Abstract

The climate trend in a dynamical seasonal forecasting system is examined using 33-yr multimodel ensemble (MME) forecasts from the second phase of the Canadian Historical Forecasting Project (HFP2). It is found that the warming trend of the seasonal forecast in March–May (MAM) over the Eurasian continent is in a good agreement with that in the observations. However, the seasonal forecast failed to reproduce the observed pronounced surface air temperature (SAT) trend in December–February (DJF). The possible reasons responsible for the different behaviors of the HFP2 models in MAM and DJF are investigated. Results show that the initial conditions used for the HFP2 forecast system in MAM have a warming trend over the Eurasian continent, which may come from high-frequency weather systems, whereas the initial conditions for the DJF seasonal forecast do not have such a trend. This trend in the initial condition contributes to the trend of the seasonal forecast in the first month. On the other hand, an examination of the lower boundary SST anomaly forcing shows that the SST trend in MAM has a negative SST anomaly along the central equatorial Pacific, which is favorable for a positive phase of the North Atlantic Oscillation atmospheric response and a warming over the Eurasian continent. The long-term SST trend used for the seasonal forecast in DJF, however, has a negative trend in the tropical eastern Pacific, which is associated with a Pacific–North American pattern–like atmospheric response that has little contribution to a warming in the Eurasian continent.

1. Introduction

The global temperature has experienced a significant warming trend since the early twentieth century, especially for the last several decades, which is likely to be a response to anthropogenic greenhouse gases (GHG) and aerosols, lower boundary conditions (e.g., SST, sea ice, etc.), or the climate system's natural internal dynamics (Stott et al. 2000; Johannessen et al. 2004). Previous studies have shown that it is important to increase the GHG concentration in the long-term model simulations to have a realistic climate trends (e.g., Meehl et al. 2004). Dynamical seasonal forecasts have been considered as a boundary condition problem. Their forecast skill mainly depends on whether the boundary forcings, such as the SST, can generate strong enough signals that can be recognized from the unpredicted atmospheric internal variability. In a standard seasonal forecasting system that only integrates for several months, the GHG is usually fixed and assumed to have a rather small impact. Even so, there are still climate trends in the seasonal forecasts output due to trends in SST forcing, or due to trends in the initial conditions.

Recently, the importance of including a realistic GHG in the seasonal forecast has been investigated (Luo and Behera 2011; Kharin et al. 2012; Cai and Shin 2009), whereas there is also debate about the influence of the GHG on seasonal forecasts. For example, Liniger et al. (2007) investigated the improvement of seasonal forecasts by including realistically varying GHG concentrations using a coupled atmosphere–ocean model. However, no clear improvement can be found in predicting the interannual variability. Zhao and Dirmeyer (2004) also found little impact for the atmospheric seasonal simulations when increasing the GHG concentrations. However, using a coupled general circulation model, Doblas-Reyes et al. (2006) examined the impact of GHG concentrations on seasonal forecasts. Two sets of experiments were performed with one using annually updated and the other using fixed GHG concentrations. Results showed that the former displays more realistic variability of temperature and has a better forecast quality that is due to a better simulation of climate trends. Cai and Shin (2009) analyzed the SAT of the retrospective seasonal climate prediction made by the National Centers for Environmental Prediction (NCEP) Climate Forecast System with the GHG fixed at the 1988 concentration level. Their results suggested that an adequate representation of the anthropogenic GHGs in coupled climate models for seasonal forecasts was essential for more accurate seasonal climate predictions, even at the regional level.

Boer (2009) examined the recent climate trends in the seasonal forecast of four numerical models from the second phase of the Historical Forecasting Project (HFP2). The long-term climate trends of the multimodel ensemble (MME) mean seasonal forecasts are compared to those in the observations for summer and winter. Results showed that the trends in the MME mean forecasts are much weaker than those in the reanalyses from the NCEP–National Center for Atmospheric Research (NCAR; Kalnay et al. 1996) and could be partially corrected using a posterior approach with the most significant improvement of the forecast skill in December–February (DJF) and over the Eurasian continent. The mechanisms accounting for the missing climate trends in the HFP2 seasonal forecasting system remain unclear. One possible explanation, as Boer argued, is that although the atmospheric initial conditions and oceanic boundary forcings used in the forecasting system contain GHG and aerosol forcing information, the forecasts were conducted without the support of radiative forcing, which can cause errors in the forecasts over land.

In this study, the HFP2 MME mean seasonal forecasts are further analyzed to explore the possible reasons accounting for the misrepresented long-term climate trends in the HFP2 seasonal forecasting system. Possibly, this analysis will help to assess the relative contributions to the long-term climate trends from different sources, such as the initial conditions, the SST boundary forcings, and/or the GHG. The study in this paper is organized as follows. The data and models are described in section 2. Section 3 presents the long-term climate trends of the SAT in the observations and in the HFP2 MME mean seasonal forecasts. In section 4, the possible reasons accounting for the underestimation of the observed climate trends in the forecast are investigated and followed by the conclusion and discussions in section 5.

