• Chang, E. K. M., 2005: The impact of wave packets propagating across Asia on Pacific cyclone development. Mon. Wea. Rev., 133, 19982015.

    • Search Google Scholar
    • Export Citation
  • Charles, M., 2008: Verification of extratropical cyclones within NCEP forecast models using an automated cyclone tracking algorithm. M.S. thesis, School of Marine and Atmospheric Sciences, Stony Brook University, 135 pp. [Available from MSRC, Stony Brook University, Stony Brook, NY 11794-5000.]

    • Search Google Scholar
    • Export Citation
  • Charles, M., , and Colle B. A. , 2009a: Verification of extratropical cyclones within the NCEP operational models. Part I: Analysis errors and short-term NAM and GFS forecasts. Wea. Forecasting, 24, 11731190.

    • Search Google Scholar
    • Export Citation
  • Charles, M., , and Colle B. A. , 2009b: Verification of extratropical cyclones within the NCEP operational models. Part II: The short-range ensemble forecast system. Wea. Forecasting, 24, 11911214.

    • Search Google Scholar
    • Export Citation
  • Elmore, K. L., , Schultz D. M. , , and Baldwin M. E. , 2006: The behavior of synoptic-scale errors in the Eta Model. Mon. Wea. Rev., 134, 33553366.

    • Search Google Scholar
    • Export Citation
  • Froude, L. S. R., 2009: Regional differences in the prediction of extratropical cyclones by the ECMWF Ensemble Prediction System. Mon. Wea. Rev., 137, 893911.

    • Search Google Scholar
    • Export Citation
  • Froude, L. S. R., 2010: TIGGE: Comparison of the prediction of Northern Hemisphere extratropical cyclones by different ensemble prediction systems. Wea. Forecasting, 25, 819836.

    • Search Google Scholar
    • Export Citation
  • Froude, L. S. R., , Bengtsson L. , , and Hodges K. I. , 2007: The prediction of extratropical storm tracks by the ECMWF and NCEP Ensemble Prediction Systems. Mon. Wea. Rev., 135, 25452567.

    • Search Google Scholar
    • Export Citation
  • Hakim, G. J., 2003: Developing wave packets in the North Pacific storm track. Mon. Wea. Rev., 131, 28242837.

  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437471.

  • Langland, R. H., , Shapiro M. A. , , and Gelaro R. , 2002: Initial condition sensitivity and error growth in forecasts of the 25 January 2000 East Coast snowstorm. Mon. Wea. Rev., 130, 957974.

    • Search Google Scholar
    • Export Citation
  • Mahoney, K. M., , and Lackmann G. M. , 2006: The sensitivity of numerical forecasts to convective parameterization: A case study of the 17 February 2004 East Coast cyclone. Wea. Forecasting, 21, 465488.

    • Search Google Scholar
    • Export Citation
  • McMurdie, L. A., , and Mass C. , 2004: Major numerical forecast failures over the northeast Pacific. Wea. Forecasting, 19, 338356.

  • McMurdie, L. A., , and Casola J. H. , 2009: Weather regimes and forecast errors in the Pacific Northwest. Wea. Forecasting, 24, 829842.

  • Mesinger, F., and Coauthors, 2006: North American Regional Reanalysis. Bull. Amer. Meteor. Soc., 87, 343360.

  • Mullen, S. L., , and Smith B. B. , 1990: An analysis of sea-level cyclone errors in NMC’s Nested Grid Model (NGM) during the 1987–88 winter season. Wea. Forecasting, 5, 433447.

    • Search Google Scholar
    • Export Citation
  • Nutter, P. A., , Mullen S. L. , , and Baumhefner D. P. , 1998: The impact of initial condition uncertainty on numerical simulations of blocking. Mon. Wea. Rev., 126, 24822502.

    • Search Google Scholar
    • Export Citation
  • Roebber, P. J., , and Tsonis A. A. , 2005: A method to improve prediction of atmospheric flow transitions. J. Atmos. Sci., 62, 38183824.

    • Search Google Scholar
    • Export Citation
  • Sellwood, K. J., , Majumdar S. J. , , Mapes B. E. , , and Szunyogh I. , 2008: Predicting the influence of observations on medium-range forecasts of atmospheric flow. Quart. J. Roy. Meteor. Soc., 134, 20112027.

    • Search Google Scholar
    • Export Citation
  • Shapiro, M. A., , and Thorpe A. J. , 2004: THORPEX International Science Plan, version 3. WMO/TD 1246, WWRP/THORPEX 2, 51 pp. [Available online at http://www.wmo.int/pages/prog/drr/publications/drrPublications/TD1246_THORPEX_International_Science_Plan/WMO_TD1246e.pdf.]

    • Search Google Scholar
    • Export Citation
  • Smith, B. B., , and Mullen S. L. , 1993: An evaluation of sea level cyclone forecasts produced by NMC’s Nested-Grid Model and Global Spectral Model. Wea. Forecasting, 8, 3756.

    • Search Google Scholar
    • Export Citation
  • Swanson, K. L., , and Roebber P. J. , 2008: The impact of analysis error on medium-range weather forecasts. Mon. Wea. Rev., 136, 34253431.

    • Search Google Scholar
    • Export Citation
  • Toth, Z., , and Kalnay E. , 1997: Ensemble forecasting at NCEP and the breeding method. Mon. Wea. Rev., 125, 32973319.

  • Wedam, G. B., , McMurdie L. A. , , and Mass C. F. , 2009: Comparisons of model forecast skill over the east and west coasts of the United States. Wea. Forecasting, 24, 843854.

    • Search Google Scholar
    • Export Citation
  • Zhang, F., , Snyder C. , , and Rotunno R. , 2003: Effects of moist convection on mesoscale predictability. J. Atmos. Sci., 60, 11731185.

  • View in gallery

    Geographical regions used for the GFS cyclone verification.

  • View in gallery

    (a) Cyclone mean absolute errors vs forecast hour for the 2002–07 cool seasons for the GFS regions specified in Fig. 1. (b) As in (a), but for the relatively deep cyclones exceeding the thresholds in Table 1. Confidence intervals at the 90% significance level are given by the vertical bars.

  • View in gallery

    Cyclone central pressure (a) MAEs (shaded in hPa) and (b) mean errors (shaded in hPa) averaged for the 72–120-h GFS forecasts during the 2002–07 cool seasons.

  • View in gallery

    As in Fig. 2a, but for cyclone displacement (km).

  • View in gallery

    Histograms showing the frequency of model cyclone positions relative to the observed position (center point) for 45° bins centered on N, NE, E, etc. in the GFS for hours (a),(c),(e) 102–120 and (b),(d),(f) 36–48 for regions (a),(b) EP, (c),(d) C, and (e),(f) WA. The dashed range circles are every 10%, from 0 to 30. The solid range circles represent the 90% confidence intervals below and above the range of the bins. The numbers for each radial represent the average displacement error (km) within that directional bin. The black vectors indicate the mean displacement vector, with each dashed range ring every 50 km.

  • View in gallery

    GFS mean SLP error normalized against the climatological error (shaded every 0.25 hPa), mean observed 500-hPa heights (dash every 60 m), and mean observed SLPs (solid every 2 hPa), averaged at each grid point for all CP cyclones with central pressure absolute errors more than 1.5 std dev greater than the mean (28 cases) for hours (a) 30, (b) 48, (c) 66, (d) 84, and (e) 102 h. (f)–(j) As in (a)–(e), but for the small error cyclone events at hour 30 over the CP (errors <0.25 std dev).

  • View in gallery

    (left) GFS 500-hPa geopotential height errors normalized against the climatological error (shaded every 5 m) and observed 500-hPa heights (dashed every 60 m) at hours (a) 30, (c) 48, (e) 66, (g) 84, and (i) 102 for relatively large (>1.5 std dev) central pressure errors at 102 h within the E region in Fig. 1. (right) GFS mean SLP error normalized against the climatological error (shaded every 0.25 hPa), mean observed 500-hPa heights (dash every 60 m), and mean observed SLPs (solid every 2 hPa), averaged at each grid point for relatively large (>1.5 std dev) central pressure errors at 102 h within the E region at hours (b) 30, (d) 48, (f) 66, (h) 84, and (j) 102.

