The representation of the CFSv2 ocean-atmosphere ensemble hindcasts is investigated during Dec-Jan-Feb (DJF) and Jun-Jul-Aug (JJA) from 1983 to 2010. The skill anomaly correlations showed that in some continents the forecasts do not have dependency with changes in the initial conditions. Also, in both seasons the model has a higher skill at the 0-month lead time with the largest spatial biases occurring over the North America, South America, and Oceania. Over the continents the largest biases in the nonlinearity of El Niño minus La Niña events are found over the eastern South Africa, part of Oceania, and central-southeastern parts of South America. During DJF the main biases are related to double-ITCZ, strengthening of SPCZ, and deepening of the Aleutian and Icelandic low pressures. The simulation of a warmer SST on the eastern of most austral oceans, the strengthening (weakening) of the Subtropical (Polar) Jet over the Southern Hemisphere, and the weakening of the zonal circulation near the Antarctic continent are also found in both seasons. Over the central-eastern Equatorial Pacific a cooler bias in SST is found during JJA. These biases are interpreted by analyses of the simulated global mean-state and their impact on the main patterns of variability.
It is fair to say that the Ocean-Atmosphere Global Climate Models (OAGCMs) have become indispensable tools for the climate sciences. Despite the complexity of the climate system, many efforts have been made to improve the climate modeling in recent years. The Climate Forecast System version 2 (CFSv2) model is one example of such progress whose hindcasts and real-time operational forecasts have been provided by the National Centers for Environmental Prediction (NCEP) since March 2011. The hindcasts (reforecasts) are designed to test the models; that is, inputs from the past climate are used to the forecasts and allow evaluating how well the predictions approach to the observed climate. CFSv2 is initialized by the CFS Reanalysis (CFSR) that cover the period from 1979 to present and the main characteristics are described by Saha et al. [
The hindcasts performance can be assessed by many metrics and some recent papers have focused on assessing the global CFSv2 ability. For the 1982–2009 period Yuan et al. [
The representation of the intraseasonal variability in CFSv2 is described by Weaver et al. [
The design of the intraseasonal, 9-month reforecasts, and the real-time operational forecasts in terms of skill among CFSv2 and its predecessor is described by Saha et al. [
The present study examined the area-average skill over the continents, the interannual variability, the global-mean state, and the main patterns of variability over the Equatorial Pacific and extratropics in both hemispheres produced by the CFSv2 model. The analyses compare the CFSv2 hindcasts with the observations and reanalyses. Emphasis is placed on a preliminary discussion of dynamical reasons for the estimated biases of CFSv2 hindcasts.
The methodology used here differs from previous studies that focus more specifically on some regions of the globe or do not show the global linkage between the simulated ocean-atmospheric circulations. In our analyses the emphasis is on Dec-Jan-Feb (DJF) and Jun-Jul-Aug (JJA) seasons of the 1983–2010 period. The area-average skill and the interannual variability of the hindcasts over the continents for 0–3 month-lead times and the global-mean spatial biases for the 0-month lead are accessed. Also, the nonlinear sign of the interannual variability in total precipitation is investigated by the difference between El Niño (EN) minus La Niña (LN) events based on the departure of the neutral events. Finally, a comparison of the main patterns of variability over the Equatorial Pacific and the extratropics in both hemispheres is done by applying the Empirical Orthogonal Function (EOF). This methodology was motivated by the need to extend the evaluation of the CFSv2 to a global analysis whereas the previous evaluation studies are focused on specific areas or do not focus on a discussion of the ocean-atmosphere interaction around the globe.
Our methodology provides important preliminary information essential for many users that need to properly interpret their seasonal forecasts. It should be emphasized that the present analysis does not focus on a comparison between the features of CFSv2 and its predecessor performance or even with previous models. The paper is organized as follows: in Section
The main technical information about the CFSv2 is described in
The atmospheric component of CFSv2 is the Global Forecast System (GFS) model [
The present analysis is based on Dec-Jan-Feb (DJF) and Jun-Jul-Aug (JJA) seasons from 1983 to 2010 CFSv2 hindcasts, analyses of observations, and reanalysis dataset. In order to compute the seasonal means for the 0-month lead time, a 24-member ensemble comprises the runs initiated in the six different days of a given month (starting on the first day and each five successive days of the month) and for the four times of each day (00, 06, 12, and 18 UTC). All runs start on the same month of each season. Specifically, mean fields for DJF (JJA) were obtained with CFSv2 hindcasts initialized at 00, 06, 12, and 18 UTC of 1st December (June). For the 1-, 2-, and 3-month lead times all the runs started on the month of Nov, Oct, and Sep (May, Apr, and Mar) for the target DJF (JJA) season, respectively.
