The link between El Niño and the California wintertime rainfall has been reported in various studies. During the winter of 1994/95, warm sea surface temperature anomalies (SSTAs) were observed in the central Pacific, and widespread significant flooding occurred in California during January 1995 and March 1995. However, the El Niño–Southern Oscillation alone cannot explain the flooding. In March 1995 California suffered flooding after the warm SSTA over the central Pacific had weakened considerably. During November and December, in spite of El Niño conditions, California was not flooded, and more than two standard deviations above normal SSTA in the North Pacific were observed. A possible link between midlatitude warm SSTA and the timing of the onset of flooding is suspected within the seasonal forecasting community.
The climate condition during the northern winter of 1994/95 is described using the National Centers for Environmental Prediction–National Center for Atmospheric Research reanalysis data. Diagnostics show the typical El Niño pattern in the seasonal mean and the link between the position of the jet exit and the flooding over California on the intraseasonal timescale.
The relationship among California floods, the Pacific jet, tropical rainfall, and SSTA is inferred from results of general circulation model (GCM) experiments with various SSTAs. The results show that the rainfall over California is associated with an eastward extension of the Pacific jet, which itself is associated with enhanced tropical convection over the warm SSTA in the central Pacific. The GCM experiments also show that rainfall over the Indian Ocean can contribute to the weakening of the Pacific jet and to dryness over California. The GCM experiments did not show significant impact of North Pacific SSTA, either upon the Pacific jet or upon rainfall over California. The agreement with diagnostics results is discussed. GCM experiments suggest the link between the tropical intraseasonal oscillation (TIO) and the flooding in March in California, since there is a strong TIO component in rainfall over the Indian Ocean.
In January 1995, intense periods of rainfall were reported at most of the stations in northern and central California. The near-record rainfall caused widespread flooding, mudslides, and heavy property loss. The rains stopped at the end of January, leaving February a relatively dry month, but heavy precipitation returned to California during the first half of March. The atmospheric and oceanic conditions are described based on the National Centers for Environmental Prediction (NCEP)–National Center for Atmospheric Research (NCAR) reanalysis (Kalnay et al. 1996). The dynamical mechanisms linked to sea surface temperature anomalies (SSTAs) for the floods of the 1994/95 winter are examined using general circulation model (GCM) experiments.
Rainfall over the west coast of the United States, especially California, has been one of the main foci of seasonal forecasts. Warm SSTAs greater than 1°C were observed in the central Pacific in October and persisted through January 1995; this SSTA pattern exemplifies the warm El Niño–Southern Oscillation (ENSO) phase. Schonher and Nicholson (1989) reported abnormally high rainfall in California during most of the type-1 warm ENSO events defined by Fu et al. (1986). These ENSO events have strong warming in the central and eastern Pacific and close to normal conditions in the western Pacific. Their results were confirmed by Cayan and Webb (1992) and Kahya and Dracup (1994), who related stream flows in the Pacific southwest of the United States to ENSO. Livezey et al. (1997) found statistically significant relationships between wet (dry) conditions in California and dry (wet) conditions in the Pacific Northwest and the warm (cold) phase of the ENSO.
However, heavy rainfall also occurred during non-ENSO years, and California was dry during many warm ENSO episodes. Ropelewski and Halpert (1986) did not find a consistent correlation between rainfall in California and ENSO events. Livezey et al. (1997) also showed that rainfall over California is significant for composites with two- to three-month windows, but the relationship is weaker for the single-month composites. Hsu (1996) showed the extratropical signal of tropical intraseasonal oscillation (TIO). His work suggested that the TIO could influence the intensity of the Pacific jet and circulation pattern over the west coast of the United States. Mo and Higgins (1998) studied the lagged statistical relationship between tropical heating and California rainfall. In this paper, in addition to ENSO, the evidence of modulation by the TIO during the 1994/95 winter is discussed. During March, heavy rainfall returned to California when the warm SSTA in the central Pacific weakened considerably. The heavy rainfall in March suggests that the impact of the TIO could be as significant as the impact of ENSO.