  • View in gallery

    Forecast cyclone tracks for WA region GFS cyclones with large (>1.5 std dev) central pressure errors for cases with (a) positive and (b) negative central pressure errors at hour 96 that are more than 1.5 std dev above (below) the mean error of all cyclones with positive (negative) error. The gray line segments indicate 6-h periods when the forecast cyclone growth rate (weakening) is slower (faster) than observed. Meanwhile, black lines indicate 6-h periods when the forecast cyclone is deepening (or filling) faster (slower) than observed. The black × symbols indicate the locations of the forecast cyclones at hour 96.

  • View in gallery

    Daily NCEP–NCAR reanalysis composite showing 500-hPa geopotential heights (color shaded and contoured every 60 m) for the relatively large (1.5 std dev) negative cyclone pressure error events in the GFS at hour 96 at (a) day −2, (c) day 0 (initialization), and (e) day +4 (hour 96). (b),(d),(f) As in (a),(c),(e) but for the positive cyclone pressure GFS error events. Tables 2 and 3 list the negative and positive error events, respectively.

  • View in gallery

    As in Fig. 9, but for the 500-hPa geopotential height anomalies (color shaded and contoured every 10 m).

  • View in gallery

    The 500-hPa geopotential height evolution for the GFS (solid every 60 m) and observed (shaded every 50 m) for the GFS run initialized at 1800 UTC 11 Jan 2004 for hours (a) 66, (b) 84, (c) 96, and (d) 102. The locations of the observed and GFS cyclones are given by the × and + symbols, respectively, with the top number the observed and the bottom the GFS.

  • View in gallery

    As in Fig. 11, but for hours (a) 60, (b) 84, and (c) 96 for the GFS run initialized at 0000 UTC 4 Dec 2003.

  • View in gallery

    As in Fig. 8, but for those GFS cyclones within the EP region with 1.5 standard deviation (a) negative and (b) positive central pressure errors at hour 72.

  • View in gallery

    Time series of GFS central pressure errors (hPa) at hour 48 for the CP region during the 2002–07 cool seasons. Each bar represents the average central pressure error for all cyclones present in the CP region at hour 48. Horizontal black dashed lines denote 2 std dev above and below the mean central pressure error. The boldface black line is the 5-day running mean of the cyclone central pressure errors. The vertical gray lines represent the warm season break between March and October.

  • View in gallery

    Monthly composite of the NCEP–NCAR reanalysis showing 500-hPa geopotential heights (contoured every 60 m) for (a) January and (b) December 2005. (c) As in (a), but for the 500-hPa geopotential height anomalies (every 20 m). (d) As in (c), but for December 2005.

  • View in gallery

    Composite of daily NCEP–NCAR reanalysis showing 500-hPa geopotential heights (shaded and contoured every 60 m) for persistent (see text for details) negative cyclone pressure errors at hour 48 in the CP region at (a) the time of initialization (day 0) and (c) hour 48 (day 2). (b),(d) As in (a),(c), but for positive GFS cyclone error events. (e),(f) As in (c),(d), but for the 500-hPa geopotential height anomalies (shaded and contoured every 10 m). (g),(h) As in (c),(d), but for the SLP (shaded and every 2 hPa).

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 93 93 17
PDF Downloads 20 20 3

Spatial Distribution and Evolution of Extratropical Cyclone Errors over North America and its Adjacent Oceans in the NCEP Global Forecast System Model

View More View Less
  • 1 School of Marine and Atmospheric Sciences, Stony Brook University, Stony Brook, New York
  • | 2 NOAA/NWS/NCEP/Climate Prediction Center, Camp Springs, Maryland
© Get Permissions
Full access

Abstract

Short- to medium-range (1–5 day) forecasts of extratropical cyclones around North America and its adjacent oceans are verified within the Global Forecast System (GFS) model at the National Centers for Environmental Prediction (NCEP) during the 2002–07 cool seasons (October–March). Cyclones in the immediate lee of the Rockies and U.S. Great Plains have 25%–50% smaller pressure errors than other regions after hour 36. The central pressure and displacement errors are largest over the central and eastern Pacific for the 42–72-h forecast, while the western and central Atlantic pressure errors for 96–120 h are similar to the central and eastern Pacific. For relatively strong cyclones, the western Atlantic and central/eastern Canada pressure errors are larger than those for the Pacific by 108–120 h. There are large spatial variations in the central pressure biases at 72–120 h, with overdeepened GFS cyclones (negative errors) extending from the northern Pacific and Bering Strait eastward to western Canada, while underdeepened GFS cyclones (positive errors) occur across northeast Canada and just east of the U.S. east coast. GFS cyclone tracks and spatial composites using the daily NCEP reanalysis are used to illustrate flow patterns and source regions for some of the large GFS cyclone errors and biases. Relatively large central pressure errors over the central Pacific early in the forecast (30 h) spread eastward over Canada by 66 h and the eastern United States by 84 h. The underdeepened GFS cyclone errors (>1.5 standard deviations) at day 4 over the western Atlantic are associated with an anomalous ridge over the western United States and trough over the eastern United States, and most of the underdeepening occurs with cyclones tracking east-northeastward across the Gulf Stream. Many of the overdeepened cyclones have tracks more parallel to the U.S. east coast. The underdeepened cyclones over the central and eastern Pacific tend to occur farther south (35°–45°N) than the overdeepened events.

Corresponding author address: Dr. Brian A. Colle, School of Marine and Atmospheric Sciences, Stony Brook University, Stony Brook, NY 11794-5000. E-mail: brian.colle@stonybrook.edu

Abstract

Short- to medium-range (1–5 day) forecasts of extratropical cyclones around North America and its adjacent oceans are verified within the Global Forecast System (GFS) model at the National Centers for Environmental Prediction (NCEP) during the 2002–07 cool seasons (October–March). Cyclones in the immediate lee of the Rockies and U.S. Great Plains have 25%–50% smaller pressure errors than other regions after hour 36. The central pressure and displacement errors are largest over the central and eastern Pacific for the 42–72-h forecast, while the western and central Atlantic pressure errors for 96–120 h are similar to the central and eastern Pacific. For relatively strong cyclones, the western Atlantic and central/eastern Canada pressure errors are larger than those for the Pacific by 108–120 h. There are large spatial variations in the central pressure biases at 72–120 h, with overdeepened GFS cyclones (negative errors) extending from the northern Pacific and Bering Strait eastward to western Canada, while underdeepened GFS cyclones (positive errors) occur across northeast Canada and just east of the U.S. east coast. GFS cyclone tracks and spatial composites using the daily NCEP reanalysis are used to illustrate flow patterns and source regions for some of the large GFS cyclone errors and biases. Relatively large central pressure errors over the central Pacific early in the forecast (30 h) spread eastward over Canada by 66 h and the eastern United States by 84 h. The underdeepened GFS cyclone errors (>1.5 standard deviations) at day 4 over the western Atlantic are associated with an anomalous ridge over the western United States and trough over the eastern United States, and most of the underdeepening occurs with cyclones tracking east-northeastward across the Gulf Stream. Many of the overdeepened cyclones have tracks more parallel to the U.S. east coast. The underdeepened cyclones over the central and eastern Pacific tend to occur farther south (35°–45°N) than the overdeepened events.

Corresponding author address: Dr. Brian A. Colle, School of Marine and Atmospheric Sciences, Stony Brook University, Stony Brook, NY 11794-5000. E-mail: brian.colle@stonybrook.edu

1. Introduction

a. Background

This paper is the third in a series verifying extratropical cyclones within the operational models at the National Centers for Environmental Prediction (NCEP). In Parts I and II of this study, Charles and Colle (2009a,b, hereafter referred to as CC09a and CC09b) highlighted the performance of the North American Mesoscale (NAM), Global Forecast System (GFS), and Short-Range Ensemble Forecast (SREF) models around North America and its adjacent oceans for the 0–60-h forecasts of extratropical cyclones during the 2002–07 cool seasons.