2. Data and models

The Canadian Climate Variability Research Network has conducted a project termed the Historical Forecasting Project (HFP). The project aimed to test how well seasonal-mean atmospheric conditions could be predicted using dynamical general circulation models in an operational environment and to produce a sequence of forecasts for many years in order to obtain statistics on the model performance (Derome et al. 2001). In the second phase, HFP2, the seasonal forecast experiments were performed using four numerical models with fixed greenhouse concentrations. The models are the second and the third generations of the general circulation models (GCM2 and GCM3), used for climate simulations, which were developed at the Canadian Centre for Climate Modeling and Analysis (CCCma; Boer et al. 1984; McFarlane et al. 1992). The other two models are reduced-resolution versions of the global spectral model (SEF; Ritchie 1991) and the Global Environmental Multiscale model (GEM; Côté et al. 1998a,b), used for medium-range weather forecasting, developed at Recherche en Prévision Numérique (RPN) in Montréal, Canada.

For each model, ten 4-month forecasts were carried out starting from the first day of each month. The initial conditions were from the 12-h interval preceding the first day of the expected forecast season. For the initial conditions, the NCEP–NCAR reanalyses were used. The SST forcing used for the seasonal forecasts was the sum of the SST anomaly of the month prior to the forecast period, persisted through the forecast period, and the monthly varying climatological SST. The SST and ice data were taken from the Seasonal Prediction Model Intercomparison Project-2 (SMIP-2) boundary data to understand the numerical models' capability in capturing the interannual variability of the atmosphere. The standard deviation of the surface air temperature (SAT) for each member of the HFP2 seasonal forecasts is examined and compared to the observations (not shown). Results show that the HFP2 forecasts can capture the basic structure of the standard deviation of the observed SAT while with relatively weaker amplitude.

In this study, we only make use of the MME mean forecasts spanning the 33 years from 1969 to 2001. Four seasons, March–May (MAM) for spring, June–August (JJA) for summer, September–November (SON) for fall, and December–February (DJF) for winter are investigated. The variables employed include the SAT, the 500-hPa geopotential height (Z500), and the precipitation. The data used for the verification of the forecasts are the Climate Research Unit monthly time series, version 2.1 (CRU TS 2.1) dataset, a set of monthly averaged observed SAT over the land surface from CRU at the University of East Anglia, United Kingdom (Mitchell and Jones 2005; http://www.cru.uea.ac.uk/cru/data/hrg.htm). The climate trend we investigated in this study is the linear trend component represented as T = at, where t is the time anomaly and the rate of change a in the trend component is estimated by fitting a linear trend to the data. The significance of the trend is examined using Mann–Kendall trend test in this study.

3. Long-term SAT trends in the observations and in the MME mean forecasts

We start by examining the long-term climate trends of SAT in the observations over the Northern Hemisphere. Figure 1 shows the spatial structures of the 33-yr (1969–2001) linear trends of the continental SAT in the CRU data for four seasons. The linear trends at the 5% significant level are shaded. Although cooling trends can be seen over some local area, the long-term climate trends in SAT during this period are dominated by pronounced warming for all the four seasons. The largest warming appears over mid- to high-latitude continents in MAM and DJF with the maximum values exceeding 5°C in some area. In DJF, pronounced positive SAT trends extend from the Canadian northwest to the East Coast of the United States and over most part of the Eurasian continent. The distribution of the warming trends in MAM has many similarities to that in DJF while with relatively weaker magnitudes. Climate trends of SAT are not distinct in JJA over North America, while moderate warming can be found over the Eurasian land areas. The weakest climate trends among the four seasons appear in SON. The SAT trends of JJA and DJF are consistent with those shown in Boer (2009, see their Fig. 1). The climate trends shown above are the combined results of both the external forcings such as GHG, SST–sea ice, and atmospheric internal processes.

Fig. 1.

The climate trend in CRU SAT (°C decade−1) for (a) MAM, (b) JJA, (c) SON, and (d) DJF. The contour interval is 0.4°C. In this and appropriate following figures, shading represents a significance level of 0.05 according to the Mann–Kendall trend test.

Fig. 1.

The climate trend in CRU SAT (°C decade−1) for (a) MAM, (b) JJA, (c) SON, and (d) DJF. The contour interval is 0.4°C. In this and appropriate following figures, shading represents a significance level of 0.05 according to the Mann–Kendall trend test.

As we mentioned above, although the GHG is fixed in the HFP2 seasonal forecasting system, there still exist climate trends in the seasonal forecasts output due to trends in SST forcing, or due to trends in the initial conditions. To evaluate the performance of the HFP2 models in capturing the observed long-term SAT trends the same analysis has been applied to the HFP2 MME mean forecasts for the same period (Fig. 2). It should be noted that the interval in Fig. 2 is only half of that in Fig. 1 for the purpose of a better presentation. It depicts that the largest warming trends of the HFP2 MME seasonal forecast appear in MAM and the overall magnitudes of the SAT trends in this season are comparable to those in the observations. In DJF, warming trends can only be found over North America with a warming pattern similar to its counterpart in the observations. The warming trends are seen to be quite weak in JJA and SON. Apparently, the HFP2 MME mean forecasts considerably underestimate the observed warming trends during this period. Also noticed are the negative climate trends of SAT over northwestern Eurasian in DJF, which will be discussed later and that are, at least partly, related to the initial conditions of the HFP2 model integrations.

Fig. 2.

As in Fig. 1, but for the four-model-averaged SAT. The contour interval is 0.2°C.

Fig. 2.

As in Fig. 1, but for the four-model-averaged SAT. The contour interval is 0.2°C.