For the CFSv2 hindcasts validation, we use the following global monthly mean observational and reanalysis datasets: total precipitation (
The most adequate procedure to validate the model outputs is being compared with true observation based data. However, the comparison with reanalyses data is also justified since there are no global data covering a long time period available for many variables. Besides, the reanalyses are a reasonable approximation of the real state of the atmosphere. CMAP data set is derived from a combination of rain gauge observations and satellite estimates and it is widely used in many studies based on regional (Silva and Mendes [
For both DJF and JJA seasons we compute the area-average of the hindcasts (considering the lead times 0, 1, 2, and 3 months) for the total precipitation and air temperature at 2 m over each continent. The hindcasts skill is obtained based on the anomaly correlations as a measure of interannual variability. These area-average anomalies hindcasts for precipitation and air temperature at 2 m were compared with the CMAP and ERA-Interim data, respectively. The four lead times were analyzed to examine whether closer initial conditions can improve the forecasts performance. For the 0-month lead time the spatial pattern of the ensemble global-mean of the hindcasts is compared with the CMAP, R2, and SST datasets for each season. This constitutes an important measure of the hindcasts quality.
To facilitate the results visualization, the total precipitation was normalized to the interval
A posterior normalization to the interval
To find
Note that now
The main mode that explains most part of the large scale climate variance in the tropics refers to the El Niño-Southern Oscillation (ENSO) [
Through Empirical Orthogonal Function analysis (EOF, [
The anomaly correlations between the area-average ensemble-mean hindcasts over the continents and their corresponding verifications dataset are shown for the target seasons of DJF (Table
Anomaly correlations between the CFSv2 hindcasts and the corresponding total precipitation from CMAP and temperature at 2 m from ERA-Interim over the continents as a function of target DJF season from 1983 to 2010 period for 0- to 3-month lead times.
Lead times | North America | South America | Europe | Africa | Asia | Oceania | Antarctica | |||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Total prec. | Temp 2 m | Total prec. | Temp 2 m | Total prec. | Temp 2 m | Total prec. | Temp 2 m | Total prec. | Temp 2 m | Total prec. | Temp 2 m | Total prec. | Temp 2 m | |
l0 | 0,5 |
|
|
|
|
0,4 | 0,3 |
|
0,3 | 0,5 |
|
|
0,2 | 0,5 |
l1 | 0,3 | 0,5 |
|
|
0,4 | 0,4 | 0,1 |
|
−0,1 | 0,3 | 0,3 | 0,4 | 0,1 | 0,3 |
l2 | 0,1 | 0,5 | 0,5 |
|
0,5 | 0,1 | 0,1 |
|
−0,1 | 0,3 | 0,2 | 0,5 | 0,0 | 0,1 |
l3 | 0,3 | 0,5 | 0,5 |
|
0,3 | 0,3 | −0,1 |
|
−0,2 | 0,4 | 0,3 | 0,5 | 0,1 | 0,4 |
The higher correlations are in bold.
As in Table
Lead times | North America | South America | Europe |
Africa |
Asia |
Oceania |
Antarctica |
|||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Total prec. | Temp 2 m | Total prec. | Temp 2 m | Total prec. | Temp 2 m | Total prec. | Temp 2 m | Total prec. | Temp 2 m | Total prec. | Temp 2 m | Total Prec. | Temp 2 m | |
l0 | 0,5 |
|
|
|
0,3 |
|
0,3 |
|
|
|
|
|
−0,2 | 0,2 |
l1 | 0,4 |
|
0,3 |
|
0,5 |
|
0,3 |
|
|
|
|
0,5 | −0,1 | −0,1 |
l2 | 0,4 |
|
0,3 |
|
0,2 |
|
0,4 |
|
0,3 |
|
|
0,5 | 0,0 | 0,2 |
l3 | 0,4 |
|
0,3 |
|
0,3 |
|
0,3 |
|
0,4 |
|
|
0,4 | 0,1 | 0,0 |
Comparing both seasons in average JJA exhibits higher skills independent of the continent, month-lead time, or variable. For DJF (Table
As in DJF, for JJA, the Antarctic continent exhibits the worst skill (Table
For all lead times, Figures
Interannual variability of anomaly hindcasts of CFSv2 at lead 0 (red), 1 (orange), 2 (blue), and 3 (green) for DJF season from 1983 to 2010. The corresponding dataset in gray are on the left column the total precipitation from CMAP; on the right column the temperature at 2 m from ERA-Interim-ERAI.
