Seasonal forecasts bridge the gap between the weather forecast and climate prediction, which play a key role in building a seamless forecasting system. In the past two decades, the dynamic seasonal forecast has developed rapidly. Many national climate prediction centers have established and developed their own seasonal and interannual numerical forecast models (Stockdale et al., 2010). China also established its own dynamic seasonal forecast system in 2003 (Ding et al., 2004; Wang et al., 2009), which serves as one of the main tools for seasonal forecasts. In recent years, the World Meteorological Organization (WMO) and other relevant agencies have extensively assessed dynamic seasonal forecasts to evaluate the progress, to reveal existing problems, and to show prospects (Wang et al., 2005; Kumar et al., 2012). The Ensemble-Based Predictions of Climate Changes and Their Impacts (ENSEMBLES) is a European multi-model ensemble system for seasonal-to-annual forecasts including five state-of-the-art global coupled atmosphere–ocean climate models; it is one of the major international prediction hindcast experiments that provide a large number of hindcasts to analyze predictability and related physical processes (van der Linden and Mitchell, 2009; Weisheimer et al., 2009; Dobla-Relyes et al., 2010). Some studies have shown that the ENSEMBLES hindcasts have high forecast skills for the Indian summer monsoon, tropical Pacific sea surface temperature (SST), and summer climate in Northwest Pacific (Weisheimer et al., 2009; Rajeevan et al., 2012; Li et al., 2014), but its ability to predict the summer monsoon precipitation in East Asia needs to be improved (Li et al., 2012; Fan et al., 2016).
Due to the influence of monsoon and topography of the Tibetan Plateau, climate variability in China and its related physical processes are extremely complex. How to reasonably simulate and predict the climate characteristics in China has always been a primary concern of Chinese meteorologists. A number of studies have assessed the capability of global climate models (GCMs) in simulating the climate in China. These results suggest that most of the climate models overestimate the magnitude of seasonal and annual precipitation in most regions of China, especially along the eastern edge of the Tibetan Plateau, and underestimate summer precipitation over the southeastern coast of China (Yu et al., 2000; Xu et al., 2010; Huang et al., 2013). Jiang et al. (2005) reported that simulation errors of surface air temperature, precipitation, and sea level pressure are generally large over and around the Tibetan Plateau based on the outputs of seven fully coupled atmosphere–ocean models. Zhou and Yu (2006) examined the surface air temperature simulated by 19 coupled GCMs used for the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC AR4) over China and the globe, and suggested that the reproducibility of the surface air temperature averaged over China is acceptable, though it is lower than that of the global and hemispheric averages. Miao et al. (2012) suggested that the annual precipitation and annual temperature generated by 24 GCMs from Coupled Model Intercomparison Project Phase 3 (CMIP3) should be used cautiously in China due to their poor performances. Chen and Frauenfeld (2014) found that the CMIP5 models can reproduce the spatial pattern of precipitation over China, which represents an improvement over CMIP3 models. Moreover, the Beijing Climate Center Climate System Model (BCC_CM1.0) of China has the capability to predict the Meiyu precipitation, but it is neither skillful in forecasting the intensity as well as decadal scale variability of precipitation nor the persistent heavy precipitation events (Si et al., 2009).
The characteristics of spring climate in China are distinct from those in summer. Due to the combined impacts of the climate systems in the tropics and mid–high latitudes, the climate variability in spring and its related physical processes are complicated. Wu and Kirtman (2007) suggested that the spring Eurasian snow cover in western Siberia is positively correlated with spring rainfall in southern China. Chen et al. (2015) indicated the contribution of the North Atlantic Ocean state to the surface air temperature and atmospheric circulation anomalies over northeastern China. You and Jia (2018) investigated the interannual variations and prediction of the leading two empirical orthogonal function (EOF) modes of spring precipitation over China, and find that the ENSO-related tropical SST anomalies in the previous winter and the North Atlantic SST anomaly dipole can contribute to the first and second EOF modes, respectively. Spring is a critical period for sowing and growing of crops, and the early planting season in southern China is heavily dependent on the amount of spring precipitation. However, the interannual variability of spring climate is large, prone to disastrous weathers such as low temperature, droughts, and floods (Feng and Li, 2011). Therefore, making a more reasonable seasonal forecast of the spring climate can aid in early decision making, prevention, and loss reduction, and thus greatly benefits the society and economy (Qiu et al., 2009; Chen et al., 2014; Feng et al., 2014; Wu and Mao, 2016). Studies on assessing the capability of the ENSEMBLES hindcasts in predicting spring climate in China are limited. In this study, we assess the forecast capability of the ENSEMBLES in predicting spring climate in China by in terms of precipitation, temperature, and circulation. These results will provide a scientific basis for further improvement of the GCMs and prediction of climate variability in China.