Another intriguing element was the timing of the onset of the flooding. During the 1994/95 winter, a warm SSTA greater than 1°C in the North Pacific was observed. This is more than two standard deviations above normal. This warm SSTA peaked in the middle of December 1994, then started to decrease as the Pacific jet extended to the east. It is suspected that this warm midlatitude SSTA affected the timing of the jet development. The cause and effect relationship between the jet and the warm SSTA is investigated using GCM experiments and diagnostics based on NCEP–NCAR reanalysis data.
2. Data and model
The primary data used in this study are the daily averaged global data from the NCEP–NCAR reanalysis from 1979 to 1995 (Kalnay et al. 1996). The data have spectral resolution with triangular truncation at wavenumber 62. In this paper, the daily annual cycle is computed for each day from daily mean data from January 1979 to December 1995 (instead of using harmonic fitting) in order to avoid any overfitting problem. Overfitting distorts the results especially for SST fields. The monthly mean of the daily anomaly will exactly agree with the anomaly computed from monthly data. To obtain low-pass time-filtered fields, a Lanczos filter (Duchon 1979) with 123 points was used.
The daily climatology of outgoing longwave radiation (OLR) data is computed using the OLR data from Liebmann and Smith (1996). During 1994/95, OLR data from the National Environmental Satellite, Data and Information Service are used (Janowiak et al. 1985). Daily station rainfall data is obtained from the climate assessment database from the National Oceanic and Atmospheric Administration’s Climate Prediction Center. The global rainfall is estimated by Xie and Arkin (1996).
The GCM used is the climate version of the NCEP operational medium-range forecast (MRF) model with a Kuo scheme (MRF9). MRF9 has a horizontal T40 spectral resolution and 18 vertical levels. The convection scheme and physics are adjusted for seasonal forecasts (Ji et al. 1994). The forecast skill and predictability based on 11-yr integrations forced by observed SSTs (SST runs) are documented by Livezey et al. (1996). The responses to SST during ENSO are realistic in comparisons with observations (Livezey et al. 1997).
All GCM experiments started on 15 November 1994 and ran through 31 March 1995. SSTs were constructed from November 1994 monthly mean SSTA and the climatological daily SSTs. The model climatologies were obtained from nine SST runs from 1982 to 1992. Anomalies were defined as departures from the model climatologies. Initial conditions for each of the nine members of SST runs and forecast runs were obtained by perturbing the atmospheric initial conditions derived from balanced model winter states from earlier simulations (Livezey et al. 1996). Four sets of ensemble forecast runs with nine members are presented. The same set of nine initial conditions was used for all experiments. Therefore, the differences in the simulated responses are only from the SSTAs. For atmospheric variables, ensemble means are formed by averaging over nine realizations in each set.
3. Circulation anomalies during the 1994/95 winter
a. Seasonal mean fields
The seasonal mean from December 1994 through February 1995 (DJF9495) shows a typical warm ENSO pattern (Fig. 1). The 500-hPa geopotential height anomalies averaged over DJF9495 showed negative anomalies in the Pacific and positive anomalies over Canada. This pattern has a large projection onto the negative tropical Northern Hemisphere (TNH) mode. The negative TNH is associated with heavy precipitation over the western United States and a warm ENSO (Livezey et al. 1997). The rainfall anomalies estimated by Xie and Arkin (1996) are shaded in Fig. 1, showing anomalous rainfall over California. Figure 2 shows total zonal wind and anomalous zonal wind at 200 hPa averaged for DJF9495. It shows that during DJF9495, the westerly wind was intensified over the eastern central Pacific. Cayan and Redmond (1994) related the westerlies along the West Coast to precipitation and found that, while the westerlies extended to the West Coast, precipitation in California was generally heavy.
b. Relationship between the jet and rainfall
The total rainfall averaged over seven California stations (San Diego, Los Angeles, San Francisco, Blue Canyon, Sacramento, Riverside, and Red Bluff) show moderate rainfall during November and December (Fig. 3). Two intense periods of heavy rainfall were observed during January with a sharp break in between. February was relatively dry, but heavy rainfall returned at the beginning of March and continued for about 15–20 days.