It is also noted that, over the Eurasian continent, the GCMs perform quite differently in capturing the SAT trends for MAM and DJF. In MAM, the observed warming trends over there are reasonably reproduced by the HFP2 MME mean forecasts. However, in DJF, the most pronounced warming trends in the observations are almost absent in the HFP2 MME mean forecasts. To see whether or not the underestimation of the observed long-term SAT trends in the seasonal forecasts in DJF are model dependent we also examine the spatial structures of the SAT trends for the four numerical models used in the HFP2 separately (not shown). Results indicate that no single model can capture the pronounced observed Eurasian warming trend over there in DJF. The long-term climate trends of SAT over the Eurasian continent for MAM and DJF are also estimated by averaging the SAT anomalies over the Eurasian region (20°–75°N, 15°E–170°W; Fig. 3). The values are constructed as an area-weighted average of all the grid points over the continents. The area-averaged SAT anomalies of the CRU SAT and the HFP2 MME mean forecasts are represented by solid and dotted curves, respectively. The CRU SAT displays a distinctive warming tendency in both MAM and DJF with SAT trends of 0.57° and 0.68°C over the examined 33 years, respectively. The MME mean forecasts clearly underestimate the observed warming trends in DJF with a trend of only 0.25°C in the 33 years, while a much better estimate can be found for MAM with a trend up to 0.46°C. The differences of the performances of the HFP2 models in representing the climate trends over the Eurasian land areas in these two seasons are obvious. In the following, we will focus on a comparison of the SAT in MAM and DJF, especially over the Eurasian continent.

Fig. 3.

The 33-yr SAT anomalies' area-weighted average over the Eurasian continent for (a) MAM and (b) DJF. The SAT anomalies of the CRU SAT and the average of 4 GCMs are represented by solid and dotted curves, respectively.

Fig. 3.

The 33-yr SAT anomalies' area-weighted average over the Eurasian continent for (a) MAM and (b) DJF. The SAT anomalies of the CRU SAT and the average of 4 GCMs are represented by solid and dotted curves, respectively.

4. The influence of the climate trends to the seasonal forecast skill of SAT

In the last section, we have examined the long-term climate trends of SAT in the observations and in the HFP2 ensemble forecasts. One challenge in dynamical seasonal forecasting is to figure out how to use the long-term climate trends to improve the seasonal forecast skill. However, first we need to understand the importance of the long-term climate trends to the dynamical seasonal forecasting system. More specifically, the question we want to know is the following: To what extent can the seasonal forecast skill potentially be improved if the GCMs can capture the observed long-term climate trends?

To understand the importance of numerical models in capturing the long-term climate SAT trends, the HFP2 MME seasonal forecast skill for MAM and DJF measured by the temporal correlation coefficients (TCC) of the SAT between the forecasts and the observations are first examined. The TCC score of the MME mean SAT forecasts for MAM and DJF are illustrated in Figs. 4a and 4b, respectively. Areas with a correlation score significant at the 5% level or better, are shaded. As is seen, in MAM, the MME forecasts have significant skills over most of the North American and Eurasian continents except some areas over central-eastern Asia. In DJF, the TCC score are generally less pronounced than that in MAM. Most of North America and the extratropical Eurasian continent are also covered by significant TCC, while little forecast skill can be found over the mid- to high-latitude Asian continent. The corresponding TCC score for the detrended SAT are illustrated in Figs. 4c and 4d for MAM and DJF, respectively. The TCC scores are generally decreased after the linear trended removed from the forecasts. Compared to the original forecast skill, the TCC over northeast North America in DJF and high-latitude Eurasian continent are obviously decreased, which is apparently the effect of long-term climate trends of SAT.

Fig. 4.

The TCC forecast skill of the HFP2 SAT seasonal forecasts for (a) MAM and (b) DJF and the TCC score of the detrended HFP2 SAT forecasts for (c) MAM and (d) DJF. The shaded areas represent correlations with a significance level of 0.05 according to a Student's t test.

Fig. 4.

The TCC forecast skill of the HFP2 SAT seasonal forecasts for (a) MAM and (b) DJF and the TCC score of the detrended HFP2 SAT forecasts for (c) MAM and (d) DJF. The shaded areas represent correlations with a significance level of 0.05 according to a Student's t test.

Our purpose here is to see by how much the seasonal forecast skill of SAT can be improved if we assume that the GCMs can totally capture the observed long-term climate trends. To achieve this object, a new set of SAT seasonal forecasts are constructed by adding the observed SAT linear trends to the detrended HFP2 seasonal forecasts. The process is done separately for each GCM and the new ensemble forecasts are obtained by averaging the new constructed SAT forecasts. To see how much improvement of the SAT forecast skill has been achieved by the addition of the observed long-term SAT linear trends, the TCC score of the new constructed SAT forecasts and the difference of the TCC between the original SAT forecasts and the new constructed SAT forecasts are calculated and depicted in Fig. 5. Note that the interval is changed to 0.1 in Figs. 5c and 5d for the purpose of a better representation.

Fig. 5.

As in Fig. 4, but for the new constructed SAT (see details in the text) for (a) MAM and (b) DJF and the differences of TCC between the original SAT and new constructed SAT for (c) MAM and (d) DJF. The difference of TCC scores > 0.1 are shaded in (c) and (d).

Fig. 5.