In Section 2, data and methods are provided. In Section 3, we evaluate the capability of the ENSEMBLES in forecasting the climatology, interannual variability, and dominant EOF modes of precipitation, 2-m air temperature, and atmospheric circulations. We also investigate the interannual spread and inter-member spread of spring precipitation and 2-m air temperature. In Section 4, through the singular value decomposition (SVD) analyses of the circulation and SST, the primary ocean–atmosphere coupled modes associated with the spring precipitation in China are identified. Section 5 provides a summary and discussion.2 Data and methods
We use the monthly precipitation data from the Global Precipitation Climatology Project (GPCP; Adler et al., 2003) on 2.5° × 2.5° grids for the period. We also use the monthly 2-m air temperature and winds from the NCEP–DOE (Department of Energy) AMIP-II reanalysis (Kanamitsu, et al., 2002) on 2.5° × 2.5° grids for the same period 1980–2005. In addition, we use the monthly SST data from the NOAA (Huang et al., 2014; Liu et al., 2015) on 2° × 2° grids for the same period. Spring average is for March, April, and May (MAM).
The ENSEMBLES project is a multi-model ensemble system developed by the European Union. It provides a series of hindcasts from 1960 to 2005 by using five fully coupled atmosphere–ocean–land models from the ECMWF, Leibniz Institute of Marine Sciences at Kiel University (IFM-GEOMAR), Météo-France (MF), UK Met Office (UKMO), and Euro-Mediterranean Center for Climate Change (CMCC-INGV). See van der Linden and Mitchell (2009) for more details. As satellite observations were available from 1979, the reanalysis data have become more reliable and homogeneous since then than in the pre-satellite era. To make a better comparison with observations, the hindcasts of all the five models for the period of 1980–2005 are used in this study. For each year, 7-month-long seasonal forecasts starting on the 1st of February, May, August, and November are carried out. Additionally, the November-initialized forecasts from all the models except CMCC-INGV are extended to 14-month-long annual forecasts. The individual model ensembles consist of nine members for each ensemble with different initial conditions. Therefore, MAM mean can be obtained from the hindcasts initialized in February (1-month lead) and November (4-month lead). Hereafter, MAM (Feb) and MAM (Nov) denote the MAM mean on the starting month of February and November, respectively.
There are several ways to construct multi-model ensemble (MME) from individual models (Krishnamurti et al., 1999; Doblas-Reyes et al., 2005). Due to the relatively small sample size of seasonal hindcasts, finding robust non-equal weights in the combination of models proves difficult. Therefore, the simplest and most straightforward method by applying equal weights to all contributing models and ensemble members is used in this study.
We use the Taylor diagram (Taylor, 2001) to assess the forecast skill of spring climatology in China by summarizing the degree of correspondence between the simulated and observed fields. On this diagram, the correlation coefficient and root-mean-square error (RMSE) between the two fields, along with the ratio of standard deviations of the two patterns, are all indicated by a single point. In addition, the EOF analysis is used to assess the capability of the models to capture the primary modes of precipitation and temperature variability. A primary goal of EOF analysis is to identify and extract the physically and dynamically independent patterns from a dataset (Wilks, 1995). These independent “modes” provide important clues as to the physics and dynamics of the system being studied.
