• Blackmon, M. L., J. E. Geisler, and E. J. Pitcher, 1983: A general circulation model study of January climate anomaly patterns associated with interannual variation of equatorial Pacific sea surface temperatures. J. Atmos. Sci, 40 , 14101425.

    • Search Google Scholar
    • Export Citation
  • Brankovic, C., T. N. Palmer, and L. Ferranti, 1994: Predictability of seasonal atmospheric variations. J. Climate, 7 , 217237.

  • Chervin, R. M., 1986: Interannual variability and seasonal climate predictability. J. Atmos. Sci, 43 , 233251.

  • Geisler, J. E., M. L. Blackmon, G. T. Bates, and S. Muñoz, 1985: Sensitivity of January climate response to the magnitude and position of equatorial Pacific sea surface temperature anomalies. J. Atmos. Sci, 42 , 10371049.

    • Search Google Scholar
    • Export Citation
  • Hoerling, M. P., and A. Kumar, 2002: Atmospheric response patterns associated with tropical forcing. J. Climate, 15 , 21842203.

  • Horel, J. D., and J. M. Wallace, 1981: Planetary-scale atmospheric phenomena associated with the Southern Oscillation. Mon. Wea. Rev, 109 , 813829.

    • Search Google Scholar
    • Export Citation
  • Kanamitsu, M., and Coauthors, 2002: NCEP dynamical seasonal forecast system 2000. Bull. Amer. Meteor. Soc, 83 , 10191037.

  • Kistler, R., and Coauthors, 2001: The NCEP–NCAR 50-year reanalysis: Monthly means CD-ROM and documentation. Bull. Amer. Meteor. Soc, 82 , 247267.

    • Search Google Scholar
    • Export Citation
  • Kumar, A., and M. Hoerling, 1995: Prospects and limitations of seasonal atmospheric GCM predictions. Bull. Amer. Meteor. Soc, 76 , 335345.

    • Search Google Scholar
    • Export Citation
  • Kumar, A., and M. Hoerling, 1997: Interpretation and implications of the observed inter–El Niño variability. J. Climate, 10 , 8391.

    • Search Google Scholar
    • Export Citation
  • Kumar, A., and M. Hoerling, 1998: Annual cycle of Pacific–North American seasonal predictability associated with different phases of ENSO. J. Climate, 11 , 32953308.

    • Search Google Scholar
    • Export Citation
  • Kumar, A., M. Hoerling, M. Ji, A. Leetmaa, and P. Sardeshmukh, 1996: Assessing a GCM's suitability for making seasonal predictions. J. Climate, 9 , 115129.

    • Search Google Scholar
    • Export Citation
  • Kumar, A., S. D. Schubert, and M. S. Suarez, 2003: Variability and predictability of 200-mb seasonal mean heights during summer and winter. J. Geophys. Res.,108, 4169, doi:10.1029/2002JD002798.

    • Search Google Scholar
    • Export Citation
  • Lau, N-C., 1985: Modeling the seasonal dependence of the atmospheric response to observed El Niños in 1962–76. Mon. Wea. Rev, 113 , 19701996.

    • Search Google Scholar
    • Export Citation
  • Livezey, R. E., and K. C. Mo, 1987: Tropical–extratropical teleconnections during the Northern Hemisphere winter. Part II: Relationships between monthly mean Northern Hemisphere circulation patterns and proxies for tropical circulations. Mon. Wea. Rev, 115 , 31153132.

    • Search Google Scholar
    • Export Citation
  • Madden, R. A., 1976: Estimates of the natural variability of time-averaged sea-level pressure. Mon. Wea. Rev, 104 , 942952.

  • Shukla, J., 1983: Comments on “Natural variability and predictability.”. Mon. Wea. Rev, 111 , 581585.

  • Shukla, J., and D. S. Gutzler, 1983: Interannual variability and predictability of 500 mb geopotential heights over the Northern Hemisphere. Mon. Wea. Rev, 111 , 12731279.

    • Search Google Scholar
    • Export Citation
  • Straus, D. M., and J. Shukla, 2000: Distinguishing between the SST-forced variability and internal variability in mid-latitudes: Analysis of observations and GCM simulations. Quart. J. Roy. Meteor. Soc, 126 , 23232350.

    • Search Google Scholar
    • Export Citation
  • Wallace, J. M., and D. S. Gutzler, 1981: Teleconnections in the geopotential height field during the Northern Hemisphere winter. Mon. Wea. Rev, 109 , 784812.

    • Search Google Scholar
    • Export Citation
  • Yarnal, B., and H. F. Diaz, 1986: Relationship between extremes of the Southern Oscillation and the winter climate of the Anglo– American Pacific coast. J. Climatol, 6 , 197219.

    • Search Google Scholar
    • Export Citation
  • View in gallery
    Fig. 1.

    Analysis of variance results for JAN. (a) Mean internal variance (m2) for the 1-month lead, which shows the 21-yr average of the variability among the 10 ensemble members in each year. Contour interval is 1000 m2. (b) As in (a) but for the 2-month lead. (c) External variance (m2) for the 1-month lead, which shows the variability among ensemble-averaged anomalies due only to interannual changes in SST. Contour interval is 1000 m2. (d) As in (c) but for the 2-month lead. (e) Signal-to-noise ratio for the 1-month lead, which is the ratio of EV to MIV. Contours for 0.25, 0.5, 1, 2, 4, 8, and 16 are shown. (f) As in (e) but for the 2-month lead

  • View in gallery
    Fig. 2.