As in Fig. 4, but for the new constructed SAT (see details in the text) for (a) MAM and (b) DJF and the differences of TCC between the original SAT and new constructed SAT for (c) MAM and (d) DJF. The difference of TCC scores > 0.1 are shaded in (c) and (d).

Improvements of the SAT forecast skill can be found over some sparse area in MAM, although not very pronounced. The correlation score for the new constructed SAT ensemble forecasts that contain the observed long-term climate trends is significantly improved over a large area of the Eurasian continent for DJF. Significant TCC score are found covering most of Eurasian continent, including the central-eastern Asia. To determine each of the models' performances in this process, the results for all GCMs involved in HFP2 were examined separately (not shown). All four models provide consistent results allowing conclusions similar to those obtained above to be drawn. Again, the correlation scores of the new constructed SAT are better than those of the original model forecasts and the largest improvement is found over the Eurasian continent. The above results illustrated that DJF is the season and the mid- to high-latitude Eurasian continent is the region that benefits most significantly if the GCMs can improve their ability in representing the observed long-term SAT trends.

5. Reasons for the underestimation of the climate trends of SAT in the forecasts

a. Long-term SAT trends in the first and in the two–three months of MAM and DJF

As we mentioned above, the pronounced observed warming trends in DJF are largely missing in the HFP2 MME mean forecasts over the Eurasian continent whereas the GCMs perform reasonably well in representing the SAT trends over there in MAM. In the following, we will further explore the possible reasons why the HFP2 models perform differently in capturing the climate trends in MAM and DJF over the Eurasian land areas. Most importantly, we want to investigate the contributions of the missing SAT trends over the Eurasian continent in DJF from different sources.

Climate trends in seasonal forecasts with a zero-month lead time can come from both the initial conditions and the slowly changing lower boundary conditions. The initial conditions can influence the seasonal forecasts over approximately the first two to three weeks whereas the lower boundary forcing plays a role mainly during the rest of the forecast period. To separate the contributions to the long-term SAT trends in the HFP2 MME mean forecasts from the initial conditions and the lower boundary conditions the trends should be examined separately for the first month and for the average of the second and the third month forecasts. The CRU SAT are first examined for the purpose of a comparison to the HFP2 MME mean forecasts. The 33-yr climate trends of the CRU SAT for March and December are shown in Figs. 6a and 6b, while those for April–May and January–February are displayed in Figs. 6c and 6d, respectively. As is seen from Fig. 6a, pronounced warming trends appear over mid- to high-latitude continents in March with weak cooling trends lying over some lower latitudes such as North Africa. The spatial distribution of the SAT trends in April–May is similar to that in March, but with much weaker amplitudes. For the first month of wintertime (Fig. 6b), cooling trends are noticed over the high-latitude Eurasian continent along the north polar region, while warming trends can be seen over central Eurasian and the high-latitude North American continent. The warming trends in January–February are much more enhanced and positive SAT trends can be seen over most part of the continent (Fig. 6d). The above results indicate that in spring, the pronounced warming trends appearing in Fig. 1a get more contributions from March than from the following two months whereas in wintertime the trends in January–February are stronger than December, especially over high-latitude Eurasian continent.

Fig. 6.

As in Fig. 1, but for the climate trends of the CRU SAT (°C decade−1) for (a) March, (b) December, (c) April–May, and (d) January–February. The contours interval is 0.4°C.

Fig. 6.

As in Fig. 1, but for the climate trends of the CRU SAT (°C decade−1) for (a) March, (b) December, (c) April–May, and (d) January–February. The contours interval is 0.4°C.

The climate trends of SAT in the HFP2 MME mean forecasts for March and December are depicted in Figs. 7a and 7b and those for April–May and January–February are displayed in Figs. 7c and 7d, respectively. For springtime, as can be seen from Figs. 7a and 7c, although with relatively weaker amplitudes, the HFP2 MME mean forecasts capture the basic structure of the SAT warming trend in the observations. The long-term SAT trend shown in Fig. 7a is obviously more pronounced than that in Fig. 7c over the mid- to high-latitude Eurasian continent indicating that the warming trend becomes weaker with time in the HFP2 MME forecasts. We need to keep in mind that even in the first month of the SAT forecasts, the SAT trends can also get some contribution from the lower boundary conditions while the influence of the initial conditions are mostly within the first two to three weeks. However, as the SAT trends shown in Fig. 7c is obviously weaker than Fig. 7a over the Eurasian continent, it is reasonabe to conclude that the SAT warming trends in HFP2 MME forecasts for MAM get more contribution from the initial conditions than the lower boundary conditions over this region.

Fig. 7.

As in Fig. 6, but for the HFP2 seasonal forecasts for (a) March, (b) December, (c) April–May, and (d) January–February. The contour interval is 0.2°C.

Fig. 7.

As in Fig. 6, but for the HFP2 seasonal forecasts for (a) March, (b) December, (c) April–May, and (d) January–February. The contour interval is 0.2°C.