SVD is a method designed to find covariant patterns in two different variable fields. It is a direct and objective method for assessing the strength of covariability between the two fields. SVD operates on the covariance matrix between the two fields and provides pairs of spatial modes with high temporal covariance. The SVD modes are ordered according to the amount of squared covariance explained. Further details on SVD analysis in meteorology can be found in Wallace et al. (1992) and Deser and Timlin (1997). For the ENSEMBLES hindcasts, an SVD analysis is performed on the five models by using the 26-yr records of MAM averaged vorticity and SST fields. For example, the SST matrix is (Nx, Ny, Nens, Nyr), where Nx and Ny are the zonal and meridional grid numbers, respectively, Nens denotes the ensemble number, and Nyr represents the number of years. We focus on the inter-member variability by subtracting the ensemble mean (nx, ny, nyr) from the matrix (Nx, Ny, Nens, Nyr). Then, the matrix (Nx × Ny, Nens × Nyr) is used for the SVD analysis. Hence, the conventional time dimension is enlarged by the ensemble size. Similarly, we obtain the interannual anomalies by calculating the deviations from the climatological mean. In addition, to obtain a clear view of the relationship between the spring precipitation, wind, and the coupled modes, the correlation coefficients between the precipitation, wind, and the corresponding SST principal component (PC) of the SVD analysis are calculated.3 Assessment of ENSEMBLES predicted spring climate in China 3.1 Climatology
To assess the capability of the ENSEMBLES in predicting the spring climatology in China (15°–55°N, 70°–140°E), Fig. 1 shows the Taylor diagram of precipitation, 2-m air temperature, and 850-hPa wind. The predicted patterns that agree well with observations lie nearest to the point marked “REF” on the x-axis. These models have relatively high correlations and low RMSEs. The models lying on the dashed arc have correct standard deviations, indicating that the pattern variations are of right amplitude. Note that the models have the best forecast skills for the 2-m air temperature, with the RMSEs less than 0.25. The relative standard deviations are generally less than 1.0, indicating that the models have less spatial variability than the observations. For precipitation, the forecast spatial pattern is of good performance with the pattern correlation coefficients greater than 0.8, significant at the 95% confidence level. The RMSEs of MME and individual models except for the MF are below 0.5, among which the MME has a slightly better prediction skill. Besides, the forecast amplitude of precipitation variation is also less than that of the observed, with the ratios of standard deviations generally below 1.0. As for the 850-hPa wind, the forecast is of good performance with the correlation coefficients greater than 0.7, which is also significant at the 95% confidence level. However, the RMSEs scatter between 0.25 and 0.75. Overall, the ENSEMBELS can well predict the main climatological characteristics of spring precipitation, 2-m temperature, and circulations in China, and their MME is generally better than the individual models. In addition, both February and November initialization results have good performances.
Differences between MME and observations for MAM precipitation and 2-m air temperature by using the hindcasts initialized in February (similar for those initialized in November) are shown in Figs. 2a, c. In most regions of China, the forecast precipitation is larger than observations, with the maxima located to the east of the Tibetan Plateau and south of the Loess Plateau. However, the precipitation in the coastal area of southern China is underestimated. As for the 2-m air temperature, warm deviation can be found in most parts of northwestern and southern China with a false high-value center located on the Tibetan Plateau, while cold deviation exists in other regions. These biases may be related to the deficiency of GCMs in representing the atmospheric responses to complex topography (Xu and Xu, 2012), as well as our lack of understanding some physical processes related to the East Asian monsoon and land–sea contrasts (Chen et al., 1997). The standard deviations among the five models are shown in Figs. 2b, d. For precipitation (Fig. 2b), the poor consistency centers at the eastern Tibetan Plateau and south of the Yangtze River. For 2-m air temperature (Fig. 2d), the largest standard deviations appear in Xinjiang Region. The uncertainties among the individual models are probably related to the coarse resolution of the climate models (Feng et al., 2011).3.2 Interannual variability
For the interannual variability of spring climate in China, the correlation coefficients between forecasted and observed interannual anomaly sequences for MME precipitation (Figs. 3a, b) and the 2-m air temperature (Figs. 3c, d) with February and November initializations are examined. For precipitation, the region with correlation coefficient significant at the 95% confidence level occupies about 32% of the whole China for MAM (Feb) and 28% for MAM (Nov). The significant correlations appear in eastern China (EC), southwestern China (SW), Xinjiang (NW1), and the joint area of Gansu and Inner Mongolia (NW2) for February initialization, while the high correlations calculated by using the November initialization are located in the central part of northeastern China, eastern China, northern Tibetan Plateau, and western Sichuan. As for 2-m air temperature, the area that reaches a 95% confidence level occupies 57% and 37% of the whole region for MAM (Feb) and MAM (Nov), respectively. Compared with precipitation, the MME has a better performance in predicting the interannual variability of 2-m air temperature. In addition, the areas with significant correlations initialized in February are obviously larger than those initialized in November, with significant correlations mainly located in Xinjiang, central and southwestern China as well as southeastern coastal regions. Generally, the selection of different initialization months has a great impact on the correlation patterns. Note that the correlations for precipitation are significant in the EC region with both February and November initializations.
