As in Fig. 1, except for interpolated forecast tracks for 04W by NOGAPS, GFDN,

UKMO, and GFS for the 5-day forecast beginning 1800 UTC 15 May 2004.



Fig. 9. As in Fig. 4, except for 500-mb streamline and isotach forecast fields for 04W by GFDN

(row 1) and NOGAPS (row 2) and verifying NOGAPS analyses (row 3) for the times indicated.

First column contains the 54- h forecasts and the second column contains the 90-h forecasts.




                                                33
Fig. 10. As in Fig. 1, except for interpolated forecast tracks for 22W by NOGAPS, GFDN,

UKMO, and GFS for the 5-day forecast beginning 0600 UTC 2 September 2004.



Fig. 11. As in Fig. 7, except for 700- mb streamline and isotach forecast fields for 22W by

GFDN (row 1) and NOGAPS (row 2) and verifying NOGAPS analyses (row 3) for the times

indicated. First column contains the 54-h forecasts and the second column contains the 114-h

forecasts.



Fig. 12. As in Fig. 1, except for interpolated forecast tracks for 25W by NOGAPS, GFDN,

UKMO, and GFS for the forecast beginning 0600 UTC 26 September 2004. The extended track

verification is represented by the heavy dashed line.



Fig. 13. Forecasts of 700- mb geopotential height (decameters) fields for 25W by NOGAPS (row

1) and verifying NOGAPS analysis (row 2) for the forecast of 0000 UTC 26 September 2004.

First (second) column contains the 66- h (90-h) forecasts.




                                                34
                                                                   GFDN
                          NOGAPS




                          GFS
          UKMet
                                                  TC 19W
                                                   Chaba
       Actual
       position
       in 5 days




Fig. 1. Interpolated forecast tracks for 19W by NOGAPS, GFDN, UKMO, and GFS for the 5-
day forecast (symbols each 24 h) beginning 0600 UTC 21 August 2004. The bold, solid line
with circles represents the tropical cyclone best track.




                                          35
                  GFDN Analysis                         GFDN Forecast (valid in 66 h)




                NOGAPS Analysis                        NOGAPS Forecast (valid in 66 h)




                                                   NOGAPS Verifying Analysis in 66 h




Fig. 2. Mean sea- level pressure (mb, contour interval 2 mb) fields for 19W at the initial time
(column 1) and the forecasts (column 2) by GFDN (row 1) and NOGAPS (row 2). The verifying
NOGAPS analysis (row 3) is at 1800 UTC 20 Aug 2004.




                                              36
                      GFDN
                                                    TC 04W
                    NOGAPS
            GFS
                                                     Nida
                                            UKMet




                                                     Actual
                                                     position
                                                     in 5 days



Fig. 3. As in Fig. 1, except interpolated forecast tracks for 04W by NOGAPS, GFDN, UKMO,
and GFS for the 5-day forecast beginning 0600 UTC 18 May 2004. The extended track
verification is represented by the heavy dashed line and solid circles.




                                          37
           GFDN Forecast (valid in 90 h)             GFDN Forecast (valid in 108 h)




          NOGAPS Forecast (valid in 90 h)           NOGAPS Forecast (valid in 114 h)




        NOGAPS Verifying Analysis in 90 h         NOGAPS Verifying Analysis in 114 h




Fig. 4. Similar to Fig. 2, except for 500- mb streamline and isotach (shaded, beginning 20 kt,
contour interval 10 kt) forecast fields for 04W by GFDN (row 1) and NOGAPS (row 2) and
verifying NOGAPS analyses (row 3) at the times indicated. First column contains the 90-h
forecasts and the second column contains the 114-h forecast for NOGAPS and the 108-h GFDN
forecast because the later forecast was not available. Note that the tropical cyclone was
approaching the eastern boundary of the GFDN domain.




                                             38
          Midlatitude System Evolutions (MSE)



         A                       A                  A                       A
 (a)                             D           (b)                          D

                                                                           E
                           E
         A                       A                  A                       A
                                                                    F
                            F
 (c)                                         (d)


Fig. 5. Schematics of the Midlatitude System Evolutions (MSEs) affec ting the subtropical
anticuyclone cells (A) and thus the steering- level flow (arrows) that may lead to large tropical
cyclone (labeled B, D, E, and F) track changes with larger or smaller poleward motions. The
deepening of the midlatitude trough from (a) to (b) depicts MCG and the reverse order [(b) to
(a)] implies MCL. Similarly, the midlatitude anticyclone change poleward of the tropical
cyclone from (c) to (d) depicts MAG and the reverse order [(d) to (c)] implies MAL (from Carr
and Elsberry 2000b).