The movement of the jet is illustrated by plotting the time–longitude cross section (Fig. 4) of the 10-day low-pass filtered 200-hPa zonal wind averaged from 30° to 35°N. Until the middle of December the jet stayed west of the date line. In the middle of December, the jet started to extend eastward, and its exit region moved toward the west coast of the United States. The jet was well established in its new position by the end of December, and rainfall started in California. The jet persisted during January with a sharp break around 15 January. During February, while California was dry, the jet moved back westward to its climatological position, and even easterly winds appeared over California. During the March episode, the jet shifted eastward again and persisted near the west coast of the United States until the middle of March. As suggested by Cayan and Redmond (1994), the subtropical jet extended to the west coast of the United States during both rainfall episodes. Comparing Figs. 3 and 4, the extension of the jet and the rainfall over California agree well.
Figure 5 shows 200-hPa divergence and zonal wind fields during wet and dry periods. During the January flood, the Pacific jet extended to the east, and its exit reached to the south of California (Fig. 5a). The 200-hPa divergence fields north of the jet exit over California agree with the rainfall over California. On the other hand, during February, this jet exit moved to the west over the Pacific Ocean, and the region of 200-hPa divergence also moved to the west (Fig. 5b). In March the jet exit moved back to the south of California, which lay beneath a region of 200-hPa divergence (Fig. 5c). This relationship between the jet and the upper-air divergence shows that the rainfall over California during these periods is associated with a secondary circulation at the north of the exit of the Pacific jet. Mo and Higgins (1998) related the anomalies of jet intensity to the rainfall over California. However, it is important to observe the total fields in order to identify the location of the jet exit.
c. Sea surface temperature anomalies
The warm ENSO conditions started to develop during the 1994 summer and continued to strengthen throughout the fall. During November (Fig. 6a), a positive SSTA greater than 1°C covered the central Pacific from the date line to the South American coast. This SSTA gradually weakened through the boreal winter. During March 1995 (Fig. 6e), there were no significant SSTAs in the central Pacific. In the extratropics, a positive SSTA greater than 1°C was centered around 35°N, 170°W in November and persisted through mid-December (Fig. 6b). After mid-December this SSTA weakened, and it diminished by mid-February. Over the Indian Ocean, positive SSTAs greater than 1°C were observed from November 1994 to January 1995, but anomalies weakened by the end of January.
d. Tropical convection
Figure 7 depicts the evolution of the 10-day low-pass filtered OLR anomaly (OLRA) averaged from 10°S to 10°N. The persistent negative OLRA in the central Pacific from November to January corresponds well with the warm SSTA in the area. In March, the negative OLRA in the central Pacific decreased when the warm SSTA in the central Pacific weakened. The time evolution of OLRA also reveals a strong TIO. During the winter of 1994/95, there were three TIO episodes: a slow event in early winter and two fast events in January and March. Each episode is clearly marked by eastward propagation of negative OLRA from the Indian Ocean to the central Pacific. In section 4b the modulation of the TIO in the California rainfall is discussed based on results from GCM experiments.
4. General circulation model experiments with various SSTAs
The GCM is used to study the linkage between SSTAs and the midlatitude circulation associated with California floods. The model and basic setting of the experiments have been described in section 2. The November 1994 monthly mean SSTA used in this experiment has a major warm SSTA in the central Pacific, unusually warm midlatitude SSTA in the northern Pacific, and moderately warm SSTA in the Indian Ocean. For each experiment, different parts of the SSTA are used to force the atmosphere to determine the impact of SSTA in different areas. Four sets of nine-member ensemble runs are diagnosed to determine the impact of each SSTA on the California floods. Global SSTAs are used for the global SST (GSST) experiment. The Tropical Pacific SST (TPSST) experiment is forced by central Pacific SSTA only, to isolate the impact of the central Pacific SSTA from other SSTAs. The midlatitude SST (MSST) experiment is forced by midlatitude SSTA only, to study the impact of midlatitude SSTA. The no Tropical Pacific SST (NTPSST) experiment is forced by all SSTAs except those of the central Pacific. The SSTA for NTPSST and TPSST added together gives the SSTA for the GSST experiment.