As in Figure
CFSv2 has a large ability in reproducing the observed interannual variability of precipitation for JJA over North America, Asia, and Oceania (Figures
The precipitation pattern is a complex field determined by the global water cycle in association with the behavior of many factors such as moisture distribution over the continents, thermodynamics, and dynamical aspects like the SLP pattern, among others. Figure
Mean total CFSv2 precipitation ((a) and (c)) and differences from CMAP ((b) and (d)) during DJF ((a) and (b)) and JJA ((c) and (d)) at 0-month lead. The climatology period is 1983–2010. Shaded interval is 2 mm/day from 1 mm/day in (a) and (c) and 1 mm/day from 0.5 mm/day in (b) and (d).
For JJA, the main observed (figure not shown) precipitation action centers are also simulated by CFSv2 (Figure
A final comment is that although there are some intensity biases, the latitudinal displacement of ITCZ towards the summer hemisphere is fairly realistic in CFSv2. In both DJF and JJA seasons, the ensemble-mean hindcast simulates a secondary maximum of wet bias between 40°S and 60°S. This indicates a large activity of transient low pressure systems in the Southern Hemisphere storm tracks region.
For the annual mean, Wang et al. [
In the following analyses we investigate the large-scales oceanic and atmospheric biases that may be connected with the precipitation biases in Figure
SST is a variable used in many climate discussions. The main spatial pattern of the ensemble-mean simulated SST during DJF and JJA (Figures
Sea surface temperature (SST) from the CFSv2 ((a) and (c)) and differences from observation ((b) and (d)) during DJF ((a) and (b)) and JJA ((c) and (d)) at 0-month lead. The climatology period is 1983–2010. Contour interval is 5°C in (a) and (c) and 0.5°C in (b) and (d).
In Figure
Figure
In Figure
Mean CFSv2 sea level pressure ((a) and (c)) and differences from R2 ((b) and (d)) during DJF ((a) and (b)) and JJA ((c) and (d)) at 0-month lead. The climatology period is 1983–2010. Contours interval is 4 hPa in (a) and (c) and 1 hPa in (b) and (d). Negative contours are dotted.
During DJF (Figure
In the ensemble-mean hindcast of CFSv2 the semipermanent high pressures over the South Atlantic, South Pacific, and Indian Oceans are weaker than in R2 (Figure
Although the configuration of the simulated oceanic high pressure centers over the Northern Hemisphere during JJA (Figure
The wind-stress represents the horizontal force of the near surface wind on the sea surface. This variable is a measure of the vertical transfer of horizontal momentum from the atmosphere to the ocean (negative momentum flux) and it is used as a boundary condition in the oceanic model. We investigate the wind-stress instead of the wind vector at low levels to illustrate the sensitivity of the oceanic model component in CFSv2 to the atmospheric forcing.
Over the North Hemisphere, R2 wind-stress for DJF shows strong cyclonic circulations over middle and high latitudes in the North Pacific and North Atlantic Oceans which are related, respectively, to the Aleutian and Icelandic lows (figure not shown). The climatological northeastern wind-stress vectors with 0.15 N/m2 of magnitude over the Equatorial Atlantic Ocean occur in response to the predominant northeastern surface winds transporting moisture from the sea to the South America (figure not shown). Other important features are the convergence in ITCZ and the southeast flow related to the meridional upwelling near western coasts in the Southern Hemisphere (figure not shown). Strong northeasterly wind-stress predominates over the whole northern part of western-northern Pacific (from the equator to 20°N). For JJA in R2 the mean anticyclonic circulation over the Indian Ocean reflects the persistence of the surface winds associated with the summer Indian monsoon (figure not shown). In this season, over the northern Equatorial Pacific the activity of northeasterly trade winds is displaced to the central basin, while southeasterly trade winds in the western of the basin result from intense monsoon activity.
CFSv2 captures the strong seasonal variation of the wind-stress between DJF and JJA (Figures
As in Figure
During DJF, the most pronounced differences between CFSv2 and R2 are found in middle and high latitudes on the North Pacific and North Atlantic basins (Figure
Over the Equatorial North Atlantic the southwesterly bias during DFJ (Figure
During JJA it is noted that wind-stress patterns of CFSv2 are generally similar to R2 over the northern parts of the Pacific and Atlantic basins (Figure
Over the western South Pacific Ocean around 20°S in JJA the wind-stress vector biases indicate an underestimation of the atmospheric forcing in the hindcast that do not explain the cold bias in SST. In both DJF and JJA seasons the simulated wind-stress shows smaller magnitude than R2 in the circumpolar latitudes of the Southern Hemisphere. Dynamically, this indicates that in CFSv2 the atmosphere is transferring less momentum to the surface implying in the weakening of the Polar Jet. The analysis of the geopotential height mean state at 500 hPa and the zonal wind at 200 hPa will be helpful in the interpretation of these biases.