The SAT trend pattern for December in the HFP2 MME forecasts share many similarities to its counterpart in the observations. The positive SAT trend over North America and the cooling trends over the high-latitude Eurasian continent appear in both the observations and the HFP2 MME mean forecasts. Some disagreement, however, can also be found. For example, the pronounced observed warming trend over the central-east Eurasian continent is not captured by the HFP2 MME seasonal forecasts during the first month. The forecast SAT trends in the following two months of wintertime, however, are found far from the observations. In contrast to the pronounced warming trends in the observations over the Eurasian continent, the warming trends in HFP2 SAT forecasts for January–February are almost absent over there. The above results indicate that, except for the positive SAT trend over the central-east Eurasian continent, the observed SAT trend in the first month of wintertime is reasonably simulated by HFP2 numerical models, which mainly get contributions from the initial conditions. However, the lower boundary conditions cannot provide enough forcing for the HFP2 seasonal forecasts to generate pronounced warming trends over the Eurasian continent for months 2–3 of the winter season.

b. Impact of initial conditions

As discussed before, for each GCM in the HFP2 seasonal forecasting system, 10 members of 4-month forecasts were carried out starting from the first day of each month. The initial conditions were taken from 10 of the 12-h intervals immediately preceding the first day of the forecast season. That is, the forecasts of MAM use the last 5 days of NCEP–NCAR reanalyses from February as the initial conditions while those from 26 to 30 November are used for the forecasts of DJF. The linear trends of SAT for the 5-day NCEP reanalysis during the 33-yr period for February and November are shown in Figs. 8a and 8b, respectively. The differences of the SAT trends for the 5-day-averaged SAT in the NCEP–NCAR reanalysis between Figs. 8a and 8b are quite obvious.

Fig. 8.

The observed climate trend for the 5-day averaged SAT (°C decade−1) in the NCEP–NCAR reanalyses for (a) February and (b) November. The contour interval is 1°C.

Fig. 8.

The observed climate trend for the 5-day averaged SAT (°C decade−1) in the NCEP–NCAR reanalyses for (a) February and (b) November. The contour interval is 1°C.

For the initial conditions used for MAM seasonal forecasts, a significant positive SAT trend and moderate warming can be found over high-latitude Eurasian and North American continents with a pattern similar to that in Fig. 7a, indicating the influence of initial conditions to the seasonal forecast in the first month. A weak cooling can also be found over the lower-latitude Eurasian continent, consistent with the cooling in Fig. 7a. For the DJF seasonal forecasts (Fig. 8b), a moderate initial warming trend can be found over high-latitude North America. However, over the Eurasian continent, a pronounced cooling SAT trend present over northwestern Russia, accounting for the significant negative SAT trends in the first month of the HFP2 MME mean forecasts as shown in Fig. 7b.

The above results illustrate that the spatial distribution of the SAT trend in the initial conditions used for HFP2 seasonal forecasts share many similarities to those of the first month of the model integrations, indicating that the initial conditions play an important role for the first month trend of the model forecasts. The most significant differences of the SAT trends for the first month appear over the central-east Eurasian continent in DJF where a warming trend can be noticed in the observations whereas it is almost absent in the first month of the HFP2 seasonal forecast. The results indicate that the missing observed warming trends over Eurasia in DJF is caused, at least partly, by the fact that the initial conditions used for the HFP2 seasonal forecasts could not provide a warming forcing in the model integrations at the very beginning. The climate trends for other 5-day-averaged SAT reanalysis in November were also examined. It was found that the patterns of the SAT trends change substantially during the month, probably because of the fast changing traveling weather systems.

c. Roles of lower boundary conditions

We then examine the role of the lower boundary SST forcing used in the HFP2 seasonal forecasts, which is the dominant external source of a skillful forced response by the GCMs. As mentioned before, the SST anomaly forcing used in the HFP2 seasonal forecasts is from the month prior to the expected forecast season and persisted through the forecast period. Therefore, the SST anomalies forcing used for MAM and DJF forecasts are from the SST anomaly of February and November, respectively. The 33-yr climate trends of SST for February and November are shown in Figs. 9a and 9c, respectively. To know how realistic the SST anomaly forcing used in HFP2 forecasting system for MAM and DJF seasonal forecasts are the linear trend of SST for MAM and DJF are also examined and illustrated in Figs. 9b and 9d, respectively, for a purpose of a comparison. It shows that the long-term climate trends for SST in MAM and DJF are strikingly similar to those in Figs. 9a and 9c, respectively, indicating the reasonability of using the persistent SST anomaly forcing.

Fig. 9.

The 33-yr (1969–2001) SST climate trend of SST (°C decade−1) for (a) February, (b) MAM, (c) November, and (d) DJF. The contour interval is 0.15°C.

Fig. 9.

The 33-yr (1969–2001) SST climate trend of SST (°C decade−1) for (a) February, (b) MAM, (c) November, and (d) DJF. The contour interval is 0.15°C.

The long-term SST trend patterns for February and November are quite similar to each as well as their magnitudes. Warming trends of the SST can be seen along the coast of continents, over the eastern subtropical Pacific in the Southern Hemisphere and over the Indian Ocean for both Figs. 9a and 9c. Differences can be seen, however, over the equatorial Pacific, which is known to be a major forcing area for the atmospheric variability on a seasonal time scale. The long-term climate trends for the SST forcing in February reveal a negative SST anomaly centered in the middle equatorial Pacific, while the corresponding SST trend in November is a narrow band of cooling over the eastern tropical Pacific. The location of the negative SST trends forcing over the tropical Pacific is critical for global atmospheric teleconnections associated with it. For instance, to obtain the dominant atmospheric patterns associated with the tropical Pacific SST forcing in winter, a singular value decomposition (SVD) analysis was conducted by Lin et al. (2005) between the observed Z500 north of 20°N and the observed SST structures in the tropical Pacific and Indian Oceans (20°N–20°S, 120°E–90°W). The first two SVD modes in the atmospheric component are found to be the Pacific–North American (PNA) and North Atlantic Oscillation (NAO) patterns, respectively. The corresponding SST structure for the negative phase of PNA has many similarities to that shown in Fig. 9c and the SST forcing associated with the NAO is quite similar to that in Fig. 9a.