To better understand the forecast skill for the interannual variation, the interannual variation curves of the four significant positive correlation regions (EC, SW, NW1, and NW2) as well as the three low correlation regions (northeastern China, NE; central China, CC; eastern Tibetan Plateau, ET) in Fig. 3a are shown in Fig. 4. For the high correlation regions (Figs. 4a–d), the interannual correlations vary with models. However, the MME is of good performance in predicting the interannual variation of precipitation, and has a better forecast skill than the individual models, with the correlation coefficients of above 0.6 in EC and SW, and above 0.45 in NW1 and NW2, all significant at the 95% confidence level. Note that the MME does not show evident interannual variations as the observations. This may be due to the inconsistent internal variability of the climate system in these individual models. The MME tends to lose some original signals as the internal variabilities of individual models offset each other. For the low correlation regions (Figs. 4e–g), large differences among the individual models can be seen, while neither MME nor single models can pass the significant test. The bad performance and uncertainty among the individual models may be caused by the lack of reflecting the complex surface properties and coarse resolution of the climate models.3.3 Primary modes
Figure 5 shows the first two EOF modes of spring precipitation and 2-m air temperature for observations and MMEs initialized in February and November, respectively. For precipitation, the MME can capture the spatial pattern of the first two modes with correlation coefficients above 0.75, significant at the 95% confidence level, and the results from both February and November initializations show good performances. Note that the observed first two modes account for 38.1% of the total variance, while the first two modes explain 86.8% and 85.1% for the MMEs initialized in February and November, respectively, with the first variance contributing above 70% in both cases. It indicates that the variance contribution of the first mode of the MME precipitation is overestimated, which may cause the MME to capture only part of the spring precipitation variability. Previous studies have found that the ENSO-related tropical Pacific SST anomalies in the preceding winter have a great impact on the leading EOF mode of spring precipitation over China (You and Jia, 2018). The large variance explained by the first mode may be caused by the dominant signal of the ENSO in the ENSEMBLES. The first mode of predicted precipitation shows a significantly positive signal in the eastern part of China. However, the region of the observed strongest positive signals is mainly located in the southeastern coastal areas, while the MME is biased toward the south of the Yangtze River. For the second mode, the MME exhibits a dipole pattern in eastern China with negative anomaly in the Yangtze–Huaihe River basins and positive anomaly in the southern area, which is consistent with the observations, only with slight deviations for both intensity and range. The leading two EOF modes of the spring precipitation over China are consistent with the results achieved by You and Jia (2018).
For the 2-m air temperature, the MME also overestimates the variance contribution of the first mode. The spatial correlation of the first mode is above 0.8, significantly higher than that of the second mode. The observed first mode shows a spatial distribution with low temperature anomalies on the Tibetan Plateau and high temperature anomalies in the other regions. While the MME generally captures the positive signals in most areas of China, it fails to reflect the negative signals on the Tibetan Plateau. For the second mode, the MME can capture the positive signals in northeastern China. However, there are some differences in the eastern part of the Tibetan Plateau. In addition, deviations of the range and intensity of the negative signals in MME also exist.