    Correlations between hindcast ensemble-averaged height anomalies and reanalysis anomalies at 200 hPa for JAN during the entire period 1980–2000 for (a) the 1-month lead and (b) the 2-month lead. Subset of ENSO warm and cold events for (c) the 1-month lead and (d) the 2-month lead, and subset of ENSO neutral years for (e) the 1-month lead and (f) the 2-month lead. Contours for correlations of 0.3 and greater are drawn in intervals of 0.1. Light shading (0.3 < AC < 0.5) indicates some skill, medium shading (0.5 < AC < 0.8) indicates significant skill, and heavy shading (AC > 0.8) indicates high skill

  • View in gallery
    Fig. 3.

    As in Fig. 1 but for JFM

  • View in gallery
    Fig. 4.

    As in Fig. 2 but for JFM

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 267 63 3
PDF Downloads 90 29 1

Potential Predictability in the NCEP CPC Dynamical Seasonal Forecast System

Michael W. PhelpsCenter for Ocean–Atmospheric Prediction Studies, The Florida State University, Tallahassee, Florida

Search for other papers by Michael W. Phelps in
Current site
Google Scholar
PubMed
Close
,
Arun KumarClimate Prediction Center, NCEP, Camp Springs, Maryland

Search for other papers by Arun Kumar in
Current site
Google Scholar
PubMed
Close
, and
James J. O'BrienCenter for Ocean–Atmospheric Prediction Studies, The Florida State University, Tallahassee, Florida

Search for other papers by James J. O'Brien in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

Monthly and seasonal predictions of mean atmospheric states have traditionally been viewed as a boundary forcing problem, with little regard for the role of atmospheric initial conditions (IC). The potential predictability of these mean states is investigated using hindcasted monthly mean January (JAN) and seasonal mean January– February–March (JFM) 200-hPa geopotential heights from the National Centers for Environmental Prediction Climate Prediction Center (NCEP CPC) Dynamical Seasonal Prediction System along with the corresponding data from the NCEP–National Center for Atmospheric Research (NCAR) reanalysis for the period 1980–2000. With lead times ranging from 1 to 4 months, analyses of variance tests are employed to separate the total variability into an unpredictable internal component, due to atmospheric dynamics, and a potentially predictable external component, due to the boundary forcing. These components represent the noise and signal, respectively, and areas where the signal exceeds the noise designate where time averages could be potentially predicted with some degree of skill. Temporal anomaly correlations (ACs) between ensemble-averaged model height anomalies and reanalysis height anomalies also provide a measure of the model skill.

Comparisons between the results of these tests for the different initialization times confirm that, for this model, the atmospheric initial conditions have little effect on the monthly and seasonal means for lead times of one month or more. The model proves to be highly skillful in the Tropics, as expected. Signal-to-noise ratios (SNRs) and ACs also show four areas in the extratropics displaying significant skill: the South Pacific Ocean, Southern Ocean, Southeast Asia, and the Pacific–North America (PNA) region. The skill found in the extratropics outside of the PNA region is highly encouraging. SNRs for JFM are approximately twice those for JAN, suggesting that seasonal forecasts are more reliable than monthly forecasts. Anomaly correlations for El Niño–Southern Oscillation (ENSO) warm and cold events are markedly higher than correlations for both the period 1980–2000 and the subset of ENSO neutral events. The model's ability to accurately capture changes in the atmosphere in response to changes in sea surface temperatures (SSTs) suggests that accurate forecasting of SSTs in the ocean could lead to more accurate forecasts of atmospheric conditions associated with ENSO warm and cold events.

Corresponding author address: Michael W. Phelps, Jacobs Sverdrup Advanced Systems Group, NRL Code 7331, Bldg. 1009, Rm. A138, Stennis Space Center, MS 39529. Email: phelps@nrlssc.navy.mil

Abstract

Monthly and seasonal predictions of mean atmospheric states have traditionally been viewed as a boundary forcing problem, with little regard for the role of atmospheric initial conditions (IC). The potential predictability of these mean states is investigated using hindcasted monthly mean January (JAN) and seasonal mean January– February–March (JFM) 200-hPa geopotential heights from the National Centers for Environmental Prediction Climate Prediction Center (NCEP CPC) Dynamical Seasonal Prediction System along with the corresponding data from the NCEP–National Center for Atmospheric Research (NCAR) reanalysis for the period 1980–2000. With lead times ranging from 1 to 4 months, analyses of variance tests are employed to separate the total variability into an unpredictable internal component, due to atmospheric dynamics, and a potentially predictable external component, due to the boundary forcing. These components represent the noise and signal, respectively, and areas where the signal exceeds the noise designate where time averages could be potentially predicted with some degree of skill. Temporal anomaly correlations (ACs) between ensemble-averaged model height anomalies and reanalysis height anomalies also provide a measure of the model skill.