The 500 hPa geopotential height field provides important information about the main action centers of troughs and ridges over the globe and it may be useful in the analysis of the CFSv2 efficiency to simulate the upper atmospheric counterparts of surface cyclones and anticyclones. During DJF, the R2 climatology of the 500 hPa geopotential height indicates that the major troughs are found over Eastern Europe, northeastern Asia, and northern North America, whereas the ridges are located over Russia, western coast of North America, and eastern Atlantic (figure not shown). Figure
As in Figure
For JJA, CFSv2 hindcast presents at 500 hPa a large-scale ridge covering the southern part of North America, the subtropical North Atlantic Ocean, and the north part of Africa (Figure
The zonal wind at 200 hPa allows an identification of the mean position of the upper-level westerly winds. The westerly jet streams are related to the surface temperature gradients and denote the amplification of the troughs and ridges. The simulated mean zonal wind at 200 hPa for DJF (Figure
As in Figure
Although the simulated upper-level zonal winds near the equator cover similar areas of R2 the magnitude is smaller, except over the central Pacific (Figure
CFSv2 minus R2 upper-level zonal wind (Figure
In this section we examined the interannual variability of the CFSv2 hindcasts related to ENSO events. In [
Figure
El Niño minus La Niña difference composites of anomalous total precipitation during DJF at 0-month lead time from (a) CFSv2 and (b) difference between CFSv2 minus CMAP. Differences are contoured every 1 mm/day from 0.5 mm/day.
An investigation on the CFSv2 capability in representing the spatial and temporal variability of the ENSO pattern is shown in Figure
El Niño-Southern Oscillation (ENSO) pattern captured from anomalous SST over 15°N–15°S and 120°–285°E during DJF at 0-month lead for (a) CFSv2; (b) Optimum Interpolation v2 SST (Obs.); (c) the corresponding EOF time series which the gray (black) line is related to Obs. (CFSv2). Contour interval is 0.5°C and negative contours are dotted.
The Northern and Southern Annular Modes (NAM and SAM, resp.) are the main extratropical patterns of low frequency variability and their existence is due to internal atmospheric dynamics in middle and high latitudes in both hemispheres. A thorough discussion about the annular patterns can be found in Thompson and Wallace [
Northern Annular Mode (NAM) captured from SLP anomalies over 20°–90°N during DJF season at 0-month lead for (a) CFSv2; (b) R2 data; (c) the corresponding EOF time series which gray (black) line is related to R2 (CFSv2). Contour interval is 1 hPa from 0.5 hPa and negative contours are dotted. (a) and (b) are adapted from IRI/LDEO Climate Data Library.
Southern Annular Mode (SAM) captured from mean geopotential height at 850 hPa over 20°–90°S during JJA season at 0-month lead for (a) CFSv2 and (b) R2 data; (c) the corresponding EOF time series which gray (black) line is related to R2 (CFSv2). Contour interval is 1 hPa from 0.5 gpm and negative contours are dotted. (a) and (b) are adapted from IRI/LDEO Climate Data Library.
The NAM is obtained from the first EOF mode of the SLP field for DJF. In the R2 data (Figure
The major spatial features related to the shape and orientation of SAM are not well captured by CFSv2 in comparison to R2 (Figures
A preliminary discussion regarding some of the important features of CFSv2 hindcasts over the globe is presented. We analyzed DJF and JJA during 1983–2010. The skill (anomaly correlation) of the ensemble-mean seasonal area-average over the continents and its interannual variability is investigated for four lead times. Further the nonlinear signal of ENSO over the globe, the spatial distribution of the global-mean state seasonal hindcasts, and its main patterns of variability on the tropics and extratropics in both hemispheres are examined.
The air temperature skill during the boreal summer does not present a clear dependency on the lead times used in the seasonal hindcasts, indicating a positive aspect of CFSv2. The hindcasts also have good ability in representing the positive air temperature trends in the interannual scale. Except for South America, the 2 m air temperature amplitude is smaller in CFSv2 than in ERA-Interim.