To further assess the capability of individual models and MME to reproduce the EOF spatial patterns and PCs, we examine both pattern correlation and temporal correlation coefficients for precipitation (Figs. 6a, b) and 2-m air temperature (Figs. 6c, d). For the precipitation, the forecast skill of the first spatial pattern is better than that of the second mode except for the UKMO, with the spatial correlation coefficients above 0.6, significant at the 95% confidence level. Note that the forecast skills of the spatial pattern for the UKMO and MF are greatly influenced by the lead time, as the correlations using the hindcasts with the February initialization are higher than those initialized from November. The predicted first PC of precipitation agrees well with that of the observations, with the correlation coefficients generally above 0.6, passing the 5% significance level test, while most of the second PCs are not significantly correlated with that of the observations. For the 2-m air temperature, the performance of the first spatial pattern is also significantly better than that of the second mode, with the spatial correlation coefficients generally above 0.7. Note that the ECMWF initialized in February and the MF initialized in November have significantly poor performances in the second spatial pattern, and the performance of PC for 2-m air temperature is generally poor.3.4 Interannual and inter-member spreads
The spring climate anomalies in China are not only related to the external forcing such as tropical Pacific Ocean SST (Zhang and Sumi, 2002) and Eurasian snow (Wu and Kirtman, 2007), but also influenced by internal dynamic processes such as the Madden–Julian Oscillation (Bai et al., 2012). Inter-member variability is associated with the uncertainty caused by internal ocean–atmosphere coupled processes (Ma et al., 2017a), which can influence the ensemble forecast skill. Therefore, we examine the interannual and inter-member spreads of MAM precipitation and 2-m air temperature initialized in February (Fig. 7). The spatial pattern in the November initialization (figure omitted) is similar to that initialized in February. The interannual and inter-member spreads are measured by standard deviations. For instance, the matrix of precipitation is defined as (Nx, Ny, Nyr, Nens). We obtain the inter-member spread by calculating the standard deviation of the rightmost dimension. Similarly, we obtain the interannual spread by calculating the temporal standard deviation at each grid point. As shown in Fig. 7, the high-value area of precipitation spread is mainly located in the southeastern part of China. Note that the maximum appears in the southeastern coastal area for the observations, while it biases toward the south of the Yangtze River in the MME. For the 2-m air temperature, both observations and forecasts feature large variabilities in the mid-to-high latitudes. However, in the forecasts, the large variabilities on the eastern Tibetan Plateau are not captured. Additionally, a significant overestimate occurs in the southern part of China. Generally, the spatial patterns of the interannual and inter-member spreads of MME are consistent with the observed spreads, which suggests that the internal dynamic processes have major impacts on the interannual variability of spring climate in China (Li et al., 2012; Ma et al., 2017a, b).
Previous studies indicate that the mechanisms that account for the variation of precipitation over China can be affected by many anomalous states of lower boundary conditions (Wang et al., 2000; Wu and Kirtman., 2007; Feng et al., 2014). Among them, the SST over Pacific and Indian ocean is one of the most important factors that impact the spring precipitation over China on the interannual time scale (Zhang and Sumi, 2002; Wang et al., 2003; Chen et al., 2014; Ma et al., 2017a). In the following, we concentrate on examining the relationship between the spring precipitation, circulation, and SST, to further explore the coupled modes that have major impacts on the interannual variability of spring climate in China.4 Relationship between the spring precipitation, circulation, and SST
To examine the relationship between the spring precipitation, circulation, and SST, we perform an SVD analysis to extract the dominant patterns of covariability between MAM 850-hPa vorticity and SST in spring and at a lead of two months based on the interannual and inter-member variability. The relationships between the spring 850-hPa wind, precipitation, and the first two coupled modes are further discussed.4.1 Connection between the spring precipitation, circulation, and contemporaneous SST
Here, the MAM vorticity (0°–55°N, 70°–165°E) and SST (50°S–55°N, 40°E–90°W) anomalies are used as the left and right fields, respectively. In the Northern Hemisphere, positive and negative vorticities are indicators of low and high pressure systems, respectively. The vorticity anomalies in the region (0°–55°N, 70°–165°E) are closely related to the Northwest Pacific subtropical anticyclone activity, which influences the precipitation over East Asia significantly (Kosaka et al., 2013; Ma et al., 2017a). Figure 8 shows the first two SVD modes for the observations, which explains 40.5% and 15.3% of the total covariance, respectively. The vorticity field of the leading mode (Fig. 8a) is characterized by a negative correlation located in Northwest Pacific (NWP), which is noted as an anomalous anticyclone. Meanwhile, there is a positive correlation in the southern region of China. In the SST field of the first mode (Fig. 8b), positive SST correlations occur in the central and eastern equatorial Pacific with intensified equatorial westerlies, while negative correlations are located in the western Pacific (WP) and north of the central equatorial Pacific (NCP). In addition, there is a significant positive correlation in the equatorial Indian Ocean. In the vorticity field of the second mode (Fig. 8c), the negative vorticity moves southeastward compared with the leading mode. In the corresponding SST field (Fig. 8d), positive SST correlations mainly appear in the Maritime Continent, northern Indian Ocean (NIO), and NWP.