a. Results from GCM experiments
The SSTAs used in the four sets of experiments are plotted in Fig. 8. Here, the SSTAs are the anomalies from the 1982–92 climatology. The SSTAs were fixed during each experiment, but note that the SST itself varies according to daily climatology. The mean anomalies from the season January–March, relative to the 1982–92 model climatology, are presented for each of the four sets of experiments and for the following variables: global rainfall patterns (Fig. 9), 200-hPa zonally asymmetric streamfunction (Fig. 10), 200-hPa zonal wind (Fig. 11), rainfall over the Pacific–North American (PNA) region with finer contour intervals (Fig. 12), and 200-hPa divergence over the PNA region (Fig. 13). The significance levels of rainfall anomalies in Fig. 12 are computed based on the internal variance of both forecast and climatology (see the appendix).
1) GSST experiment
In the GSST experiment, the response to warm SSTA (Fig. 8a) in the central Pacific is modified by the warm SSTA in the North Pacific and in the Indian Ocean. The net result is below-normal rainfall over California (Fig. 12a). In midlatitudes, warm SSTAs greater than 2°C are located in the North Pacific near 35°N, 170°W. In the Tropics, over the warm SSTA, 400 mm month−1 rainfall anomalies are formed in both the Pacific and Indian oceans. Although the SSTA over the Indian Ocean is not strong, its impact on the rainfall was large. Note that this intense rainfall over the warm SSTA is heavily dependent on the convection scheme used for MRF9, which tends to overrespond to warm total SST in the Indian Ocean. There is a slight intensification of the subtropical jet near the date line, but the 200-hPa zonal wind anomalies over California are negative, the positive anomaly being located farther south (Fig. 11a).
Some significant positive rainfall anomalies are found over the warm SSTAs in the midlatitude North Pacific (Fig. 12a). The rainfall anomaly east of the date line is weakly associated with 200-hPa divergence fields (Fig. 13a). Therefore, there is a possibility that this SSTA could affect the large-scale circulation. The rainfall anomaly over the western Pacific is not associated with the 200-hPa divergence anomaly (Fig. 13a). Therefore, it does not have an impact on upper-air circulation features such as the subtropical jet.
The streamfunction pattern over the central Pacific in Fig. 10a shows a similarity to the streamfunction during El Niño (Hoskins et al. 1989), showing a clear anticyclonic pair over the tropical Pacific. There is a weak anticyclonic circulation over the Indian Ocean as well. The west coast of the United States is covered by an anticyclonic circulation, and the rainfall anomaly over California was negative (Fig. 12a).
This experiment shows that if global persistent SSTA (GSST) is used for seasonal forecasts, the rainfall anomaly for the 1994/95 winter becomes dry over California. Figure 6 showed that SSTAs varied in various timescales, and SSTAs over the Pacific midlatitudes are weakened by January. Therefore, if midlatitude SSTAs cause a major impact on the circulation, persisting midlatitude SSTA will degrade the seasonal forecasts. During late December, in spite of warm SSTA, a cooling anomaly was observed over the Indian Ocean (Fig. 7) while the GSST experiment shows heavy rainfall (Fig. 9). TPSST, MSST, and NTPSST are performed to evaluate the impact of various persisting SSTAs.