CC09a showed that the GFS analysis had more accuracy for the initialized cyclone pressures than the NAM and the North American Regional Reanalysis (Mesinger et al. 2006), especially over the oceanic regions. In addition, the NCEP GFS had a more accurate cyclone intensity and position for the 0–60-h forecasts than the NAM over the continental United States and adjacent oceans, especially over the eastern Pacific, where the NAM underdeveloped cyclones on average. These results are consistent with other recent GFS and NAM forecast comparisons of sea level pressure around North America (Wedam et al. 2009; McMurdie and Casola 2009). CC09a showed little improvement in the 0–2-day cyclone forecasts during the past 5 yr over the eastern United States, while there has been a relatively large improvement in the cyclone pressure predictions over the eastern Pacific in the NAM.

CC09b also verified the GFS against the NCEP SREF for the 2004–07 cool seasons. The GFS has more accuracy than the SREF mean for cyclone central pressure and position for most forecast times and regions around North America. CC09b also showed that the SREF had slightly greater probabilistic skill than the combined GFS and NAM for central pressure; however, for the higher probabilities of cyclones exceeding a threshold the SREF was overconfident (probabilities too high) in all regions.

The GFS is also run in an ensemble configuration at NCEP using the breeding technique (Toth and Kalnay 1997). Froude et al. (2007) verified extratropical cyclone tracks in the GFS-based ensemble and the European Centre for Medium Range Weather Forecasting (ECMWF) Ensemble Prediction System (EPS) between 6 January 2005 and 5 April 2005. The ECMWF EPS consisted of 50 perturbed members out to 10 days, with a spectral resolution of T255L40, while the 10-member NCEP EPS was run at a resolution of T126L28 (T62L28) for the first 7.5 (last 8.5) days. They found that the ECMWF ensemble had greater skill than the NCEP ensemble for cyclones in the Northern Hemisphere, while the NCEP ensemble had more skill in the Southern Hemisphere. The ECMWF ensemble mean had greater skill than the control member, although the ECMWF ensemble was underdispersed. Froude (2010) compared day 1–7 cyclone predictions of nine global ensemble prediction systems for February–July 2008 and found that the ECMWF mean had the greatest cyclone skill of all models.

The cyclone errors in the operational model have large spatial variability. For example, Froude (2009) showed that ECMWF cyclone position errors in the Northern Hemisphere were largest in the Atlantic Ocean region for 1–7-day forecasts, since storms moved too slowly in this region. Cyclones also tended to be overpredicted (too deep) over the oceans and underpredicted over the land areas. CC09a found that cyclones in the GFS model over the eastern Pacific have larger central pressure absolute errors than other regions around North America at forecast hours 24–60, while the central United States had generally smaller errors.

The sea level pressure errors along the U.S. west coast are frequently larger than those of the U.S. east coast (Wedam et al. 2009). McMurdie and Mass (2004) focused on a select number of cyclone events over the eastern Pacific to illustrate the large errors that can be four to eight cyclone events per year on average with errors greater than 10 hPa at hour 48 in the GFS, and this number has not decreased in recent years.

Several previous studies have shown that model forecast errors can spread eastward from the Pacific and impact the eastern United States (Langland et al. 2002; Elmore et al. 2006; Swanson and Roebber 2008; Sellwood et al. 2008). For example, Elmore et al. (2006) found that the initialization errors of upper-level short-wave troughs along the Pacific Coast persist with the short wave as it crosses the United States. Langland et al. (2002) showed that the poor numerical forecast of the 24–25 January 2000 “surprise” snowstorm may have resulted from errors in the initial conditions over the eastern Pacific on 21 January. For the years 1991–2001, Swanson and Roebber (2008) showed that NCEP and ECMWF analysis errors over the Pacific can significantly impact forecast skill in the medium range (5 days) over the continental United States. Hakim (2003) and Chang (2005) showed that forecast errors over the northern Pacific can spread eastward associated with eastward-propagating wave packets. Sellwood et al. (2008) showed that the impacts of assimilating additional targeted observations over the eastern Pacific can spread downstream over eastern North America by day 3 of the forecast.

The model predictability has been linked to particular large-scale flow regimes. Several studies have shown that higher-amplitude flows are inherently more predictable than low-amplitude or zonal jet flows (Roebber and Tsonis 2005; Nutter et al. 1998; McMurdie and Casola 2009). For example, McMurdie and Casola (2009) found that the largest sea level pressure errors along the U.S. west coast were associated with a ridge axis over the Rocky Mountains and nearly zonal flow over the Pacific. They concluded that the shape of the upper-level flow is more important than the magnitude of the wind for impacting short-term sea level pressure errors.

b. Motivation

CC09a,b focused on the performance of the NCEP GFS cyclones during the 0–2.5-day forecasts. A major goal of The Observing System Research and Predictability Experiment (THORPEX) is to better understand atmospheric predictability from 1 to 12 days (Shapiro and Thorpe 2004). With the exception of the Froude et al. (2007) and Froude (2010) studies, which evaluated the GFS within the context of other ensembles, the cyclones in the GFS for days 3–5 have not been comprehensively verified, especially around North America.

CC09a suggested that there are relatively large biases in cyclone position and strength during the day 0–2 forecasts, which vary spatially around North America. It needs to be determined whether these same biases exist for the GFS medium-range forecasts. CC09a also did not investigate whether GFS cyclone biases were linked to any particular synoptic-scale flow regimes or cyclone tracks. If a particular flow pattern or set of cyclone tracks favor large errors in the GFS, this may help forecasters anticipate a potential predictability problem at least a day or two in advance. Finally, it needs to be determined whether relatively large cyclone errors early in the forecast over the Pacific are associated with relatively large cyclone errors downstream over North America. If there is a connection, a forecaster should be aware of recent short-term (24–48h) model performance over the eastern Pacific before interpreting the longer-range model guidance over the eastern United States. The questions this study addresses are the following:

  • What are the cyclone position and strength errors in the NCEP GFS model for the 0–5-day forecasts, and how do the errors vary spatially from the central Pacific eastward to the central Atlantic?
  • Are relatively large cyclone errors over the Pacific associated with poor cyclone predictability over the eastern United States 3–5 days later?
  • Are cyclone errors in the GFS model associated with coherent large-scale flow patterns and cyclone tracks?
Section 2 of this paper will describe the data and methods used in the analysis. Section 3 will show the GFS central pressure and displacement errors, while section 4 will show how these errors can propagate from the Pacific to the Atlantic. Sections 4b and 4c will composite the large-scale flow associated with some of the systematic biases in the GFS.

2. Data and methods

The GFS was verified for the cool season (October–March) from October 2002 to March 2007 over a large portion of North America and adjacent oceans (Fig. 1). Although this study does not include more recent years (e.g., 2008–09), there is still enough data to test whether the GFS model errors have some coherent flow patterns associated with them. Also, previous studies of operational cyclone errors (Froude 2010) only looked at several months of cyclones (February–July 2008), so our results still include several more cool seasons of data. During the first three cool seasons (2002–05), the GFS grids were available from NCEP’s National Operational Model Archive and Distribution System (NOMADS; information online at http://nomads.ncdc.noaa.gov/data.php) 4 times daily every 6 h at ∼95-km grid spacing. For the 2005–07 period, the GFS was available from NOMADS to hour 120 4 times daily on a 1° grid. The GFS is run at a resolution of T382 before hour 180 (T254 before June 2005), which is roughly equivalent to a horizontal resolution of ∼0.3°. The GFS has undergone several changes since 2002 (see Table 1 of CC09a).

Fig. 1.
Fig. 1.

Geographical regions used for the GFS cyclone verification.

Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222422.1

Cyclones were identified using the approach outlined in CC09a, which involved using a 2-hPa closed-contour threshold and a sea level pressure (SLP) gradient of 1.5 hPa per 1000 km anywhere within 300 km of this pressure minimum. Each forecast cyclone was paired with the closest observed cyclone (not to exceed 800-km separation). More details on this automated cyclone identification and pairing with observations can be found in CC09a. To get a GFS cyclone track, first, a mean-depth steering flow was calculated using the winds averaged from 850 to 500 hPa at a particular GFS forecast time. Then, the existing cyclone position is advected downstream according to this flow for 6 h to guess a new cyclone position to compare with the GFS forecast 6 h in the future. The closest cyclone to this guess position completes the track.