Comparisons between the results of these tests for the different initialization times confirm that, for this model, the atmospheric initial conditions have little effect on the monthly and seasonal means for lead times of one month or more. The model proves to be highly skillful in the Tropics, as expected. Signal-to-noise ratios (SNRs) and ACs also show four areas in the extratropics displaying significant skill: the South Pacific Ocean, Southern Ocean, Southeast Asia, and the Pacific–North America (PNA) region. The skill found in the extratropics outside of the PNA region is highly encouraging. SNRs for JFM are approximately twice those for JAN, suggesting that seasonal forecasts are more reliable than monthly forecasts. Anomaly correlations for El Niño–Southern Oscillation (ENSO) warm and cold events are markedly higher than correlations for both the period 1980–2000 and the subset of ENSO neutral events. The model's ability to accurately capture changes in the atmosphere in response to changes in sea surface temperatures (SSTs) suggests that accurate forecasting of SSTs in the ocean could lead to more accurate forecasts of atmospheric conditions associated with ENSO warm and cold events.

Corresponding author address: Michael W. Phelps, Jacobs Sverdrup Advanced Systems Group, NRL Code 7331, Bldg. 1009, Rm. A138, Stennis Space Center, MS 39529. Email: phelps@nrlssc.navy.mil

1. Introduction

Predictability of atmospheric means on monthly or seasonal time scales has been historically viewed as a boundary forcing problem, with little attention paid to the possible effects of atmospheric initial conditions (ICs). Such boundary forcings include sea surface temperature (SST), sea ice, snow cover, soil moisture, and other land surface conditions. It is the slow changes in these forcings, and the subsequent response to these changes, that are extensively studied in atmospheric general circulation model (AGCM) simulations. The analysis of AGCM simulations indicate that slowly varying anomalies in lower boundary forcing [i.e., sea surface temperature anomalies (SSTAs)] can have a significant effect on the atmospheric response and are the basis of potential predictability of monthly or seasonal means (Brankovic et al. 1994).

Potential predictability (PP) of model-simulated monthly or seasonal means can be determined by an analysis of the interannual variability of monthly or seasonal means (Chervin 1986). Atmospheric mean states comprise a naturally varying component and a boundary-forced component (Kumar et al. 1996). The naturally varying component is due to the internal dynamics of the atmosphere and is referred to as the internal variability. The boundary-forced component is external to the atmosphere and is referred to as the external variability. Under the assumption that internal variability is a measure of the unpredictable climate noise and the external variability is a measure of the potentially predictable signal, separation of the total variability into its internal and external components allows for the determination of a model's potential skill and can be quantified by means of a signal-to-noise ratio (SNR). The potential for predictability can then be assessed by the amount that the signal exceeds the noise (Shukla and Gutzler 1983).

One of the earliest studies of potential predictability was by Madden (1976). He estimated the so-called natural variability using time-averaged sea level pressure analyses. This is analogous to IV and was referred to as “natural” since it would be present in an unchanging boundary forcing. Madden concluded that potential predictability is low because the total variability was not appreciably larger than his estimates of natural variability. Shukla (1983) later pointed out that Madden's estimates of natural variability were too high, and as such the potential for predictability was underestimated. He noted that Madden's estimate of PP should be viewed as a lower bound for PP.

Kumar and Hoerling (1995) used the separation of total variability methodology on a nine-member ensemble of monthly mean 200-hPa eddy height anomalies for January from an AGCM forced with monthly mean observed SSTs to evaluate the PP of atmospheric mean states. They found that large-scale atmospheric patterns associated with anomalous boundary forcings observed in El Niño–Southern Oscillation (ENSO) extreme events were produced at times in the extratropics. However, skill in the model simulations was not large away from the Tropics, primarily due to large background climate noise. They concluded that the PP in the extratropics is low. Based on the idea that time averaging has a similar effect as ensemble averaging on the internal variance, they suggested that seasonal predictions should be improved over monthly predictions. Internal variance was found to decrease with increasing ensemble size, and therefore time averaging on longer time scales should produce a similar decrease in the internal variance, provided each month in a seasonal average is characteristically similar (Kumar and Hoerling 1995). This, in turn, would result in a higher SNR and a larger potential for predictability for seasonal means. They also noted that maxima in the boundary-forced signal coincided with two centers of the Pacific–North America (PNA) pattern and that this region would offer the best chance for predictability of wintertime climate patterns in the extratropics (Kumar and Hoerling 1998). Predictability was found to be better in boreal winter because the strongest signal in model studies has consistently been found during Northern Hemisphere winter for both Tropics and extratropics. The wintertime signal in the PNA region has also been seen to increase with increasing strength of ENSO events, with a stronger response in warm events compared to cold events (Kumar and Hoerling 1997).