                                               39
                                        GFS


                        NOGAPS                                  Actual
                                                              position
                                                             in 5 days




        GFDN

                        UKMet                   TC 27W
                                                Tokage


Fig. 6. As in Fig. 1, except for interpolated forecast tracks for 27W by NOGAPS, GFDN,
UKMO, and GFS for the 5-day forecast beginning 1800 UTC 17 October 2004. The extended
track verification is represented by the heavy dashed line.




                                         40
           GFDN Forecast (valid in 42 h)              GFDN Forecast (valid in 66 h)



                        G                                               G




          NOGAPS Forecast (valid in 42 h)            NOGAPS Forecast (valid in 66 h)
                         N                                              N




         NOGAPS Verifying Analysis in 42 h          NOGAPS Verifying Analysis in 66 h

                       V                                                V




Fig. 7. Similar to Fig. 4, except for 700- mb streamline and isotach forecast fields for 27W by
GFDN (row 1) and NOGAPS (row 2) and verifying NOGAPS analyses (row 3) for the times
indicated. First column contains the 42- h forecasts and the second column contains the 66-h
forecasts. Troughs in the GFDN, NOGAPS, and verifying analyses are labeled G, N and V.




                                              41
                                               GFDN
                TC 04W
                 Nida                                     UKMet

                                                    NOGAPS
                                                       Actual
                                                     position
                                                GFS in 5 days




Fig. 8. As in Fig. 1, except for interpolated forecast tracks for 04W by NOGAPS, GFDN,
UKMO, and GFS for the 5-day forecast beginning 1800 UTC 15 May 2004.




                                         42
       GFDN Forecast (valid in 54 h)               GFDN Forecast (valid in 90 h)




     NOGAPS Forecast (valid in 54 h)              NOGAPS Forecast (valid in 90 h)




     NOGAPS Verifying Analysis in 54 h         NOGAPS Verifying Analysis in 90 h




Fig. 9. As in Fig. 4, except for 500-mb streamline and isotach forecast fields for 04W by GFDN
(row 1) and NOGAPS (row 2) and verifying NOGAPS analyses (row 3) for the times indicated.
First column contains the 54- h forecasts and the second column contains the 90-h forecasts.




                                             43
          TC 22W
                                                           Actual
          Songda                                         position
                            NOGAPS                      in 5 days

                           GFS




         GFDN       UKMet




Fig. 10. As in Fig. 1, except for interpolated forecast tracks for 22W by NOGAPS, GFDN,
UKMO, and GFS for the 5-day forecast beginning 0600 UTC 2 September 2004.




                                          44
      GFDN Forecast (valid in 54 h)              GFDN Forecast (valid in 114 h)




     NOGAPS Forecast (valid in 54 h)          NOGAPS Forecast (valid in 114 h)




    NOGAPS Verifying Analysis in 54 h         NOGAPS Verifying Analysis in 114 h




Fig. 11. As in Fig. 7, except for 700- mb streamline and isotach forecast fields for 22W by
GFDN (row 1) and NOGAPS (row 2) and verifying NOGAPS analyses (row 3) for the times
indicated. First column contains the 54-h forecasts and the second column contains the 114-h
forecasts.




                                            45
              GFS
                                                           Actual
                                                         position
                                                        in 5 days
                                      NOGAPS



                                                  TC 25W
                                                   Meari
        UKMet           GFDN



Fig. 12. As in Fig. 1, except for interpolated forecast tracks for 25W by NOGAPS, GFDN,
UKMO, and GFS for the forecast beginning 0600 UTC 26 September 2004. The extended track
verification is represented by the heavy dashed line.




                                          46
         NOGAPS Forecast (valid in 66 h)           NOGAPS Forecast (valid in 90 h)




        NOGAPS Verifying Analysis in 66 h         NOGAPS Verifying Analysis in 90 h




Fig. 13. Forecasts of 700- mb geopotential height (decameters) fields for 25W by NOGAPS (row
1) and verifying NOGAPS analysis (row 2) for the forecast of 0000 UTC 26 September 2004.
First (second) column contains the 66- h (90-h) forecasts.




                                            47
                                       LIST OF TABLES



Table 1. Number of cases of large forecast errors at 96 h and 120 h for 2004. First total indicates

number of verifying forecast positions from best track data. Second total, in parentheses,

includes verifying positions extended beyond declared extratropical transition to maximize 96-h

and 120-h model verifications.