The GFS analysis was used to verify the forecasts given its greater accuracy for cyclones as compared to the NAM analysis and the North American Regional Reanalysis (NARR) (CC09a). This resulted in at least 300 cyclones per forecast hour for 2002–07 in each of the eight regions shown in Fig. 1; however, to increase statistical significance of some of the results below, some of the forecast hours and regions were combined. To test for statistical significance, a bootstrapping approach was used to resample the data and obtain 90% confidence intervals around the means as described in CC09a.

For certain combinations of cyclone error events (e.g., large error days), the spatial errors across a 1° grid were also calculated by subtracting the forecast pressure or geopotential height with the corresponding GFS analysis at each grid point. Composite cyclone central pressure absolute errors were normalized against the climatological error (Pclim) by averaging the absolute difference between the forecast (Pft) and observed cyclone (Pot) central pressure over all [m in Eq. (1)] cyclones during the entire 2002–07 period for a particular grid point. The Pclim was subtracted from the average absolute difference between the forecast (Pfc) and observed (Poc) cyclone central pressure for the select [number of cases is n in Eq. (2)] composite cases [Pcomp in Eq. (2)] to create the normalized error [E in Eq. (3)]:
e1
e2
e3

To determine the large-scale flow associated with particular cyclone errors in a region, composite averages of several variables were completed using the daily NCEP–National Center for Atmospheric Research (NCAR) reanalysis at 2° resolution (Kalnay et al. 1996). The closest reanalysis time to the time of the cyclone error was used in the composite analysis.

3. Results

a. GFS regional errors

1) Cyclone central pressure

Figure 2 shows the mean absolute errors (MAEs) of cyclone central pressure averaged every 12 h for seven of the eight regions in Fig. 1. Because of sea level pressure reduction difficulties over steep terrain, the results over the western United States (W) are not presented here, but can be found in Charles (2008). During the first 72 h, the eastern and central Pacific (regions EP and CP) errors increase more rapidly than the other regions. By hour 72, regions EP and CP errors (4–4.5 hPa) are 1–2 hPa larger than other regions on average. Between 72–120 h, the western (WA) and central Atlantic (CA) errors increase more rapidly than the other regions, and become as large as the central Pacific errors by 108–120 h. The eastern United States (region E) and central/eastern Canada (EC) errors are smaller than the oceanic regions by 120 h, while the central United States (region C) has the smallest errors at all forecast times.

Fig. 2.
Fig. 2.

(a) Cyclone mean absolute errors vs forecast hour for the 2002–07 cool seasons for the GFS regions specified in Fig. 1. (b) As in (a), but for the relatively deep cyclones exceeding the thresholds in Table 1. Confidence intervals at the 90% significance level are given by the vertical bars.

Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222422.1

The central pressure errors for hours 72–120 were put on a 2° × 2° grid and the error was shaded for grid boxes with at least five cyclones verified during the 2002–07 cool seasons (Fig. 3a). The average MAEs between 72 and 120 h are largest (6–8 hPa) along the oceanic storm track across the north-central Pacific (35°–50°N) and from the east coast of North America to the northern Atlantic (Fig. 3a). Other areas of relatively large MAEs are over Alaska, western Canada, and just north of Hudson Bay. The smallest errors (2–3 hPa) are across the northern plains and the central part of Canada.

Fig. 3.
Fig. 3.

Cyclone central pressure (a) MAEs (shaded in hPa) and (b) mean errors (shaded in hPa) averaged for the 72–120-h GFS forecasts during the 2002–07 cool seasons.

Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222422.1

Figure 3b shows the spatial distribution of cyclone mean errors between hours 72 and 120. A positive error corresponds to the forecast cyclones that are too weak and a negative error corresponds to the forecast cyclones that are too intense. The forecast cyclones during 72–120 h tend to be too weak on average by 2–3 hPa at many points within the entrance of the storm track south of 50°N and between 175°E and 175°W as well as near the U.S. east coast (80°–60°W) (Fig. 3b), and over much of eastern Canada. In contrast, GFS cyclones tend to be too deep (1–3 hPa) on average over the northern Pacific, Alaska, western Canada, and parts of the northern Atlantic around Greenland. The pressure bias over eastern Canada for 72–120 h is nearly as large as the first 72 h of the forecast (not shown), while the bias increases between 24 and 72 h in the other regions (not shown).

If only the relatively deep cyclones are verified (Fig. 2b), which are greater than one standard deviation deeper in central pressure than the mean for each region (Table 1), the central pressure errors in the WA and EC regions increase more rapidly than the other regions after day 3, and these two regions have larger MAEs (5.5–6.5 hPa) than other regions by 120 h. The MAEs for deep cyclones in the WA and EC regions are about as large as for all cyclone events (Fig. 2a). In contrast, the MAEs for the relatively deep cyclones over the CA and CP regions are 1–2 hPa less than all cyclones, since errors for the deep cyclones only increase slowly after 72 h. This suggests that relatively deep cyclones in the medium range (72–120 h) are less predictable in the entrance of the storm track (e.g., WA); thus, the growth of deep cyclones is difficult to predict.

Table 1.

The deep cyclone thresholds (hPa) for the regions in Fig. 1.

Table 1.

2) Cyclone displacement

Figure 4 shows the displacement errors versus forecast hour for all GFS cyclones within the regions in Fig. 1. The cyclone displacement error is the distance between the forecast position and the observed cyclone. During the 18–72-h period, the cyclone displacement errors for all regions fall within a 30–60-km (15%–20%) window. During 18–36 h, the central Great Plains area (region C) has the largest displacement errors, while from 42 to 72 h the CP and EP regions have the largest errors and the CA region the smallest. By 96–120 h, the EP and WA regions have slightly larger errors than EC, while the CA and central U.S. (C) regions have the smallest errors. The variation in displacement errors between the regions does not change for the deeper cyclone events (not shown).

Fig. 4.
Fig. 4.

As in Fig. 2a, but for cyclone displacement (km).

Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222422.1

To quantify the distribution of displacement errors, Fig. 5 shows the position-error histograms of GFS cyclone displacement over the WA, C, and EP regions. Cyclones are slightly shifted to the north in the EP for 108–120 h (Fig. 5a), while there is little bias in the 36–48-h forecast (Fig. 5b). Over the central United States (C region), the GFS cyclones tend to be too far to the north (Figs. 5c and 5d), with a mean vector error of 50–60 km from the observed cyclone. Cyclones displaced to the north have mean errors of 400 and 220 km for the 102–120- and 36–48-h forecasts, respectively, which are ∼20% greater than other octants. Previous operational models, such as the Nested Grid Model, and earlier versions of the GFS (i.e., the Aviation Model), also had this problem (Mullen and Smith 1990; Smith and Mullen 1993). This has been attributed to the relatively smooth model terrain; thus, the lee troughs are underdeveloped and the resulting cyclogenesis occurs too far to the north. Cyclones are also slightly shifted too far north in the EP for 108–120 h, but there is little bias in the 36–48-h forecast (not shown).

Fig. 5.
Fig. 5.

Histograms showing the frequency of model cyclone positions relative to the observed position (center point) for 45° bins centered on N, NE, E, etc. in the GFS for hours (a),(c),(e) 102–120 and (b),(d),(f) 36–48 for regions (a),(b) EP, (c),(d) C, and (e),(f) WA. The dashed range circles are every 10%, from 0 to 30. The solid range circles represent the 90% confidence intervals below and above the range of the bins. The numbers for each radial represent the average displacement error (km) within that directional bin. The black vectors indicate the mean displacement vector, with each dashed range ring every 50 km.

Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222422.1

At 102–120 h over the WA region (Fig. 5e), the most common displacement bias (18%) is to the northeast of the observed cyclone, and there is also a secondary peak (15%) to the southwest. Cyclones in these two octants have mean displacement errors (450–475 km), which are 50–100 km greater than the other directions. As the lead time decreases to 36–48 h over the WA (Fig. 5f), the GFS displacement bias shifts more to the west and southwest, with a mean error of ∼200 km on average. Since a majority of cyclones move from southwest to northeast over the WA storm track, this suggests that these cyclones move too fast in the GFS at longer lead times, and there is more of a tendency for WA cyclones to be too slow and too far west for shorter forecast periods.

b. West-to-east error evolution

Another goal of this paper was to obtain some insights into the upstream and downstream evolution of the cyclone errors across North America and its adjacent oceans. This was accomplished by showing the large-scale flow patterns and normalized pressure errors (outlined in section 2) associated with certain cyclone errors in specific regions. Figures 6a–e show the evolution of the average SLP and normalized SLP MAEs for 28 cyclones in the CP region with cyclone central pressure MAEs at 30 h more than 1.5 standard deviations greater than the mean cyclone error in this region (>6.2-hPa error threshold). Hour 30 errors were used over the CP, since these errors early in the forecast can spread downstream and impact the East Coast errors by days 3–4. At 30 h (Fig. 6a), there are relatively large SLP errors (1–2 hPa) around the south coast of Alaska and northern Pacific in association with an average 994-hPa cyclone centered over the Aleutians. During the 48–66-h forecast period (Figs. 6b and 6c), these errors spread northward and eastward over Alaska and into central Canada, with some of these errors moving quickly to the east around the 500-hPa trough centered over western Canada and eastern North America. By 84 h (Fig. 6d), the Canadian SLP errors spread east and south over the Great Lakes region and the northeast United States as the upper-level trough amplifies along the east coast. However, the errors do not amplify further by hour 102 over the eastern United States, and there is no well-defined composite surface cyclone at this time along the U.S. east coast. If only the smallest cyclone error events (<0.25 std dev) are chosen over the CP at hour 30 (Fig. 6f), there is no apparent downstream movement of errors for 48–96 h across North America (Figs. 6g–i). Overall, large cyclone errors early in the forecast over the Pacific do impact the medium-range surface pressure predictions over the eastern United States, but they may not necessarily be tied to a well-defined east coast cyclone event.

Fig. 6.
Fig. 6.

GFS mean SLP error normalized against the climatological error (shaded every 0.25 hPa), mean observed 500-hPa heights (dash every 60 m), and mean observed SLPs (solid every 2 hPa), averaged at each grid point for all CP cyclones with central pressure absolute errors more than 1.5 std dev greater than the mean (28 cases) for hours (a) 30, (b) 48, (c) 66, (d) 84, and (e) 102 h. (f)–(j) As in (a)–(e), but for the small error cyclone events at hour 30 over the CP (errors <0.25 std dev).

Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222422.1

The reverse approach was also attempted, in which composites were created for 25 eastern U.S. (region E) cyclone events with 102-h cyclone central pressure MAEs ∼1.5 standard deviations greater than the mean central pressure error for this region (>10.3-hPa error threshold). We chose the 102-h forecast period because this is when the errors are centered over the northeast United States for the cases with large errors over the CP at hour 30. At 30 h (Figs. 7a and 7b), the SLP errors are within ∼0.25 hPa of the climatological error over much of the Pacific and North America. Meanwhile, at 500 hPa there are some geopotential height errors 5–15 m greater than climatology over the central Pacific and near the west coast of Canada, where a mean ridge was located; however, these errors are not statistically significant at this time (not shown). By hour 48 (Figs. 7c and 7d), the errors are much larger than climatology over the northern and central Pacific associated with a surface cyclone centered over the Aleutians (significant at the 90% level). Meanwhile, a weak short-wave trough associated with near-climatological errors is located over the Intermountain West. By 84 h (Figs. 7e and 7f), this short-wave trough at 500 hPa moves eastward to the Great Lakes region and develops large errors relative to climatology at 500 hPa and at the surface (significant at the 95% level), where a mean surface cyclone (1010 hPa) is developing over the eastern Great Lakes. Meanwhile, from 30 to 84 h the errors over the northern Pacific increase by a factor of 2–4 at the surface and 500 hPa as the mean cyclone over the Gulf of Alaska deepens by a few hectopascals. At the time of the largest error over the eastern United States (102 h; see Figs. 7e and 7f), there are large errors over the northern Pacific, a downstream ridge over northwest Canada, and a trough over the eastern United States. However, unlike the previous example with large errors over the CP at hour 30 that subsequently spread eastward, the large errors over the eastern United States at hour 102 do not show clear downstream spreading from the Pacific. Rather, the Pacific and eastern U.S. errors amplify more simultaneously, thus suggesting more local and rapid error growth later in the forecast related to baroclinic instability within the Pacific and eastern U.S. storm tracks.

Fig. 7.
Fig. 7.

(left) GFS 500-hPa geopotential height errors normalized against the climatological error (shaded every 5 m) and observed 500-hPa heights (dashed every 60 m) at hours (a) 30, (c) 48, (e) 66, (g) 84, and (i) 102 for relatively large (>1.5 std dev) central pressure errors at 102 h within the E region in Fig. 1. (right) GFS mean SLP error normalized against the climatological error (shaded every 0.25 hPa), mean observed 500-hPa heights (dash every 60 m), and mean observed SLPs (solid every 2 hPa), averaged at each grid point for relatively large (>1.5 std dev) central pressure errors at 102 h within the E region at hours (b) 30, (d) 48, (f) 66, (h) 84, and (j) 102.

Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222422.1

The next two sections will focus on the synoptic flow patterns that favor relatively large biases in cyclone strength over the western Atlantic and central Pacific regions. These two regions were chosen since they are more within the active storm tracks than the other regions.

c. Western Atlantic errors

The cyclone tracks for the GFS forecasts were diagnosed for significant (>1.5 std dev) central pressure errors at hour 96 over the WA. Cyclone tracks and errors are not shown after 96 h, since there are not enough events to be statistically significant. Figure 8 shows the GFS cyclone forecast tracks plotted every 6 h associated with relatively large positive (underdeepen) and negative (overdeepen) central pressure errors at 96 h. The gray line segments indicate periods when the forecast cyclone is filling faster than observed or deepening slower than observed (i.e., under prediction of cyclone strength), while the black lines indicate 6-h periods when the forecast cyclone is deepening faster than observed or filling slower than observed (i.e., over prediction of cyclone strength).

Fig. 8.
Fig. 8.

Forecast cyclone tracks for WA region GFS cyclones with large (>1.5 std dev) central pressure errors for cases with (a) positive and (b) negative central pressure errors at hour 96 that are more than 1.5 std dev above (below) the mean error of all cyclones with positive (negative) error. The gray line segments indicate 6-h periods when the forecast cyclone growth rate (weakening) is slower (faster) than observed. Meanwhile, black lines indicate 6-h periods when the forecast cyclone is deepening (or filling) faster (slower) than observed. The black × symbols indicate the locations of the forecast cyclones at hour 96.

Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222422.1

For the large positive cyclone errors at 96 h, a majority of GFS cyclones track east and northeast along the eastern edge of the Gulf Stream (Fig. 8a). The positive errors for most cyclones begin as the cyclone approaches the immediate coast and continue increasing over the Atlantic. A few of the cyclones originate over southern Canada and move southeastward across the Great Lakes. In contrast, many GFS cyclones with negative errors track northeastward closer to the coast than positive error events (Fig. 8b). The mean cyclone displacement errors are different for the large positive and negative error cyclones. The negative events are displaced 230 km to NNE (30°) of the observed position on average (not shown), while the large positive error events are displaced to the south of observed by 175 km on average.