The vectors shown in Fig. 8 indicate the correlation coefficients between the MAM 850-hPa winds and the first two SST PCs from the SVD analysis reaching at 90% confidence level with a t-test. For the first mode, the southeastern China is under the control of intensified southwesterlies, with significantly positive precipitation anomalies sitting in the eastern part of China. The significant negative precipitation anomalies are mainly located in southeastern Asia and NWP, with weak signals in southwestern China and east of the Tibetan Plateau. For the second mode, positive precipitation anomalies are mainly located in northwestern and northern China. Therefore, the observed first mode may play a major role in the spring precipitation anomaly in eastern China, while the second mode associates with the precipitation anomaly in northwestern and northern China.
Figure 9 shows the SVD results for the ensemble mean forecast with the November initialization. The first two modes account for 81.1% and 9.5% of the total covariance, respectively. Note a significant overestimation of the first variance contribution. Indeed, the first PC of the SVD mode is highly correlated with the Nino3.4 index with the correlation coefficients of above 0.9 (figure omitted), significant at the 99% confidence level, and the second SVD PC is in poor correlation with the Nino3.4 index with the correlation coefficients below 0.2, indicating that the large variance explained by the first mode in the MME may be caused by the dominant signal of the ENSO in the ENSEMBLES. The predicted SVD results resemble the observations, identifying the first mode as positive SST correlations over the Indian Ocean and the central and eastern equatorial Pacific with intensified westerly anomalies, negative SST correlations in the WP and NCP (Fig. 9b). Meanwhile, a negative vorticity anomaly exists over the NWP, and southeastern China is under the control of anomalous positive vorticity (Fig. 9a). The significant southwesterly wind and positive precipitation anomaly are located in southeastern China, while the negative precipitation anomaly mainly occurs in South Asia and the NWP, with a slight impact on southwestern China. Many previous studies have revealed that ENSO is one of the most important factors that influence precipitation anomalies in southern China, and the Philippine Sea anticyclone (PSAC) is the key system that bridges the remote ENSO forcing to East Asian climate variation (Wang et al., 2000; Xie et al., 2009; Jia et al., 2014), which can also be seen from the leading SVD mode in Fig. 9.
In the second mode, positive SST correlations are mainly located in the Maritime Continent, Bay of Bengal, and NWP, while negative SST correlations are in the central equatorial Pacific accompanied by anomalous northeasterly trade. The positive SST anomaly and corresponding weak westerlies in the eastern equatorial Pacific also exist (Fig. 9d). In addition, the anomalous anticyclone in the second mode shows an eastward and southward position compared with the first mode (Fig. 9c), which is consistent with the observations. However, there is a difference of the location between the predicted anomalous precipitation and observations, as the former shows a significantly positive signal of precipitation and corresponding positive vorticity in the southeastern coastal areas of China.
The MAM SVD results with the initialization of February (figure omitted) closely resemble those for the November initialization. Table 1 summarizes the correlation coefficients between the observed and ensemble mean SVD PCs. For the SVD from hindcasts in February and November, both PC1 and PC2 are correlated with the counterparts from observation, significant at the 90% confidence level. Thus, the coupled modes of the spring circulation and SST predicted by the MME are closely correlated with their counterparts from the observations, which give us confidence to use ENSEMBLES hindcasts to investigate the spring climate predictability. And the first mode may play a leading role in the climate variability over eastern China, though the predicted locations of anomalous southwesterlies and precipitation are generally biased toward the south compared to the observations.