It is know that the SST anomaly in the equatorial ocean, for example, that is related to ENSO is closely related to changes in precipitation and diabatic heating that generate anomalous vertical motion and upper-level divergence leading to extratropical Rossby waves and global teleconnections (Wallace and Guztler 1981; Sardeshmukh and Hoskins 1988). It is, therefore, important for numerical models to have a realistic precipitation response over the tropical ocean to have a skillful seasonal forecast. To see how the HFP2 models respond to the SST anomaly forcing trends shown in Figs. 9a and 9c the long-term climate trends of the HFP2 seasonal precipitation forecasts over the tropics are examined and depicted in Figs. 10a and 10c for MAM and DJF, respectively. It can be seen that the long-term linear trend of the HFP2 MME precipitation forecasts for MAM are dominated by a significant dry anomaly over the western tropical Pacific. For the HFP2 seasonal forecast precipitation in DJF, a significant negative precipitation trend was found over central-eastern tropical Pacific. The differences of the location of the negative precipitation anomaly along the equator can significantly influence the atmospheric circulation over the extratropics as depicted in Jia et al. (2009). Using a simple atmospheric model, they examined the atmospheric response to the thermal forcing over different locations of equatorial Pacific. Results indicate that the negative thermal forcing over the western tropical Pacific is effective in generating an Arctic Oscillation (AO)-like, or NAO-like, atmospheric response while those over the central tropical Pacific are favorable for a negative PNA-like atmospheric response. To see how realistic the precipitation response in HFP2 MME seasonal forecasts is comparing to the real atmosphere, the long-term linear trend of the NCEP–NCAR precipitation for MAM and DJF during the same period are also examined and presented in Figs. 10b and 10d, respectively. Positive precipitation trends can be found in both the HFP2 MME forecasts and the observations while the magnitudes are somehow different. Although differences can also be found for the long-term precipitation trends over the tropical Pacific Ocean, the negative precipitation anomaly trends in the observations can be captured to a certain extent by the HFP2 MME forecasts.

Fig. 10.

The climate trends of precipitation (mm day−1 decade−1) in (a),(c) HFP2 MME forecasts and (b),(d) NCEP–NCAR reanalysis for (a),(b) MAM and (c),(d) DJF. The contour interval is 0.3 mm day−1 for (a),(b) and 0.5 mm day−1 for (c),(d).

Fig. 10.

The climate trends of precipitation (mm day−1 decade−1) in (a),(c) HFP2 MME forecasts and (b),(d) NCEP–NCAR reanalysis for (a),(b) MAM and (c),(d) DJF. The contour interval is 0.3 mm day−1 for (a),(b) and 0.5 mm day−1 for (c),(d).

To get the associated Z500 response to the tropical Pacific forcing trend we examine the climate trends of Z500 in the HFP2 MME mean forecasts for MAM and DJF and present in Figs. 11a and 11c, respectively. The long-term climate trends of HFP2 Z500 for MAM has a negative center around Greenland and positive anomalies along the midlatitude North Atlantic, which share many similarities to the NAO pattern over this region. Positive anomalies can be found over the North Pacific and northern Eurasia. The corresponding Z500 trend pattern for DJF is dominant by positive anomalies over the North Pacific and over the east coast of North America, which are consistent with the two centers of the PNA around these regions. Previous studies showed that the distribution of the NAO-like pattern can bring strong anomalous wind along the extratropical North Atlantic Ocean from west to east along the isobars that penetrates into the mid- to high-latitude Eurasian continent, leading to anomalous warm advection that may be responsible for a warming trends over the Eurasian continent, consistent with our results.

Fig. 11.

The climate trends of the Z500 (m decade−1) of (a),(c) HFP2 MME forecasts and (b),(d) NCEP–NCAR Z500 for (a),(b) MAM and (c),(d) DJF. The contour interval is 3 m for (a) and (c) and 5 m for (b) and (d).

Fig. 11.

The climate trends of the Z500 (m decade−1) of (a),(c) HFP2 MME forecasts and (b),(d) NCEP–NCAR Z500 for (a),(b) MAM and (c),(d) DJF. The contour interval is 3 m for (a) and (c) and 5 m for (b) and (d).

The long-term climate trends of Z500 in the NCEP–NCAR reanalysis are also examined and presented in Figs. 11b and 11d, respectively, for the purpose of comparison. Although differences are very obvious between the HFP2 Z500 forecasts and the observations, the trend pattern shown in Fig. 11b also has significant positive anomalies over extratropical North America, the western European, and high-latitude eastern Eurasian continent while a negative trend anomaly can be found over the polar region, consistent with that in Fig. 11a. The long-term Z500 trend in DJF in the observations is an AO-like distribution. The differences between the Z500 trends in the HFP2 forecasts and the observations are probably caused by the internal dynamics of the atmosphere. As we mentioned before, the HFP2 MME forecasts only represent the signal from the external forcings, while that of the observations are the result of both the boundary forcings and the internal dynamics of the atmosphere.