Table 2. 96-h and 120-h error mechanisms for NOGAPS and GFDN occurring in 2004. The

first (second) number listed is the number of times the phenomenon occurred excessively

(insufficiently).



Table 3. Cases of model-predicted E-DCI-t in the western North Pacific in 2004. A total of 31

cases of E-DCI-t occurred in five tropical cyclones during 2004. Intensity is measured in knots.

A probable tropical circulation is indicated by ―PTC‖.



Table 4. Cases of model-predicted E-DCI-m in three western North Pacific tropical cyclones in

2004.



Table 5. Two groups of Midlatitude System Evolution-related errors from Table 2.




                                                48
Table 1. Number of cases at 96 h and 120 h for 2004. First total indicates number of verifying
forecast positions from best-track data. Second total, in parentheses, includes verifying positions
extended beyond declared extratropical to maximize 96-h and 120-h model verifications.

   Model           Year       No. Storms        No. 96-h forecast          No. 120-h forecast


 NOGAPS            2004            32                367 (422)                  277 (354)


   GFDN            2004            32                283 (318)                  211 (262)




                                                49
Table 2. 96-h and 120-h error mechanisms for NOGAPS and GFDN occurring in 2004. The
first (second) number listed is the number of times the phenomenon occurred excessively
(insufficiently).

                                                          No. of         No. GFDN
         Phenomenon name                   Acronym      NOGAPS           forecasts
                                                        forecasts
    Large Errors due to Tropical

              Influences

Direct cyclone interaction (tropical)       DCI-t          20-0             11-0
Reverse trough formation                    RTF            0-0              3-0
Beta effect propagation                     BEP            0-5              0-0
   Large Errors due to Midlatitude

              Influences

Direct cyclone interaction (midlatitude)    DCI-m          6-0              5-0
Response to vertical wind shear              RVS           26-0             0-0
Baroclinic cyclone interaction               BCI           6-0              0-0
Midlatitude cyclogenesis                    MCG            6-53            28-46
Midlatitude cyclolysis                       MCL           12-0             2-0
Midlatitude anticyclogenesis                MAG            6-0              9-6
Midlatitude anticyclolysis                   MAL           2-4              0-0
False Alarm                                                 6                4
Tracker Error                                               8                4
Fields not available                                        2                16
Total of all large-error forecasts                         162              134




                                           50
Table 3. Cases of model-predicted E-DCI-t in the western North Pacific in 2004. A total of 31
cases of E-DCI-t occurred in five tropical cyclones during 2004. Intensity is measured in knots.
A probable tropical circulation is indicated by ―PTC‖.


  TC        Start time of         No.       Synoptic       Intensity      Nature of     Location
  No.      affected model        cases:    environme        during         second       of second
                runs           NOGAPS          nt of      interaction     cyclone        cyclone
                                (GFDN)     affected TC
 08W      1800 UTC 08
          Jun-                     2           S/TE          35-40          07W           NNW
          0600 UTC 09 Jun
 10W      1800 UTC 23
          Jun-                     2           S/TE          35-45        Pre-11W           E
          0000 UTC 24 Jun
 19W      1200 UTC 19
          Aug-                   7(9)          S/TE          45-90       Pre-20W           W
          1200 UTC 21 Aug                                                   20W
 20W      1800 UTC 19
          Aug-                   8 (2)         S/TE          30-65          19W             E
          1200 UTC 21 Aug

 31W      0000 UTC 09 Dec          1           S/TE            30            PTC            E




                                              51
Table 4. Cases of model-predicted E-DCI-m in three western North Pacific tropical cyclones in
2004.

TC     Start time of    Occurrences   Synoptic    Intensity             Nature/location
No.      affected        NOGAPS     environment    before                 of second
       model runs         (GFDN)     of affected interaction               cyclone
                                         TC
04W    0000 UTC 17
           May               6 (2)          S/PF          130-135        Midlatitude
       0600 UTC 18                           M/PF                          to NNW
           May
09W    1800 UTC 23                                                        Midlatitude
          June                (2)           S/PF          115-120         to NNW
       0000 UTC 24                           M/PF
          June
11W    0000 UTC 29
          June                (1)           S/PF             80          Midlatitude
                                             M/PF                          to NNW




                                             52
Table 5. Two groups of Midlatitude System Evolution-related errors from Table 2.

          Phenomenon                      No. of NOGAPS              No. of GFDN
                                            forecasts                 forecasts
Erroneous prediction of
environmental flow dominated by                   75                      57
a ridge
Erroneous prediction of
environmental flow dominated by                   8                       34
a trough
Total of all degraded forecasts                   83                      91




                                             53

								
To top