|Correlation coefficient||FEB PC1||FEB PC2||NOV PC1||NOV PC2|
In the following, SVD analyses are conducted to extract the dominant coupled modes between the inter-member anomalies of MAM 850-hPa vorticity and SST. Figure 10 presents the first two SVD modes of the inter-member variability from the hindcasts initialized in November, which account for 33.4% and 22.6% of the total covariance, respectively. The vorticity fields resemble the first two SVD modes of ensemble mean interannual variability, identifying the anomalous anticyclone located over the NWP for the first mode and a southeastward position for the second mode (Figs. 10a, c). The SST field of the first mode features significant positive SST correlations in the equatorial Indian Ocean and central-eastern equatorial Pacific, while the second mode shows positive SST correlations in the NIO and NWP, as well as negative correlations in the central Pacific (Figs. 10b, d). Figures 10a and 10c present the correlation coefficients between the inter-member 850-hPa wind, precipitation, and SST PCs of the SVD analysis. For the first mode, intensified anomalous southwesterlies and positive precipitation occur over southeastern China, while the second mode presents weak northeasterlies with precipitation confined to the south of the mainland. The SVD modes between the inter-member vorticity and SST anomalies are similar to those between the interannual anomalies. It suggests that the prediction spread in the spring climate forecast of eastern China is mainly derived from the internal dynamics of ocean–atmosphere interaction over the tropical Pacific, NCP, and Indian Ocean. As we mentioned before, anomalous SST associated with El Niño in the central-eastern equatorial Pacific can induce anomalous descent over the Philippine Sea, favoring the development of an anomalous anticyclone. The anomalous anticyclone acts as a medium bridging remote El Niño forcing and climate variations in East Asia as it enhances southwesterlies to its northwest flank and generates positive precipitation over China. The anomalous anticyclone can persist through local air–sea interactions over the NWP, causing rainfall anomalies over eastern China in the ensuing spring.4.2 Connection between the spring precipitation, circulation, and preceding SST
We find that the 2-month preceding SST is well coupled with the MAM circulation and precipitation in eastern China. Figure 11 shows the observed first two SVD modes of MAM vorticity and SST at a lead of two months, accounting for 39.4% and 18.7% of the total covariance, respectively. Note that the modes calculated by preceding SST resemble those in Fig. 8. For the first mode (Figs. 11a, b), the SST field presents positive SST correlations over the central-eastern equatorial Pacific and negative correlations in the WP and NCP, while the vorticity field shows an anomalous anticyclone over the NWP with a significant impact on the precipitation in eastern China. For the second mode (Figs. 11c, d), positive SST correlations occur over the NWP, Indian Ocean, and eastern equatorial Pacific. An anomalous anticyclone with a southeastward position compared with the leading mode is shown in the second vorticity field. Note that the second mode has little connection with the precipitation in eastern China, but gives a positive precipitation anomaly in the northwest.
For the predicted SVD results of the ensemble mean forecast with the November initialization (figure omitted), the first two modes account for 83.6% and 10.5% of the total covariance, respectively. Similar to the coupled modes in Fig. 9, the first mode features the anomalous anticyclone over the NWP with the corresponding intensified southwesterlies and positive precipitation anomaly in southeastern China. The related negative SST anomaly occurs in the WP and NCP, while the positive SST anomaly occurs in the central-eastern equatorial Pacific and Indian Ocean. The second mode features a southeastward position of the negative vorticity anomaly with the positive precipitation anomaly confined to the south of 25°N, and the positive SST anomaly mainly in the NIO and NWP.
Finally, the first two SVD modes between inter-member anomalies of MAM 850-hPa vorticity and SST at a lead of two months are calculated (figure omitted), which account for 33.5% and 27.0% of the total covariance, respectively. Note that the correlations for the inter-member variability in 850-hPa wind and precipitation resemble those for ensemble mean interannual anomalies, indicating that the first coupled mode plays a more important role in the climate variability of eastern China, with a significant precipitation anomaly located in the southeast. No more tautology here as the first two modes as well as the correlation fields of precipitation and 850-hPa winds are similar to those calculated by the contemporaneous SST. During positive ENSO events, though the positive SST anomalies in the eastern tropical Pacific decay rapidly, the anomalous anticyclone over NWP can be maintained in the ensuing spring and summer through local air–sea interactions. With regard to the ocean memory that maintains the anomalous anticyclone, Wang et al. (2003) emphasized the SST cooling in the easterly trade wind regime of the NWP. More specifically, on the east side of the anticyclone, anomalous northeastern winds strengthen trade winds, enhancing