6. Summary and discussion

A set of 33-yr MME seasonal forecasts from HFP2 are used to investigate the ability of dynamical models in capturing the observed long-term climate trends of SAT. Results showed that the most pronounced warming trends appear in MAM and DJF over the mid- to high-latitude continents. However, the magnitudes of the climate trends of SAT in the HFP2 MME mean forecasts are much weaker than those in the observations in DJF especially over the Eurasian continent. In contrast, the SAT trends over there in MAM are reasonably well represented by the HFP2 MME mean forecasts. The possible reasons accounting for the different behaviors of the HFP2 models in capturing the warming trends over the Eurasian continent in MAM and DJF were investigated. The SAT trends of the initial condition used for the seasonal forecasts in MAM are pronounced warmings over Eurasian continent, which can contribute to warming trends in the first month of the seasonal forecasts. During the remaining two months of the seasonal forecasts in MAM, the SST boundary forcing, which has a cooling trend over the middle equatorial Pacific Ocean, is responsible for negative precipitation trends over the western tropical Pacific. This precipitation forcing is associated with a NAO-like atmospheric response over the North Atlantic region and favorable for warming advection to the Eurasian continent. For seasonal forecast in DJF, the SAT trend in the initial conditions appear to have a pronounced cooling over northwest Russia, consistent with negative SAT trends over there for the first month of HFP2 seasonal forecasts. However, the initial conditions for the DJF seasonal forecasts cannot provide a warming forcing over the central-east Eurasian continent where a pronounced positive SAT trend can be noticed in the observations. The long-term trend of the SST forcing for DJF seasonal forecast is a negative SST anomaly lying over the eastern equatorial Pacific, which corresponds to a dry precipitation trend over the middle-eastern tropical Pacific. This precipitation forcing is associated with a PNA-like atmospheric response over the Pacific–North American region and its influence to the Eurasian continent is quite weak.

We mentioned before that the SST anomaly forcing can influence the ensemble seasonal forecast mainly over the second and the third months in the HFP2 output. However, we should keep in mind that the initial conditions are to some extent influenced by the SSTs, so the trends in the initial conditions and that in the SSTs cannot be treated as totally independent sources of a trend in a seasonal forecast. Although the climate trends of SAT in MAM are much better represented in the HFP2 MME mean forecasts than those in DJF, the above results indicate that both the initial conditions and the lower boundary forcing were unable to fully produce the observed pronounced warming. As we mentioned above, it is not possible to separate the influence coming from the initial conditions as the lower boundary conditions during the first month of the model forecasts. To further understand the impact of the initial conditions to the linear SAT trends, the 1-month lead HFP2 MME forecasts for MAM and DJF are also analyzed and presented in Figs. 12a and 12b, respectively. Positive SAT trends can be found over the mid- to high-latitude North American and Eurasian continent for both MAM and DJF indicating the contribution of the SST forcing to the SAT warming trend. Again, the magnitudes are much weaker than those in the observations. The differences between Figs. 12a and 2a and the differences between Figs. 12b and 2b may reflect, to some extent, the impact of the initial conditions. Also notice that the SAT trend shown in Fig. 12b is more significant than that in Fig. 2b, suggesting the possibility that the trend in the initial condition in November (Fig. 8b) was somehow unrealistic and could have degraded the SAT trend in the lead-0 seasonal forecast for DJF. However, it also needs to be noted that the 1-month lead HFP2 seasonal forecasts for MAM and DJF use the persistent SST anomaly forcing from January and October, respectively, which may also cause some differences of SAT response comparing to the SST forcing from February and November.

Fig. 12.

The climate trends of SAT for (a) MAM and (b) DJF.

Fig. 12.

The climate trends of SAT for (a) MAM and (b) DJF.

Other mechanisms can also contribute to the underestimation of the observed warming trends in the HFP2 data. For instance, the GHG concentration was kept constant during the forecasts. Although the warming signal associated with the GHG can partly be incorporated into the forecast in the observation-based initial and boundary conditions, the forecasts lack radiative forcing that can produce errors over land as discussed by Boer (2009). Kumar and Yang (2003) examined the impact of snow on atmospheric seasonal variability over the extratropics in winter using a general circulation model. They found that when the snow is allowed to evolve freely in the model integrations, the interannual variability of the SAT was larger than when snow is simply prescribed to be a seasonally varying climatology. The lower boundary conditions, such as the sea ice and snow cover, are not well represented in the HFP2 forecasting system because of the capabilities of different models and the availability of data. Sea ice extent was initialized to be the observed one during the previous month and relaxed to climatology over a period of 15 days. The snow cover was specified with the NCEP weekly observations. It is possible that these can also account, at least partly, for the underestimation of the observed climate warming trends in the HFP2 MME seasonal forecasts.

Acknowledgments

This research was jointly funded by the National Natural Sciences Fundation of China (Grant 41105037) and by the Natural Science and Engineering Research Council of Canada (NSERC). Jia was also supported by the Fundamental Research Funds for the Central Universities (Grant 2012XZZX012) and the R&D Special Fund for Public Welfare Industry (meteorology) (Grant GYHY201106035). We are grateful to Prof. Jacques Derome for his helpful suggestions for improving our paper.