Given the relatively different set of tracks for the GFS positive and negative pressure errors at hour 96 over the western Atlantic, it was hypothesized that the large-scale flow regimes are different between these two sets of cyclone errors. A spatial composite was constructed using the daily NCEP–NCAR global reanalysis for the 24 overdeepened and 24 underdeepened GFS cyclones over the western Atlantic at hour 96 (Tables 2 and 3). The mean pressure error for the over- and underdeepened events are −10.3 and 11.8 hPa, respectively, and there are several events with >±14-hPa errors. The climatology for each day (1968–96 average) was used to calculate the geopotential height anomalies. Significance was calculated by selecting several points within an anomaly and testing using a bootstrap method as discussed above in section 2. Only those anomalies that are significant at least at the 90% level are discussed below.

Table 2.

The initialization time and date, observed cyclone pressure, and GFS forecast error (hPa) at forecast hour 96 for the largest (>1.5 std dev) negative (overdeepening) cyclone errors over the WA.

Table 2.
Table 3.

As in Table 2, but for the positive (underdeepened) cyclone errors over the WA at hour 96.

Table 3.

Two days before the initialization of the 96-h forecast (day 2), both the positive and negative error composites have a mean 500-hPa ridge near the west coast of North America and a 500-hPa trough over eastern North America (Figs. 9a and 9b). The 500-hPa ridge and trough amplitudes are near climatology in the negative error events (Fig. 10a), while there are positive 50–60-m height anomalies extending southward from the Gulf of Alaska and 70–80-m positive anomalies to the southwest of Greenland. In contrast, the positive error events have 60–90-m positive height anomalies over the northeast Pacific eastward to the northern plains (Fig. 10b), and there are 30–50-m negative anomalies with the 500-hPa trough from north of Hudson Bay southeastward to the northwest Atlantic.

Fig. 9.
Fig. 9.

Daily NCEP–NCAR reanalysis composite showing 500-hPa geopotential heights (color shaded and contoured every 60 m) for the relatively large (1.5 std dev) negative cyclone pressure error events in the GFS at hour 96 at (a) day −2, (c) day 0 (initialization), and (e) day +4 (hour 96). (b),(d),(f) As in (a),(c),(e) but for the positive cyclone pressure GFS error events. Tables 2 and 3 list the negative and positive error events, respectively.

Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222422.1

Fig. 10.
Fig. 10.

As in Fig. 9, but for the 500-hPa geopotential height anomalies (color shaded and contoured every 10 m).

Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222422.1

At the time of GFS initialization of the 96-h large-error forecast (Figs. 9c,d and 10c,d), the 500-hPa ridge and associated positive height anomalies near Greenland in the positive error cases are more amplified (80–110 m) than the negative error events in response to the more well-defined upstream trough (30–60-m anomaly) near Nova Scotia in the positive events. Meanwhile, the maximum positive height anomalies (40–50 m) with the 500-hPa ridge shift eastward to the Pacific Northwest. In contrast, for the negative error events the ridge flattens and a weak negative 500-hPa height anomaly develops over this same Pacific Northwest region.

For the positive cyclone error events at 96 h (Fig. 9f), a well-defined series of height anomalies at 500 hPa exists from the central Pacific to the northern Atlantic. More specifically, there is a positive height anomaly (70–80 m) over the western United States, a negative anomaly (50–60 m) over the western Atlantic, and a large positive anomaly over Greenland. In contrast, negative error events have more zonal flow impacting the western United States and a 20–50-m negative anomaly over the western Atlantic, and a positive height anomaly (80–100 m) over southeast Canada. Overall, these composite results suggest that many underdeepened cyclones over the western Atlantic are related to a high-amplitude wave pattern that extends eastward from the eastern Pacific to the Atlantic. Meanwhile, the negative error events have a less amplified trough over eastern North America.

To illustrate in more detail the evolution of these relatively large positive and negative GFS error events at hour 96, Figs. 11 and 12 show the 500-hPa geopotential height evolution for two representative events. For an underdeepened GFS cyclone event initialized at 1800 UTC 11 January 2004, there is a short-wave trough over the upper Midwest associated with an Alberta clipper cyclone moving southeastward downstream of a high-amplitude ridge over western North America and deep trough over eastern Canada at hour 66 (1200 UTC 14 January 2004; see Fig. 11a). At this time the cyclone central pressure error is only ∼1 hPa, but the GFS is 3–6 h slow with the short-wave trough and surface cyclone. By 84 h (0600 UTC 15 January; see Fig. 11b), the observed cyclone deepens ∼2 hPa, while the GFS cyclone weakens by ∼1 hPa (∼4 hPa error). At 96 h (1800 UTC 15 January; see Fig. 11c), the observed cyclone (983 hPa) explosively deepens 20 hPa in 12 h as it tracks eastward over the Gulf Stream, while the GFS cyclone lagging 400 km to the west only deepened ∼10 hPa (∼14-hPa error). The GFS geopotential height errors at 500 hPa along the East Coast are also positive around 80–120 m at this time. By 102 h (0000 UTC 16 January; see Fig. 11d), the GFS cyclone error increases to ∼19 hPa. Even as the GFS cyclone moves over the Gulf Stream near 40°N, 60°W by 108 h (not shown), it still has an 18-hPa central pressure error.

Fig. 11.
Fig. 11.

The 500-hPa geopotential height evolution for the GFS (solid every 60 m) and observed (shaded every 50 m) for the GFS run initialized at 1800 UTC 11 Jan 2004 for hours (a) 66, (b) 84, (c) 96, and (d) 102. The locations of the observed and GFS cyclones are given by the × and + symbols, respectively, with the top number the observed and the bottom the GFS.

Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222422.1

Fig. 12.
Fig. 12.

As in Fig. 11, but for hours (a) 60, (b) 84, and (c) 96 for the GFS run initialized at 0000 UTC 4 Dec 2003.

Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222422.1

In contrast, Fig. 12 shows a representative case of an overdeepened cyclone over the western Atlantic at 96 h that was initialized by the GFS at 0000 UTC 8 December 2003. At 60 h (1200 UTC 10 December 2003; see Fig. 12a), there is an amplified trough along the U.S. east coast at 500 hPa, and a shorter-wavelength (more progressive) ridge upstream than the underdeveloped event. At this time the GFS cyclone (1006 hPa) is 2 hPa weaker than observed, but the GFS cyclone is over a favorable area of development over the Gulf Stream east of the South Carolina coast. During the next 24 h (1200 UTC 11 December 2003; 84 h), the GFS cyclone deepens 21 hPa (Fig. 12b), while the observed cyclone ∼300 km to the northwest deepens 14 hPa (5-hPa error). Meanwhile, the ridge at 500 hPa to the west of the coastal cyclone extends from the Midwest to eastern Canada, with the GFS ridge more pronounced than observed. By 96 h (1000 UTC 12 December 2003; see Fig. 12c), the GFS cyclone is 13 hPa deeper than observed and the trough at all levels is displaced too far to the east.

d. Central Pacific errors

It was hypothesized that the GFS errors over the Pacific may also have some coherent large-scale flow differences between the positive and negative central pressure errors. For example, GFS cyclone tracks centered at hour 72 over the eastern Pacific are clustered differently between the relatively large (1.5 std dev or ∼7.5 hPa) underdeepened and overdeepened cases (Fig. 13). The negative error events (overdeepened) tend to have tracks from the eastern Pacific into the Gulf of Alaska. Perusal of many of these events revealed that they are mature cyclones that the GFS overdeepens and/or the cyclolysis in the model is not fast enough. In contrast, nearly half of the underdeepened (positive error) cyclone events (X positions) are south of 45°N, as compared to only five events in the overdeepened tracks. Several of these positive error events track toward the U.S. West Coast. A similar track distribution occurred over the central Pacific (not shown), with many more negative error cases tracking northward toward the Aleutians.

Fig. 13.
Fig. 13.

As in Fig. 8, but for those GFS cyclones within the EP region with 1.5 standard deviation (a) negative and (b) positive central pressure errors at hour 72.

Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222422.1

Figure 14 shows a time series of cyclone central pressure errors at hour 48 over the central Pacific (CP region). We chose to look at the evolution of these short-term (48 h) errors over the Pacific, since as shown in Fig. 2 and Wedam et al. (2009), the Pacific errors grow large early in the forecast (42–72 h) given the fairly data-sparse Pacific. Also, to look for possible flow regimes attached to cyclone errors, a short-term (48 h) forecast is needed for a more continuous record of simulated cyclones matched with observations. The annual frequency of relatively large (2 std dev) error events remained similar from 2002 to 2007, with many years having 8–12 events of greater than 10-hPa error (2 std dev) at hour 48 h. There are periods with relatively small cyclone errors (nearly all <6 hPa), such as October 2003 and January 2005, and other periods with persistent large errors (several with >8 hPa), such as 20 November–20 December 2005. A comparison of the monthly composites of the 500-hPa geopotential heights for January 2005 and December 2005 shows a deeper trough over the western and central Pacific in December 2005 than January 2005 (Figs. 15a and 15b). January 2005 has a more of a split flow over the eastern Pacific and a weaker jet in the western Pacific than does December 2005.There is a larger area of negative height anomalies (>100 m) relative to climatology within the storm track over the western and central Pacific in December 2005 than in January 2005 (Figs. 15c and 15d), as well as more amplified downstream ridge over western Canada and a negative height anomaly over the eastern United States. Thus, the larger-error GFS events in December 2005 were associated with an eastward shift in the storm track over the Pacific that extended eastward across North America. The October 2003 period was similar to January 2005 (not shown), with no anomalous negative heights over the western Pacific.

Fig. 14.
Fig. 14.

Time series of GFS central pressure errors (hPa) at hour 48 for the CP region during the 2002–07 cool seasons. Each bar represents the average central pressure error for all cyclones present in the CP region at hour 48. Horizontal black dashed lines denote 2 std dev above and below the mean central pressure error. The boldface black line is the 5-day running mean of the cyclone central pressure errors. The vertical gray lines represent the warm season break between March and October.

Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222422.1

Fig. 15.
Fig. 15.

Monthly composite of the NCEP–NCAR reanalysis showing 500-hPa geopotential heights (contoured every 60 m) for (a) January and (b) December 2005. (c) As in (a), but for the 500-hPa geopotential height anomalies (every 20 m). (d) As in (c), but for December 2005.

Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222422.1

The time series over the central Pacific at hour 48 in Fig. 14 also reveals that there are many periods of negative and positive central pressure errors that last at least a few days, as suggested by the oscillatory nature of the 5-day running mean. Some periods are quite persistent, such as the negative errors in October 2006, December 2005, and October 2005, while there are favored positive error periods in early to mid-March 2003, late March 2007, and mid-January 2004. To determine the large-scale flow regimes attached to these positive and negative errors in the 48-h GFS forecast, a spatial composite of events was constructed using the following criteria. For the negative events, over at least a 2.5-day period (2.5 days of GFS forecasts made every 6 h valid at hour 48) at least 80% of the cyclones over the CP region had negative errors, with no more than three consecutive events of positive errors. This resulted in 35 events, with the composite event defined as the first time that meets the above criteria. Only one event is counted if it lasts more than 60 h, and there has to be at least 2 days between events. The positive error periods used the same criteria, but only a 2-day period was considered (28 events), since these periods tended to be more short lived and a reasonable sample was needed for a composite. Since there are 30% fewer cases and larger gaps in the time series for the EP than the CP at hour 48, this same approach was not applied to the EP region.

Figure 16 shows the daily composite using the NCEP–NCAR reanalysis dataset centered (day 0) around the GFS initialization time that started the period of persistent 48-h negative or positive errors. At the time of initialization at 500 hPa (Figs. 16a and 16b), both the positive and negative events have a trough along the east coast of Asia and a ridge along the west coast of North America. By day 2 (Figs. 16c and 16d), the 500-hPa trough in the negative events extends eastward to 160°W in the negative events, while a ridge develops in the positive error periods over the north-central Pacific and a weak trough is located just north of Hawaii. As a result, there is a 40–80-m negative height anomaly at 500 hPa centered around the western Aleutians in the negative error events (Fig. 16e), while the positive error events have a 80–120-m positive height anomaly over this same region and a 40–60-m negative height anomaly centered over the central Pacific at 32°N, 170°W (Fig. 16f). Meanwhile, at the surface on day 2 (Fig. 16g), the negative error periods have a 990-hPa cyclone at the western Aleutians, which is 8–10 hPa deeper than climatology (not shown). In contrast, the sea level pressures for the positive error periods at day 2 are 15–18 hPa larger than the negative error events over the northern Pacific (Fig. 16h), which is a 4–6-hPa positive SLP anomaly (not shown). To the south in the positive error periods, there is a weak surface trough extending southward of 40°N along 170°W (Fig. 16h), which is a 4–6-hPa negative anomaly compared to climatology (not shown).

Fig. 16.
Fig. 16.

Composite of daily NCEP–NCAR reanalysis showing 500-hPa geopotential heights (shaded and contoured every 60 m) for persistent (see text for details) negative cyclone pressure errors at hour 48 in the CP region at (a) the time of initialization (day 0) and (c) hour 48 (day 2). (b),(d) As in (a),(c), but for positive GFS cyclone error events. (e),(f) As in (c),(d), but for the 500-hPa geopotential height anomalies (shaded and contoured every 10 m). (g),(h) As in (c),(d), but for the SLP (shaded and every 2 hPa).

Citation: Weather and Forecasting 26, 2; 10.1175/2010WAF2222422.1

Overall, the negative error periods have an active northern storm track into the Aleutians, while the positive errors have more split flow, with weaker northern storms and more cyclones at lower latitudes. This is consistent with the negative and positive error cyclone tracks over the eastern Pacific, which showed many positive error events tracking farther south toward the West Coast. Perusal of many events suggests the positive errors are developing systems that are relatively small in scale as they traverse at a lower latitude, as compared to the larger cyclones wrapped up over the northern Pacific. These smaller systems may be more difficult to initialize and predict in the model. The average of many of these fast-moving southern systems over many different locations in the positive error events results in the weak trough in the composite average.

4. Summary and conclusions

The 5-day forecasts of extratropical cyclones around North America and its adjacent oceans are evaluated within the NCEP Global Forecast System (GFS) model during the 2002–07 cool seasons (October–March). Cyclone tracks and spatial composites using the daily NCEP–NCAR reanalysis are also used to illustrate flow patterns and source regions for some of the large GFS cyclone errors and biases.

Cyclone central pressures in the immediate lee of the Rockies and U.S. Great Plains have 20%–30% smaller pressure errors than other regions at >36 h. The central pressure and displacement errors are largest over the central and eastern Pacific for the 42–90-h forecast, which is likely the result of the model initialization over the relatively data-sparse Pacific. The number of relatively large errors (>∼10 hPa) over the central Pacific at day 2 did not decrease from 2002 to 2007 even with improvements to the GFS model. By 108–120 h, the western and central Atlantic errors are more comparable to the Pacific regions. For the relatively strong cyclones, the western Atlantic central pressure errors are larger than the Pacific by 108–120 h. Thus, it was hypothesized that errors from the Pacific may degrade the cyclone forecasts over the eastern United States and western Atlantic in the medium range. The cyclone errors over the north-central Pacific early in the forecast (hour 30) spread eastward into Canada, and then southward into the Great Lakes and northeast U.S. regions. However, for the largest medium-range (day 4) cyclone errors over the eastern United States, much of this error grows locally within the developing baroclinic wave during days 2–3. Some of this cyclone error growth may be from diabatic precipitation processes, such as those highlighted by the cyclone predictability results using idealized simulations (Zhang et al. 2003), and numerical experiments using different convective parameterizations (Mahoney and Lackmann 2006). Future work needs to determine how much of this cyclone error growth in the GFS is from upstream initial conditions versus uncertainties in model physical parameterizations.

There are large spatial variations in the central pressure absolute errors and biases at 72–120 h. The GFS cyclone positions to the east of the Rockies are too far to the north. This error has been in the GFS for many years, and it may be related to the GFS’s relatively smooth topography. Future work should run the GFS at higher resolution and steeper topography to see if this problem can be mitigated. The largest pressure errors are within the Pacific and Atlantic storm tracks. The GFS tends to overdeepen cyclones (negative errors) from the northern Pacific and Bering Strait eastward to western Canada, while underdeepened GFS cyclones (positive errors) tend to occur more across northeast Canada and just east of the U.S. east coast. Composites suggest that frequent negative errors over a 2–3-day period over the central Pacific occur during an active western Pacific storm track and these cyclones track northeastward over the northern Pacific. Many of the underdeepened cyclones over the Pacific are associated with those storms farther to the south at 35°–40°N.