In both seasons CFSv2 has a higher skill at the 0-month lead time, with the largest biases occurring over North America, South America, and Oceania. Such feature is an important indicative that the skill and bias relationship in CFSv2 should be applied in relative rather than absolute terms. Delsole and Shukla [
The CFSv2 precipitation at the 0-month lead time exhibits similarities with CMAP even though large biases occur over the oceans. Improvements in CFSv2 were not enough to eliminate the double-ITCZ bias during DJF. This error is mainly associated with the zonally elongated SPCZ that is a common and persistent error of coupled models as reported in previous studies. In response to this error there is a strengthening of SPCZ and dry biases over part of the Equatorial Atlantic. Our results revealed that over the western Pacific (near Indonesia) the convergence of southeasterly and northeasterly wind-stress biases intensifies the convection in SPCZ and its zonal alignment favors cooler SST conditions in the western Pacific.
Part of the warm SST bias over eastern oceans during DJF and JJA could be a result of a slightly upwelling reduction due to the wind-stress bias. For JJA the SST cold bias over the central-equatorial Pacific may be related to a strengthening of ITCZ over the Equatorial Pacific in association with the wind-stress bias near 15°N in the central-eastern basin. More intense and persisting convection in ITCZ in this region inhibits the local solar warming leading to colder SST and also explains part of the negative SLP bias in these regions.
During DJF CFSv2 presents a cold bias in the troposphere, mainly over the central-eastern North Pacific, which would be forced by the atmosphere due to the wind-stress bias. The anticyclonic gyre in the wind-stress bias indicates a deeper Aleutian low in CFSv2 than in R2, contributing to wet biases over this basin. The bias in wind-stress is also connected with a colder CFSv2 SST in the western North Pacific Ocean due to the transport of cold waters from high latitudes.
Another remarkable feature in CFSv2 is the strengthening of the subtropical jet that leads to precipitation overestimations by the persistence of low pressures over subtropical and mid-latitude regions. The circumpolar zonal circulation around the Antarctic is weaker in CFSv2 than in R2 compromising the natural variability representation over the extratropics on the Southern Hemisphere.
Regarding the nonlinearity of EN minus LN events, CFSv2 shows large precipitation biases over the eastern South Africa and Oceania. The simulated ENSO pattern is in reasonable agreement with the observations, a result in agreement with Kim et al. [
Considering the reasonable ability of CFSv2 in representing the ENSO interannual variability it is possible to suggest the use of the first EOF time series as explanatory variables in the transfer function downscaling approach. Schubert and Henderson-Sellers [
Regarding the CFSv2 extratropical variability, the errors in the simulated Aleutian low seem to degrade the simulated NAM, with CFSv2 simulating larger percentage of the explained variance (31%) than R2 (26%). The EOF time series of NAM presents moderate correlation with R2 due to the deficiency in the representation of Aleutian low pressure.
The wave pattern associated with SAM is not well reproduced by CFSv2. The model simulates two large geopotential height centers over the Pacific and the anomalies have opposite sign compared with R2 over the tropical Indian and Atlantic. Such deficiency partly explains the biases in the upper-level zonal wind over the tropics and mid-latitudes of the Southern Hemisphere. By comparing the mean-state atmospheric bias presented here a barotropic structure associated with the strengthening (weakening) of the subtropical (polar) jets in CFSv2 is noted.
The major spatial features related to the shape and orientation of SAM are not properly captured by CFSv2 in comparison to R2. The zonal wavenumber in CFSv2 is smaller than in R2. According to Lefebvre et al. [
Overall evaluations show that although there are large improvements in CFSv2 further investigations are still needed. The model skill and biases identified here are essential for further investigations related to impacts on the prediction skill pattern. Thus, detailed investigations should be carried out to help in understanding in further details the reasons for the CFSv2 deficiencies shown here. A special emphasis on surface fluxes is also important. Furthermore the applications of statistical techniques are required as complementary tool for ensemble improvements as in [
The authors declare that there is no conflict of interests regarding the publication of this paper.
The authors thank the editor and the three anonymous reviewers whose constructive comments help us improve the paper. Also, they thank NCEP for providing the observational, reanalyses, and modeling data and the ECMWF for their reanalysis. Gyrlene A. M. Silva would like to thank Michael Bell from the International Research Institute for Climate and Society (IRI) for the help in computing the EOF analysis and the Group of Climate Studies from University of São Paulo (GrEC/USP) for providing the physical locations for the model dataset preparation. The authors also thank the INterdisciplinary CLimate INvestigation cEnter (INCLINE/USP) for the support received. Tércio Ambrizzi also acknowledges São Paulo Research Foundation (FAPESP) (Proc. No. 08/58101-9) and National Council for Scientific and Technological Development (CNPq) for partial support.