REFERENCES

REFERENCES
Boer
,
G.
,
2009
:
Climate trends in a seasonal forecasting system
.
Atmos.–Ocean
,
47
,
123
138
.
Boer
,
G.
,
N. A.
McFarlane
,
R.
Laprise
,
J. D.
Henderson
, and
J.-P.
Blanchet
,
1984
:
The Canadian Climate Centre spectral atmospheric general circulation model
.
Atmos.–Ocean
,
22
,
397
429
.
Cai
,
M.
, and
C.-S.
Shin
,
2009
:
The role of long-term trends in seasonal predictions: Implication of global warming in the NCEP CFS
.
Wea. Forecasting
,
24
,
965
973
.
Côté
,
J.
,
S.
Gravel
,
A.
Méthot
,
A.
Patoine
,
M.
Roch
, and
A.
Staniforth
,
1998a
:
The operational CMC-MRB Global Environmental Multiscale (GEM) model. Part I: Design considerations and formulation
.
Mon. Wea. Rev.
,
126
,
1373
1395
.
Côté
,
J.
,
J.-G.
Desmarais
,
S.
Gravel
,
A.
Méthot
,
A.
Patoine
,
M.
Roch
, and
A.
Staniforth
,
1998b
:
The operational CMC-MRB Global Environmental Multiscale (GEM) model. Part II: Results
.
Mon. Wea. Rev.
,
126
,
1397
1418
.
Derome
,
J.
, and
Coauthors
,
2001
:
Seasonal predictions based on two dynamical models
.
Atmos.–Ocean
,
39
,
485
501
.
Doblas-Reyes
,
F.
,
R.
Hagedorn
,
T. N.
Palmer
, and
J.-J.
Morcrette
,
2006
: Impact of increasing greenhouse gas concentrations in seasonal ensemble forecasts. Geophys. Res. Lett.,33, L07708,
doi:10.1029/2005GL025061
.
Jia
,
X.
,
H.
Lin
, and
D.
Jacques
,
2009
:
The influence of tropical Pacific forcing on the Arctic Oscillation
.
Climate Dyn.
,
32
,
495
509
.
Johannessen
,
O.
, and
Coauthors
,
2004
:
Arctic climate change: Observed and modelled temperature and sea-ice variability
.
Tellus
,
56A
,
328
341
.
Kalnay
,
E.
, and
Coauthors
,
1996
:
The NCEP/NCAR 40-Year Reanalysis Project
.
Bull. Amer. Meteor. Soc.
,
77
,
437
471
.
Kharin
,
V. V.
,
G. J.
Boer
,
W. J.
Merryfield
,
J. F.
Scinocca
, and
W.-S.
Lee
,
2012
:
Statistical adjustment of decadal predictions in a changing climate
.
Geophys. Res. Lett.
,
39
, L19705,
doi:10.1029/2012GL052647
.
Kumar
,
A.
, and
F.
Yang
,
2003
:
Comparative influence of snow and SST variability on extratropical climate in northern winter
.
J. Climate
,
16
,
2248
2261
.
Lin
,
H.
,
J.
Derome
, and
G.
Brunet
,
2005
: Correction of atmospheric dynamical seasonal forecasts using the leading ocean-forced spatial patterns. Geophys. Res. Lett.,32, L14804,
doi:10.1029/2005GL023060
.
Liniger
,
M.
,
H.
Mathis
,
C.
Appenzeller
, and
F.
Doblas-Reyes
,
2007
: Realistic greenhouse gas forcing and seasonal forecasts. Geophys. Res. Lett.,34, L04705,
doi:10.1029/2006GL028335
.
Luo
,
J.-J.
, and
K. S.
Behera
,
2011
:
Impact of global ocean surface warming on seasonal-to-interannual climate prediction
.
J. Climate
,
24
,
1626
1646
.
McFarlane
,
N. A.
,
G. J.
Boer
,
J.-P.
Blanchet
, and
M.
Lazare
,
1992
:
The Canadian Climate Centre second-generation general circulation model and its equilibrium climate
.
J. Climate
,
5
,
1013
1044
.
Ritchie
,
H.
,
1991
:
Application of the semi-Lagrangian method to a multilevel spectral primitive-equation model
.
Quart. J. Roy. Meteor. Soc.
,
117
,
91
106
.
Sardeshmukh
,
P.
, and
B. J.
Hoskins
,
1988
:
The generation of global rotational flow by steady idealized tropical divergence
.
J. Atmos. Sci.
,
45
,
1228
1251
.
Stott
,
P.
,
S.
Tett
,
G.
Jones
,
M.
Allen
,
J.
Mitchell
, and
G.
Jenkins
,
2000
:
External control of twentieth century temperature by natural and anthropogenic forcings
.
Science
,
290
,
2133
2137
.
Wallace
,
J. M.
, and
D.
Guztler
,
1981
:
Teleconnections in the geopotential height field during the Northern Hemisphere winter
.
Mon. Wea. Rev.
,
109
,
784
812
.
Zhao
,
M.
, and
P. A.
Dirmeyer
,
2004
: Pattern and trend analysis of temperature in a set of seasonal ensemble simulations. Geophys. Res. Lett.,31, L11202, doi:10.1029/2003GL019298.