The PNA pattern (Wallace and Gutzler 1981; Horel and Wallace 1981) is one of the two most prominent teleconnections in Northern Hemisphere winter. This pattern has also been linked to tropical SST variability. A train of anomaly centers of alternating signs emanates from the tropical Pacific Ocean. In its positive phase, negative height anomalies are found over the North Pacific Ocean and U.S. Southeast, while positive anomalies are found over Hawaii and western Canada. The centers of action over the North Pacific and western Canada are generally the strongest (Horel and Wallace 1981), and it is these two centers that are associated with the maxima in the boundary-forced signal noted above. Seasonal-mean height anomalies are more likely to prefer a PNA pattern during ENSO winters, although the PNA pattern has been observed in non-ENSO years as well (Yarnal and Diaz 1986). It has also been noted that, during the strong El Niño years, an eastward shift in the atmospheric response pattern in the extratropical latitudes occurs and the atmospheric response, instead of being the PNA pattern, is closer to the tropical Northern Hemisphere (TNH) pattern (Hoerling and Kumar 2002; Livezey and Mo 1987).

Reproduction of the atmospheric response patterns in past model studies is not unprecedented. Blackmon et al. (1983) noted a PNA pattern in 200-, 500-, and 700-hPa geopotential height anomalies produced by an AGCM run in perpetual January mode with a representative SSTA in the tropical Pacific typical of warm ENSO events. Geisler et al. (1985) showed that the PNA pattern was a typical extratropical response in 200-hPa height anomalies for ENSO warm events of varying prescribed strengths. Lau (1985) also produced a PNA pattern similar to that seen in observations when forcing a model with observed SST in the tropical Pacific. An eastward shift in the extratropical response during strong El Niño winters was noted by Hoerling and Kumar (2002).

Apart from the well-documented influence of SSTs on the interannual variability of the extratropical climate, other sources of predictability are continually being sought. One possible candidate is the low-frequency component of the atmospheric ICs and their possible influence on subsequent monthly and seasonal means (Shukla 1983; Straus and Shukla 2000). In this study, hindcasted 200-hPa geopotential heights from the second generation of the National Center for Environmental Prediction Climate Prediction Center (NCEP CPC) Dynamical Seasonal Forecast System starting from ICs with different lead times are examined to determine the potential of predicting monthly and seasonal mean atmospheric states in the NCEP dynamical model. Potential skill of model simulations is determined by SNRs and temporal anomaly correlations (ACs) between ensemble-averaged height anomalies and NCEP–National Center for Atmospheric Research (NCAR) reanalysis anomalies. If the atmospheric ICs have any positive influence, it is expected that simulations with a shorter lead time will provide a more skillful representation of the monthly and seasonal mean observed states of the upper atmosphere.

In section 2, a description of the model and NCEP– NCAR reanalysis data used in this study is presented. Section 3 outlines the methods used to separate the total variability into internal and external components and to determine the level of skill obtained in the model hindcasts. Results are presented in section 4, with their discussion in section 5.

2. Data

The data used in this study are AGCM hindcasted 200-hPa geopotential heights for January–March (JFM) for the period 1980–2000 (Table 1) produced by the NCEP Dynamical Seasonal Forecast System from different lead times. Implemented in April 2000, the second generation system was designed with a primary goal of refining seasonal predictions in the winter season. A detailed description of this system can be found in Kanamitsu et al. (2002).

Prior to the forecasts being made, 21 yr of hindcasts are made for each of the six full months being forecast. Hindcasts are produced by an atmosphere-only model that is initialized with the observed atmospheric conditions from 0000 and 1200 UTC on the first 5 days of the initialization month for each of the 21 yr for a total of 10 members in each ensemble. The only external boundary forcing is observed global monthly mean SSTs. Since SSTs are updated monthly, each simulation for a particular target month, for example, January 1980, is subjected to identical lower boundary forcing regardless of whether the AGCM integrations start in September or December. The use of hindcasts forced with the observed SST conditions provides an estimation of the upper limit of a model's forecast skill. In accordance with past studies, the 200-hPa height anomaly fields are considered to be representative of the midlatitude response to the tropical Pacific SSTs (Geisler et al. 1985).

For January (JAN) monthly means, integrations starting from September, October, November, and December are used. The daily model fields at 2.5° × 2.5° latitude– longitude resolution are averaged to produce monthly mean JAN 200-hPa geopotential height fields. From the four sets of hindcast runs, a total of 840 JAN simulations are available. For JFM seasonal means, integrations from September, October, November, and December are also used. The 3-month daily output is averaged to produce seasonal-mean JFM 200-hPa geopotential height fields. For the JFM seasonal means, a total of 840 simulations are also available. For consistency between JAN and JFM means, the AGCM simulations from December ICs are referred to as the 1-month lead time, the simulations from November ICs as the 2-month lead time, the simulations from October ICs as the 3-month lead time, and the simulations from September ICs as the 4-month lead time.

For each of the 21 yr in the hindcasts from a particular month's ICs, an ensemble mean 200-hPa height field is calculated by averaging the 10 members in the ensemble. The mean of these 21 ensemble averages determines the model climatology for that particular simulation's lead time. Hindcast height anomalies (HAs) are obtained by subtracting the 21-yr climatology from each individual ensemble member.

For comparisons to model data, the observed monthly mean JAN and seasonal mean JFM 200-hPa geopotential heights for 1980–2000 are derived from the NCEP– NCAR 50-yr reanalysis (Kistler et al. 2001). These data are readily available from the NOAA-Cooperative Institute for Research in Environmental Sciences (CIRES) Climate Diagnostics Center. The reanalysis climatologies for JAN and JFM are computed by averaging the 21 yr of JAN and JFM observed data, respectively. Height anomalies for the reanalysis data are calculated in the same manner as described above for the model data.