2) TPSST experiment
The second set of experiments simulates the atmospheric responses to the combination of warm SSTA in the tropical Pacific but normal conditions in midlatitudes and in other tropical oceans (Fig. 8b). The SSTAs in the tropical Pacific from 15°S to 15°N are the same as in the GSST experiment. SSTAs are damped to zero in 10° transition zones from the Tropics to midlatitudes in both hemispheres. Outside the Pacific basin, the climatological SSTs are used. Figure 11b shows positive 200-hPa zonal wind anomalies extending to the east around 30°N, similar to the wind anomalies during DJF 1994/95. This experiment shows the relationship between the warm SSTAs in the Tropics and the rainfall in California. Figure 10b shows the anticyclonic dipole pattern of anomalous streamfunction over the warm SSTA in the Pacific. This streamfunction pattern resembles the response to the heating over the date line presented by Jin and Hoskins (1995) using a simple GCM. The zonal wind intensified between 30° and 40°N. This experiment showed that tropical Pacific warm SSTAs and rainfall anomalies caused significant above-normal rainfall in California (Fig. 12b) and dry conditions in Washington and British Columbia. The pattern is significantly correlated to El Niño events (Livezey et al. 1997).
3) MSST experiment
This set of experiments simulates the atmospheric responses to the SSTA in midlatitudes (Fig. 8c). The climatological SSTs are used in the Tropics from 15°S to 15°N. The SSTAs south of 25°S and north of 25°N are the same as in the GSST experiment. Between these zones are transition zones.
There is no significant tropical rainfall anomaly in this experiment. The weak anomalies in either the 200-hPa streamfunction or in the zonal flow do not produce a significant rainfall anomaly over California (Figs. 10c, 11c, and 12c). This experiment shows that midlatitude SSTAs do not generate significant atmospheric circulation anomalies on the seasonal timescale. These results depend on the GCM used in this experiment, and the results will be evaluated through the diagnostics of observation.
4) NTPSST experiment
In this experiment, the tropical Pacific SSTs are set to the climatology (Fig. 8d). SSTAs of experiments TPSST and NTPSST combined make the SSTA used in experiment GSST. In midlatitudes, the location of the rainfall anomaly pattern over Pacific midlatitudes and the SSTA do not agree. Over the warm SSTA in the western Pacific the rainfall anomaly is positive, but over the warm SSTA in the eastern Pacific the rainfall anomaly is negative. NTPSST as well as MSST experiments show that midlatitude SSTAs do not produce a significant impact on the large-scale circulation. The most significant contributing factor to make the difference is tropical convection over the warm SSTAs in the Indian Ocean (Fig. 9d), which is absent in the MSST and TPSST experiments. Figure 10d shows the anticyclonic pair of asymmetric streamfunction anomalies at 200 hPa over the warm SSTA in the Indian Ocean and a cyclonic pair over the Pacific. Note that, in the TPSST experiment, heating over the central Pacific produces an anticyclonic pair in the same area and intensifies the Pacific subtropical jet. The cyclonic pair over the Pacific in NTPSST experiment is consistent with the weakening of the Pacific subtropical jet as well as with negative rainfall anomaly over California (Fig. 12d).
Sardeshmukh and Hoskins (1988) showed that divergence between the tropical western Pacific and the date line produces a similar anticyclonic pair of anomalies over the date line. On the other hand, divergence over the Indian Ocean produces a cyclonic pair in the same area. These results are confirmed by Jin and Hoskins (1995) using a multilevel model. They show that heating over the Indian Ocean and heating over the Central Pacific produce opposite impacts over the PNA region. Therefore, warm (cold) SSTAs over the tropical Indian Ocean contribute to weaken (intensify) the Pacific jet.
b. Summary and discussion of GCM experiments
The results of GCM experiments will be discussed with diagnostics using NCEP–NCAR reanalysis data. During the experiments, SSTAs were fixed; however, the actual SSTAs changed on various timescales. Therefore, the impact of some SSTAs may be overestimated. The modifications in the convection scheme introduced by Ji et al. (1994) cause convection to occur only over warm SST. As a result, there is less convection over midlatitudes and excessive convection over the Indian Ocean. Therefore, it is important to confirm the results of GCM experiments through diagnostics using observation.