Many of the underdeepened (positive error) GFS cyclones for the day 4 forecast are located over the western Atlantic just east of the Gulf Stream. They are associated with an amplifying ridge over the western United States and a large-scale trough over the eastern United States. Many of these events involve a short wave within the northern branch of the flow tracking southward out of Canada to the mid-Atlantic, with a cyclone developing along the coast and tracking east-northeastward over the Gulf Stream. There is a well-defined trough–ridge pattern at 500 hPa during these positive error events, suggesting they occur within the Rossby wave packet. The fact that many of these underpredicted cyclone errors occur as they cross the Gulf Stream suggests there may also be too little surface heat or moisture fluxes within the GFS, or too little convective heating as cyclogenesis occurs. In contrast, the overdeepened GFS events over the western Atlantic at 96 h track more northeastward along the U.S. east coast.

Although only the deterministic GFS was verified in this study, the results do have implications for the GFS ensemble run every 6 h at NCEP. Since there are fairly large biases in cyclone strength and position, the observed cyclone may fall outside the ensemble envelope or the ensemble may be underdispersed as a result of GFS systematic errors. Future studies will look more carefully at the model physics (moist and planetary boundary layer processes) for some representative events with cyclone biases. Also, the ability of the GFS to accurately predict Rossby wave packets from the Pacific to the Atlantic will be evaluated.

Acknowledgments

The research was supported by UCAR-COMET (Grant S07-66814) and NOAA-CSTAR (NA10NWS4680003). We appreciate feedback and constructive comments by Dr. David Novak and the three anonymous reviewers. We thank Evan Goldaper from Smithtown High School for his preliminary analysis of the composite flow patterns associated with Pacific cyclone GFS errors as part of his semifinalist Intel (2009) project. Some of the composite images were provided by the NOAA/ESRL/Physical Sciences Division, Boulder, Colorado, on their Web site (http://www.esrl.noaa.gov/psd/).

REFERENCES

  • Chang, E. K. M., 2005: The impact of wave packets propagating across Asia on Pacific cyclone development. Mon. Wea. Rev., 133, 19982015.

    • Search Google Scholar
    • Export Citation
  • Charles, M., 2008: Verification of extratropical cyclones within NCEP forecast models using an automated cyclone tracking algorithm. M.S. thesis, School of Marine and Atmospheric Sciences, Stony Brook University, 135 pp. [Available from MSRC, Stony Brook University, Stony Brook, NY 11794-5000.]

    • Search Google Scholar
    • Export Citation
  • Charles, M., , and Colle B. A. , 2009a: Verification of extratropical cyclones within the NCEP operational models. Part I: Analysis errors and short-term NAM and GFS forecasts. Wea. Forecasting, 24, 11731190.

    • Search Google Scholar
    • Export Citation
  • Charles, M., , and Colle B. A. , 2009b: Verification of extratropical cyclones within the NCEP operational models. Part II: The short-range ensemble forecast system. Wea. Forecasting, 24, 11911214.

    • Search Google Scholar
    • Export Citation
  • Elmore, K. L., , Schultz D. M. , , and Baldwin M. E. , 2006: The behavior of synoptic-scale errors in the Eta Model. Mon. Wea. Rev., 134, 33553366.

    • Search Google Scholar
    • Export Citation
  • Froude, L. S. R., 2009: Regional differences in the prediction of extratropical cyclones by the ECMWF Ensemble Prediction System. Mon. Wea. Rev., 137, 893911.

    • Search Google Scholar
    • Export Citation
  • Froude, L. S. R., 2010: TIGGE: Comparison of the prediction of Northern Hemisphere extratropical cyclones by different ensemble prediction systems. Wea. Forecasting, 25, 819836.

    • Search Google Scholar
    • Export Citation
  • Froude, L. S. R., , Bengtsson L. , , and Hodges K. I. , 2007: The prediction of extratropical storm tracks by the ECMWF and NCEP Ensemble Prediction Systems. Mon. Wea. Rev., 135, 25452567.

    • Search Google Scholar
    • Export Citation
  • Hakim, G. J., 2003: Developing wave packets in the North Pacific storm track. Mon. Wea. Rev., 131, 28242837.

  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437471.

  • Langland, R. H., , Shapiro M. A. , , and Gelaro R. , 2002: Initial condition sensitivity and error growth in forecasts of the 25 January 2000 East Coast snowstorm. Mon. Wea. Rev., 130, 957974.

    • Search Google Scholar
    • Export Citation
  • Mahoney, K. M., , and Lackmann G. M. , 2006: The sensitivity of numerical forecasts to convective parameterization: A case study of the 17 February 2004 East Coast cyclone. Wea. Forecasting, 21, 465488.

    • Search Google Scholar
    • Export Citation
  • McMurdie, L. A., , and Mass C. , 2004: Major numerical forecast failures over the northeast Pacific. Wea. Forecasting, 19, 338356.

  • McMurdie, L. A., , and Casola J. H. , 2009: Weather regimes and forecast errors in the Pacific Northwest. Wea. Forecasting, 24, 829842.

  • Mesinger, F., and Coauthors, 2006: North American Regional Reanalysis. Bull. Amer. Meteor. Soc., 87, 343360.

  • Mullen, S. L., , and Smith B. B. , 1990: An analysis of sea-level cyclone errors in NMC’s Nested Grid Model (NGM) during the 1987–88 winter season. Wea. Forecasting, 5, 433447.

    • Search Google Scholar
    • Export Citation
  • Nutter, P. A., , Mullen S. L. , , and Baumhefner D. P. , 1998: The impact of initial condition uncertainty on numerical simulations of blocking. Mon. Wea. Rev., 126, 24822502.

    • Search Google Scholar
    • Export Citation
  • Roebber, P. J., , and Tsonis A. A. , 2005: A method to improve prediction of atmospheric flow transitions. J. Atmos. Sci., 62, 38183824.

    • Search Google Scholar
    • Export Citation
  • Sellwood, K. J., , Majumdar S. J. , , Mapes B. E. , , and Szunyogh I. , 2008: Predicting the influence of observations on medium-range forecasts of atmospheric flow. Quart. J. Roy. Meteor. Soc., 134, 20112027.

    • Search Google Scholar
    • Export Citation
  • Shapiro, M. A., , and Thorpe A. J. , 2004: THORPEX International Science Plan, version 3. WMO/TD 1246, WWRP/THORPEX 2, 51 pp. [Available online at http://www.wmo.int/pages/prog/drr/publications/drrPublications/TD1246_THORPEX_International_Science_Plan/WMO_TD1246e.pdf.]

    • Search Google Scholar
    • Export Citation
  • Smith, B. B., , and Mullen S. L. , 1993: An evaluation of sea level cyclone forecasts produced by NMC’s Nested-Grid Model and Global Spectral Model. Wea. Forecasting, 8, 3756.

    • Search Google Scholar
    • Export Citation
  • Swanson, K. L., , and Roebber P. J. , 2008: The impact of analysis error on medium-range weather forecasts. Mon. Wea. Rev., 136, 34253431.

    • Search Google Scholar
    • Export Citation
  • Toth, Z., , and Kalnay E. , 1997: Ensemble forecasting at NCEP and the breeding method. Mon. Wea. Rev., 125, 32973319.

  • Wedam, G. B., , McMurdie L. A. , , and Mass C. F. , 2009: Comparisons of model forecast skill over the east and west coasts of the United States. Wea. Forecasting, 24, 843854.

    • Search Google Scholar
    • Export Citation
  • Zhang, F., , Snyder C. , , and Rotunno R. , 2003: Effects of moist convection on mesoscale predictability. J. Atmos. Sci., 60, 11731185.

Save