3. Analysis procedure

Analyses of variance of simulated HA are performed in order to assess the potential predictability of monthly and seasonal atmospheric conditions by isolating the potentially predictable signal produced by the external boundary forcing from the unpredictable background climate noise. For each grid point, let Ahαi represent a hindcast HA, where the subscript α denotes a particular year with a unique SST state and the subscript i denotes a particular member within an ensemble. The ensemble-averaged anomaly for the unique SST state is defined as
i1520-0442-17-19-3775-e1
Despite the fact that each member in an ensemble is subjected to the same SST forcing, the atmospheric anomalies are a blend of the response due to anomalous SSTs and atmospheric internal variability. It is the ensemble average that represents the atmospheric response and is the most likely outcome for the observed anomalies (Kumar and Hoerling 1995). The variability among the 10 ensemble members in a particular year is termed the internal variability (IV) and is given by
i1520-0442-17-19-3775-e2
The IV can differ from year to year due to differences in SST states used to force the AGCM. The mean internal variability (MIV) across all SST states, then, is the average of the IV over the entire 21-yr period and is defined as
i1520-0442-17-19-3775-e3
This MIV is a measure of the background climate noise, which is not predictable from the knowledge of SSTs.
The external variance (EV) is the spread among the ensemble-averaged anomalies and represents the variance due to interannual changes in SST. It is defined as
i1520-0442-17-19-3775-e4
where Ah is the mean anomaly of the entire population and is, by definition, zero. The EV is a measure of the boundary-forced signal, which is potentially predictable. The total variance can be approximated as the sum of the mean internal and external variances. A signal-to-noise ratio can be defined as the ratio of external variance to mean internal variance and is given by
i1520-0442-17-19-3775-e5
Correlations in time between ensemble-averaged hindcast anomalies (A) and reanalysis anomalies (A) are calculated at each grid point using
i1520-0442-17-19-3775-e6
in an effort to determine areas where the model skillfully hindcasted the observed climate anomalies. This is done for the period 1980–2000 and also for subsets of ENSO warm and cold events and ENSO neutral events. Both extremes of ENSO are treated as a whole because of the small number of warm and cold events in this period.

Classification of ENSO events is provided by NCEP CPC and is summarized for 1980–2000 in Table 2. The classification, then, is the consensus of individual evaluations. For this study, the strong (W+) and moderate (W) warm events from the NCEP CPC classification are considered as the “warm” events; the weak warm (W−), neutral (N), and weak cold (C−) events are considered as the “neutral” events; and the moderate (C) and strong (C+) cold events are considered as the “cold” events. Since ENSO events begin in the summer of a classified year, the classification for JAN and JFM appears to lag by one year. For example, the extraordinary 1982–83 warm event shows up as 1983 for JAN and JFM, even though it began in the middle part of 1982. For the period 1980–2000, there are a total of 4 warm events, 3 cold events, and 14 neutral events.

4. Results

a. Analysis for JAN means

Analysis of variance techniques described above are used to separate the total variability into internal and external components to determine signal-to-noise ratios. Because results were so similar for all four lead times, only those from the 1- and 2-month leads are shown. Internal variability (Figs. 1a,b) in the Northern Hemisphere extratropics is found to be about twice that in the Southern Hemisphere. Within 25°–30° of the equator, MIV is very small. Maximum MIV in the Northern Hemisphere is found over the North Atlantic and the Arctic Ocean and just north of the Ross Sea in the Southern Hemisphere. The pattern for MIV is fairly consistent regardless of lead time.

Maps of EV (Figs. 1c,d) show four main features: a dual-lobe pattern in the eastern tropical Pacific straddling the equator, a local maximum in the South Pacific, a slightly stronger local maximum centered just off the coast of China, and the dominant maximum in the North Pacific just south of Alaska, corresponding to one of the centers of action in the PNA pattern. These four features are consistent with each lead time, although there is a markedly higher increase in EV over Antarctica for the 1-month lead compared to the 2-month lead (Fig. 1c).

Signal-to-noise ratios (Figs. 1e,f) are strongest over the eastern tropical Pacific, corresponding to one of the local maxima in EV described above. Areas where signal exceeds noise are generally contained in the Tropics (within ∼20° of the equator) and span the entire tropical belt. In the extratropics, the lone area where SNR > 1 for all lead times is in the North Pacific just south of Alaska, in the same area of the dominant maximum in EV. An interesting feature is the appearance of SNR > 1 over Antarctica for the 1-month lead (Fig. 1e), largely the result of the noted higher EV in the same area. The areal coverage of SNR > 0.25 tends to increase as the lead time decreases. No such increase is observed in the subtropics in the total area where the signal exceeds the noise. Noticeable increases in SNR over the eastern tropical Pacific and equatorial Africa and the western Indian Ocean can be seen as the lead time decreases.