1) Forcing from tropical Pacific SSTA during January
The influence of tropical heating propagates into midlatitudes if it forms a Rossby wave source (RWS; Sardeshmukh and Hoskins 1988) by interacting with ambient flow. The RWS is computed from NCEP–NCAR reanalysis data in order to identify the origin of the forcing for the midlatitude waves. RWS is defined as
where ζ is the absolute vorticity and Uχ is the divergent wind.
Significantly above normal rainfall in California and off the coast in the TPSST experiment links the California floods to enhanced tropical convection in the Pacific. This agrees with the fact that the tropical Pacific SSTA during the warm ENSO conditions is related to California rainfall. During 1–15 January 1995, the first outbreak of excessive rainfall, the positive anomalous RWS was located about 35°N in the central Pacific (Fig. 14). This RWS was caused by convergence associated with the sinking of the local Hadley cell produced north of the warm SSTA in the central Pacific. This RWS pattern is not associated with the RWS pattern in the TIO (Hsu 1996). Therefore, the RWS pattern showed that during the first half of January, the central Pacific warm SSTA did cause the disturbance in midlatitude. It is interesting to note that the anomalous RWS is not formed directly by anomalous divergence over the warm SSTA, but by the convergence associated with the sinking of the anomalous local Hadley cell caused by warm SSTA. Note that the TNH pattern observed in Fig. 1 is formed downstream of this RWS.
2) Midlatitude SSTA
The forcing from the warm SSTA over midlatitudes during December 1994 was suspected as a cause of the delay in the onset of the flooding over California. Figure 15 shows SSTA, upward heat flux, and 200-hPa divergence anomaly for December 1994. If the forcing of the jet is caused by the warm SSTA, the heat flux anomaly needs to be upward over the warm SSTA in order to show that the sense of the forcing is from the ocean to the atmosphere. The upper-air divergence over the warm SSTA shows that the influence of the heating caused by the warm SSTA penetrates to the jet level. These are the necessary conditions to show that forcing from a warm SSTA influences the jet. During December 1994, over the warm SSTA in the central North Pacific, there is no clear relationship between the divergence, heat flux, and SSTA pattern. In fact, within the month, the SSTA peaked in the middle of December while, simultaneously, the direction of the heat flux changed from downward to upward. As regards monthly averages, there is no obvious relationship between the heat flux and SSTA patterns. This indicates that the central North Pacific warm SSTA is responding to heating from the atmosphere, rather than the converse (Cayan 1992). In the western North Pacific, the heat flux pattern agrees with the pattern of warm SSTA, but there is no upper-air divergence field associated with it. Therefore, this heating from the warm SSTA may have caused low-level large-scale rain, but its influence did not penetrate as far as the jet level. The results from the diagnostics of observed data support the results from GCM experiments.
3) Heating over the Indian Ocean and intraseasonal variability
The March rainfall episode occurred after the ENSO conditions had weakened. During the TIO, the region of heating moves from the middle of the Indian Ocean to the central Pacific Ocean (Hsu 1996). Figure 7 shows this migration clearly. There were three TIO events during the 1994/95 winter. GCM experiments showed that heating over both the Indian and central Pacific Oceans affects the rainfall over California, but in opposite ways. Therefore, an impact of the TIO on the California floods is expected. Figure 16 shows the RWS pattern averaged over 1–15 March. This RWS pattern is what is observed during the TIO (Hsu 1996), while the RWS pattern during January (Fig. 14) is not related to the TIO. Therefore, the March episode was caused by the TIO, while the January episode was caused by the warm central Pacific SSTA.
These diagnostics showed that California rainfall is affected by the TIO as much as by the ENSO since the intensity of the March episode is as strong as the January episode (Fig. 3). In fact, planetary waves of a shorter timescale, with periods of 15–20 days, also have a strong influence on the California rainfall. During the sharp cessation of flooding in mid-January, the intraseasonal waves with periods of 15–20 days were active over the North Pacific. These are the waves that were presented by Kiladis and Weickmann (1992) from diagnostics of observed data and, independently, by Hoskins and Ambrizzi (1993) in experiments using a numerical model. Statistical diagnostics of the TIO (Mo and Higgins 1998;Hsu 1996) are affected by the presence of the 15–20-day waves; the detailed phase relationship between the TIO and California rainfall deserves a careful study.
During the winter of 1994/95, California was subjected to widespread severe flooding, and a warm SSTA formed in the central Pacific that was strong enough to produce ENSO-related rainfall over the western United States. The salient features associated with this warm event were a warm SSTA in the North Pacific and the return of floods in March 1995 after the SSTA over the central Pacific had weakened. There was a significant warm SSTA over the Indian Ocean also. During both heavy rainfall episodes in January and March, the Pacific jet extended from the North Pacific to the west coast of the United States. The March episode could not be linked to the ENSO since the SSTA in the central Pacific had, by then, already weakened. There are strong indications that the March episode was influenced by the TIO.
GCM experiments show the positive link among central Pacific warm SSTA, intensification of the Pacific jet, and California rainfall. They also show the negative impact of heating over the Indian Ocean on California rainfall. The North Pacific SSTA by itself did not produce a significant impact on the large-scale circulation. The results are confirmed by observations.
The relationship between SSTA and rainfall anomaly depends on the GCM used (Masutani 1998). However, the relation between the tropical heating anomaly and the circulation anomaly is robust. The basic pattern of circulation anomalies that are produced by MRF9 agrees with results from a one-layer model experiment (Sardeshmukh and Hoskins 1988) and simple GCM experiments (Jin and Hoskins 1995), and also with observations (Hoskins et al. 1989). The GSST and NTPSST experiments showed that correct heating over the Indian Ocean is important for long-range forecasts over the United States. Since MRF9 overresponds to the SSTA over the Indian Ocean, persisting SSTA over the Indian Ocean throughout the forecasting period distorted the rainfall pattern over the United States. For the operational seasonal forecasts at NCEP, midlatitude SSTAs and Indian and Atlantic Ocean tropical SSTAs are damped toward climatology (Masutani 1997).
We greatly appreciate assistance from Drs. James Purser and Kingste Mo during the preparation of the manuscript. The GCM experiment was originally performed to evaluate seasonal forecasts at NCEP by Dr. Ming Ji. We appreciate his making the data available for this work. We would like to thank Dr. George Kiladis and Mr. John Janowiak, who provided us with the OLR data. Daily data for the NCEP–NCAR reanalysis were obtained from an archive maintained by Dr. Wayne Higgins. Dr. R. J. Purser provided compact high-order finite-difference programs for this work. This work is partially supported by the NOAA Office of Global Programs.
A Significance Test for the Ensemble Anomaly Forecasts
Student’s t-test is performed to evaluate the significance of the signals. To compute t values, the variance in both forecast and climatology are considered:
where is the climatological average of the ensemble mean computed from an 11-yr integration with observed SSTs (SST runs), f is the ensemble average of the forecasts, σ2clm is the variance in climatology, σ2fcst is the internal variance in forecast, and m is the number of realizations in the forecast. When the variance in climatology is evaluated, the variance of the ensemble mean, σ2EM, can be decomposed into two parts (Rowell et al. 1995): the variance due to external forcing, inthis experiment SST (σ2sst); and the internal variance [(1/n)σ2clm_int],
where n is the number of realization in SST runs.
Since the variance of SST mainly originates from the variance of SST in the central Pacific, the internal variance is the most relevant part to be used for the significance test for these experiments. For the GSST and the TPSST experiments the levels of significance do not change enough to alter the conclusion. The level of significance for NTPSST increases by using (1/n)σ2clm_int. In actual forecasts, σ2EM is used.
* Additional affiliation: Research and Data Systems Corporation, Greenbelt, Maryland.
Corresponding author address: Dr. Michiko Masutani, NOAA/NWS/NCEP/EMC, Rm. 207, 5200 Auth Road, Camp Springs, MD 20746.