Anomaly correlations for all 21 yr (Figs. 2a,b) show that the highest correlations between ensemble-averaged model-simulated JAN height anomalies and reanalysis height anomalies are found in the tropical belt consistent with high SNR. The maximum AC, exceeding 0.9, is found in the tropical eastern Pacific. Local maxima in AC also correspond to the other three features described earlier in regard to EV. AC > 0.5 is found in the South Pacific and off the coast of China. In the extratropics, the dominant feature is the AC center south of Alaska in the same region as the maximum EV. The strongest extratropical correlations are consistently found here. A new feature is the appearance of AC > 0.5 in the Southern Ocean between 180° and 60°W for the 2- and 1-month leads. Correlations are not as strong for the September and October runs (not shown), but the pattern of increased AC in this area is persistent. Over the entire globe, AC is consistently the same, regardless of lead time.

Computing the AC for the seven ENSO warm and cold years (Figs. 2c,d) shows that correlations are generally larger for the entire globe when compared to AC for 1980–2000. Areas of AC > 0.9 are much larger and extend into the area of maximum EV just south of Alaska and off the coast of China. Correlations over the Southern Ocean now exceed 0.8 and are much more expansive. Also, significant AC greater than 0.8 are now found over portions of North America and the North Atlantic that were not present when all 21 yr were considered as a whole.

For ENSO neutral years (Figs. 2e,f), there is a sharp decrease in AC compared to ENSO warm and cold years over the entire globe, most noticeably in the vicinity of the PNA center south of Alaska. The only area with AC > 0.8 for neutral years is in the eastern tropical Pacific. Very little skill is exhibited in the Southern Ocean or off the coast of China. A local maximum in skill is noted in the South Pacific but is not nearly as large as in ENSO warm and cold years.

b. Analysis for JFM means

Because JFM data are 3-month seasonal averages, IV (Figs. 3a,b) is reduced by about one-third when compared to JAN alone. This is especially evident in the Northern Hemisphere. Reduction in the Southern Hemisphere is not quite as drastic, however, and the IV in the Northern Hemisphere is now comparable to that in the Southern Hemisphere. A smaller than expected reduction in the IV for the Southern Hemisphere for the seasonal means could be due to a strong seasonality in the amplitude of internal variability, for example, an increase in internal variability during the months of February and March (Kumar et al. 2003). Variability in the Tropics within ∼30° of the equator is still very small and the maximum IV is still found over the North Atlantic. This is valid for all lead times.

The EV for the seasonal means (Figs. 3c,d) reveals the same four main features shown for JAN. These are the dual-lobe pattern in the eastern tropical Pacific, the local maximum in the South Pacific, another local maximum over Southeast Asia that extends across to India for JFM, and the dominant maximum in the North Pacific in the area of a PNA center. A small reduction in the external variability for the seasonal means could have contributions from several factors, such as a decrease in the SST anomalies from boreal winter to spring or changes in the extratropical response to tropical SST forcings because of seasonal changes in the atmospheric mean state (Kumar et al. 2003). Overall, EV is slightly reduced compared to JAN, but not as much so as the IV detailed above. The patterns in the EV are fairly consistent, regardless of lead time.

As seen in JAN, the SNR for JFM (Figs. 3e,f) shows a maximum for all lead times in the eastern tropical Pacific. The area where the tropical signal exceeds the noise (SNR > 1) is more expansive for JFM, extending out to ∼30° from the equator. Two regions where local maxima in EV were noted, the South Pacific and the area over China and India, also exhibit SNR > 1. An increase in SNR over the PNA region is also noted, with SNR > 1 over portions of the continental United States and SNR > 2 in the North Pacific. On the whole, the SNR for JFM is more consistent across lead times than the SNR for JAN.

Anomaly correlations for JFM (Figs. 4a,b) for the period 1980–2000 are stronger than for JAN, especially in the Tropics where AC > 0.8, and generally larger than 0.9, within 15° of the equator with only a few exceptions. The area of largest AC is not confined to the eastern tropical Pacific as is found in JAN. Correlations increase to beyond 0.7 for the Southern Ocean, the South Pacific, and over China and India. In the PNA region, the AC is greater than 0.8 and the area of AC > 0.5 is larger as well. Correlations over the United States exceed 0.6 in some instances, a feature not seen in JAN. This pattern of AC is persistent for all lead times.

Isolating the ENSO warm and cold years (Figs. 4c,d) reveals AC > 0.9 for the entire tropical belt within 20°– 25° of the equator. The areal coverage of AC > 0.3 is much larger, as well, when compared to all 21 years. Correlations over the Southern Ocean, the South Pacific, and over China and India increase to greater than 0.8, with isolated areas of AC > 0.9. Correlations in the PNA region exceed 0.9 in the North Pacific and much more of the North American continent is covered by AC > 0.6 when just ENSO warm and cold years are considered. As seen before, the AC pattern for ENSO warm and cold years is similar for all lead times.

Anomaly correlations for ENSO neutral years for JFM (Figs. 4e,f) are reduced when compared to AC for 1980–2000. Correlations in the Tropics still exceed 0.7, which is quite skillful, but areas of AC > 0.9 are nearly nonexistent. The largest reduction is seen over China and India and the continental United States, where correlations drop below 0.3. Correlations over the Southern Ocean and the South Pacific still exceed 0.5, but the areal coverage of such correlations is small. The largest extratropical AC are still found in the PNA region in the North Pacific. This pattern, too, shows little variation with different lead time.

5. Discussion

Hindcasted 200-hPa geopotential heights from the second generation NCEP Dynamical Seasonal Forecast System are used to assess the potential predictability and skill of model-simulated wintertime monthly JAN and seasonal JFM means. In particular, the role of atmospheric initial conditions is investigated by comparing the AGCM's signal-to-noise and simulation skill from lead times varying from 1 up to 4 months. As demonstrated earlier, no significant changes are noted for any of the parameters in this study as lead time decreased for both monthly and seasonal means. Patterns in anomaly correlations and signal-to-noise ratios are consistent for different lead times. This suggests that, for lead times of one month or more, atmospheric initial conditions generally have very little influence on the monthly or seasonal mean variability of upper-level atmospheric circulation during boreal winter. Operational constraints at NCEP CPC require a minimum lead time of one month. NCEP CPC products are released to the public on the Thursday closest to the middle of the month. Thus, to allow adequate time for data processing and analysis, this 1-month lead is the shortest possible lead time. Perhaps lead times of less than a month may show more influence from atmospheric initial conditions on the first month of the simulations.

Based primarily on a markedly increased signal-to-noise ratio, predictions on seasonal time scales are more skillful than monthly forecasts. External variability of ensemble-averaged anomalies is larger for monthly means compared to seasonal means. That is, the predictable signal is slightly reduced for the seasonal means. However, an effect of the time averaging to produce the seasonal mean is that the mean internal variance is reduced by approximately one-third. This reduction in the unpredictable noise is greater than the decrease in predictable signal. This reduction in noise is the primary cause of the increased predictability of seasonal means.

Results from correlations between hindcasted ensemble-averaged anomalies and reanalysis (i.e., “observed”) anomalies suggest that the model simulates the changes in 200-hPa heights associated with ENSO warm and cold events quite well. As expected, the AGCM more accurately simulates upper-air conditions for ENSO warm and cold events than when compared to both the entire period 1980–2000 and the subset of ENSO neutral events. The improvement for ENSO warm and cold events compared to ENSO neutral events is considerable. The finding that extreme ENSO warm and cold events are simulated more skillfully than neutral events is robust. The skill exhibited in hindcasting monthly and seasonal means using observed SSTA to force the model also suggests that accurate forecasts of forcing fields in a coupled system should yield positive results for climate forecasts, especially for the extreme events associated with ENSO, though the skill exhibited in such climate forecasts will be limited by the accuracy of the forecasted SSTs used as forcing fields.

The most encouraging result is the identification of areas outside of the Tropics and the PNA region where the model skillfully simulated the atmosphere. The tropical atmosphere is generally assumed to be mostly driven by fluctuations in SSTA in the tropical Pacific Ocean. Thus, the high correlations and signal-to-noise ratios in the global Tropics are not surprising. It has also been shown that effects of the interannual changes in SSTA in the tropical Pacific are teleconnected to the PNA region, and the level of skill displayed there in the AGCM simulations was also expected. However, the skill exhibited in the South Pacific, the Southern Ocean, and Southeast Asia was not necessarily anticipated. The identification of additional areas where the model skillfully replicates the upper atmosphere, especially in the extratropics away from the primary forcing mechanism, provides some confidence in the forecasting of monthly and seasonal atmospheric anomalies over those regions.

Acknowledgments

Funding for this project was provided in the form of a NOAA/NCEP/FSU fellowship in conjunction with the COAPS/Applied Research Center and the NCEP Climate Prediction Center. COAPS receives its base funding from the NOAA Office of Global Programs. Comments from two anonymous reviewers are also greatly appreciated.

REFERENCES

  • Blackmon, M. L., J. E. Geisler, and E. J. Pitcher, 1983: A general circulation model study of January climate anomaly patterns associated with interannual variation of equatorial Pacific sea surface temperatures. J. Atmos. Sci, 40 , 14101425.

    • Search Google Scholar
    • Export Citation
  • Brankovic, C., T. N. Palmer, and L. Ferranti, 1994: Predictability of seasonal atmospheric variations. J. Climate, 7 , 217237.

  • Chervin, R. M., 1986: Interannual variability and seasonal climate predictability. J. Atmos. Sci, 43 , 233251.

  • Geisler, J. E., M. L. Blackmon, G. T. Bates, and S. Muñoz, 1985: Sensitivity of January climate response to the magnitude and position of equatorial Pacific sea surface temperature anomalies. J. Atmos. Sci, 42 , 10371049.

    • Search Google Scholar
    • Export Citation
  • Hoerling, M. P., and A. Kumar, 2002: Atmospheric response patterns associated with tropical forcing. J. Climate, 15 , 21842203.

  • Horel, J. D., and J. M. Wallace, 1981: Planetary-scale atmospheric phenomena associated with the Southern Oscillation. Mon. Wea. Rev, 109 , 813829.

    • Search Google Scholar
    • Export Citation
  • Kanamitsu, M., and Coauthors, 2002: NCEP dynamical seasonal forecast system 2000. Bull. Amer. Meteor. Soc, 83 , 10191037.

  • Kistler, R., and Coauthors, 2001: The NCEP–NCAR 50-year reanalysis: Monthly means CD-ROM and documentation. Bull. Amer. Meteor. Soc, 82 , 247267.

    • Search Google Scholar
    • Export Citation
  • Kumar, A., and M. Hoerling, 1995: Prospects and limitations of seasonal atmospheric GCM predictions. Bull. Amer. Meteor. Soc, 76 , 335345.

    • Search Google Scholar
    • Export Citation
  • Kumar, A., and M. Hoerling, 1997: Interpretation and implications of the observed inter–El Niño variability. J. Climate, 10 , 8391.

    • Search Google Scholar
    • Export Citation
  • Kumar, A., and M. Hoerling, 1998: Annual cycle of Pacific–North American seasonal predictability associated with different phases of ENSO. J. Climate, 11 , 32953308.

    • Search Google Scholar
    • Export Citation
  • Kumar, A., M. Hoerling, M. Ji, A. Leetmaa, and P. Sardeshmukh, 1996: Assessing a GCM's suitability for making seasonal predictions. J. Climate, 9 , 115129.

    • Search Google Scholar
    • Export Citation
  • Kumar, A., S. D. Schubert, and M. S. Suarez, 2003: Variability and predictability of 200-mb seasonal mean heights during summer and winter. J. Geophys. Res.,108, 4169, doi:10.1029/2002JD002798.

    • Search Google Scholar
    • Export Citation
  • Lau, N-C., 1985: Modeling the seasonal dependence of the atmospheric response to observed El Niños in 1962–76. Mon. Wea. Rev, 113 , 19701996.

    • Search Google Scholar
    • Export Citation
  • Livezey, R. E., and K. C. Mo, 1987: Tropical–extratropical teleconnections during the Northern Hemisphere winter. Part II: Relationships between monthly mean Northern Hemisphere circulation patterns and proxies for tropical circulations. Mon. Wea. Rev, 115 , 31153132.

    • Search Google Scholar
    • Export Citation
  • Madden, R. A., 1976: Estimates of the natural variability of time-averaged sea-level pressure. Mon. Wea. Rev, 104 , 942952.

  • Shukla, J., 1983: Comments on “Natural variability and predictability.”. Mon. Wea. Rev, 111 , 581585.

  • Shukla, J., and D. S. Gutzler, 1983: Interannual variability and predictability of 500 mb geopotential heights over the Northern Hemisphere. Mon. Wea. Rev, 111 , 12731279.

    • Search Google Scholar
    • Export Citation
  • Straus, D. M., and J. Shukla, 2000: Distinguishing between the SST-forced variability and internal variability in mid-latitudes: Analysis of observations and GCM simulations. Quart. J. Roy. Meteor. Soc, 126 , 23232350.

    • Search Google Scholar
    • Export Citation
  • Wallace, J. M., and D. S. Gutzler, 1981: Teleconnections in the geopotential height field during the Northern Hemisphere winter. Mon. Wea. Rev, 109 , 784812.

    • Search Google Scholar
    • Export Citation
  • Yarnal, B., and H. F. Diaz, 1986: Relationship between extremes of the Southern Oscillation and the winter climate of the Anglo– American Pacific coast. J. Climatol, 6 , 197219.

    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

Analysis of variance results for JAN. (a) Mean internal variance (m2) for the 1-month lead, which shows the 21-yr average of the variability among the 10 ensemble members in each year. Contour interval is 1000 m2. (b) As in (a) but for the 2-month lead. (c) External variance (m2) for the 1-month lead, which shows the variability among ensemble-averaged anomalies due only to interannual changes in SST. Contour interval is 1000 m2. (d) As in (c) but for the 2-month lead. (e) Signal-to-noise ratio for the 1-month lead, which is the ratio of EV to MIV. Contours for 0.25, 0.5, 1, 2, 4, 8, and 16 are shown. (f) As in (e) but for the 2-month lead

Citation: Journal of Climate 17, 19; 10.1175/1520-0442(2004)017<3775:PPITNC>2.0.CO;2

Fig. 2.
Fig. 2.

Correlations between hindcast ensemble-averaged height anomalies and reanalysis anomalies at 200 hPa for JAN during the entire period 1980–2000 for (a) the 1-month lead and (b) the 2-month lead. Subset of ENSO warm and cold events for (c) the 1-month lead and (d) the 2-month lead, and subset of ENSO neutral years for (e) the 1-month lead and (f) the 2-month lead. Contours for correlations of 0.3 and greater are drawn in intervals of 0.1. Light shading (0.3 < AC < 0.5) indicates some skill, medium shading (0.5 < AC < 0.8) indicates significant skill, and heavy shading (AC > 0.8) indicates high skill

Citation: Journal of Climate 17, 19; 10.1175/1520-0442(2004)017<3775:PPITNC>2.0.CO;2

Table 1.

Schematic representation of model output for this study. Bold months represent the monthly data used and italicized months are the initialization months

Table 1.
Table 2.

Classification of ENSO events for JAN and JFM for the period 1980–2000. “Warm” denotes the strong (W+) and moderate (W) warm events from the NCEP CPC classification; “Neutral” de notes the weak warm (W−), neutral (N), and weak cold (C−) events; and “Cold” denotes the moderate (C) and strong (C+) cold events

Table 2.
Save