the latent heat flux from the ocean to the atmosphere and cooling the underlying SST. Negative SST anomalies excite westward propagating Rossby waves, which reinforce the anticyclone in return. Southerlies on western flank of the anticyclone transport warm air from low-latitude oceans and contribute to the positive SST anomalies over the South China Sea and surrounding regions. In addition, El Niño can induce the Indian Ocean warming, which persists into the following spring (Du et al., 2009). And modeling studies also support that the Indian Ocean can force the anomalous anticyclone over NWP (Huang et al., 2010; Chowdary et al., 2011). Thus, the SST anomalies in the leading SVD mode at a lead of two months may be a predictor for the spring climate variability in eastern China. Additionally, the forecast skill can mainly arise from the internal dynamics of the ocean–atmosphere interaction over the tropical Pacific, NCP, and Indian Ocean.5 Summary and discussion
The ENSEMBELS can forecast the climatological characteristics of the spring precipitation, 2-m air temperature, and 850-hPa wind in China with a good performance by both February and November initializations, and the MME has a slightly better performance than individual models. The spring precipitation is overestimated in most parts of China, while that in coastal southern China is underestimated systematically, and the 2-m air temperature shows a false high-value center on the Tibetan Plateau. The five models present poor consistency for precipitation on the eastern Tibetan Plateau and south of the Yangtze River, while poor consistency for 2-m air temperature is located in Xinjiang Region. The MME forecast of the interannual variability is of good performance in some regions of China such as eastern China with the February initialization. The selection of different initialized months has a great impact on the correlation results. The MME tends to lose some original interannual signals as the internal variabilities of individual models offset each other. Additionally, the MME can reproduce the first two EOF spatial patterns of the precipitation and 2-m air temperature in China, with slight differences in intensity and range; and the performances using February and November initializations are similarly good. For the individual models, the performance of the first spatial pattern and its PC is better than that of the second mode, and the correlation of PCs for 2-m air temperature is generally poor. Moreover, the forecast skills of individual models are greatly influenced by initialization time.
The spatial patterns of the interannual and inter-member spreads of MME for the precipitation and 2-m air temperature are consistent with the observed spread, which suggests that the internal dynamic processes have major impacts on the interannual anomaly of spring climate in China. We identify two SVD coupled modes between the interannual and inter-member anomalies of the 850-hPa vorticity in spring and SST both in spring and at a lead of two months. The first mode features an anomalous anticyclone in Northwest Pacific with positive precipitation and southwesterly anomalies in eastern China. The second mode is characterized by a southeastward position of the negative vorticity anomaly, with less connection with the precipitation in eastern China. The SST at a lead of two months may be a good predictor for the spring climate in eastern China, and the predictability is mainly rooted in the ocean–atmosphere interaction over the tropical Pacific, NCP, and Indian Ocean, which is highly correlated with ENSO. Specific mechanisms of the coupled processes need to be revealed in future studies.
Acknowledgments. The authors acknowledge the ENSEMBLES project for providing the model outputs. We also acknowledge the organizations that provided the observations for this study: the NCEP Reanalysis Derived data, NOAA_ERSST_V4 data, and GPCP precipitation data, which are provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA (at https://www.esrl.noaa.gov/psd/). The authors thank the anonymous reviewers for their constructive and thoughtful comments, which have helped improve this manuscript.
|Adler, R. F., G. J. Huffman, A. Chang, et al., 2003: The version-2 Global Precipitation Climatology Project (GPCP) monthly precipitation analysis (1979–present). J. Hydrometeor., 4, 1147–1167. DOI:10.1175/1525-7541(2003)004<1147:TVGPCP>2.0.CO;2|
|Bai, X. X., C. Y. Li, and L. Li, 2012: Numerical simulation study of the Madden–Julian Oscillation influences on spring precipitation in China. Acta Meteor. Sinica, 70, 986–1003. DOI:10.11676/qxxb2012.083|
|Chen, J. P., Z. P. Wen, R. G. Wu, et al., 2014: Interdecadal changes in the relationship between southern China winter–spring precipitation and ENSO. Climate Dyn., 43, 1327–1338. DOI:10.1007/s00382-013-1947-x|
|Chen, L., and O. W. Frauenfeld, 2014: A comprehensive evaluation of precipitation simulations over China based on CMIP5 multimodel ensemble projections. J. Geophys. Res. Atmos., 119, 5767–5786. DOI:10.1002/2013JD021190|
|Chen, Q. Y., Y. Q. Yu, and Y. F. Guo, 1997: Simulation of East Asian summer monsoon with IAP CGCM. Adv. Atmos. Sci., 14, 461–472. DOI:10.1007/s00376-997-0064-3|
|Chen, S. F., R. G. Wu, and Y. Liu, 2015: Dominant modes of interannual variability in Eurasian surface air temperature during boreal spring. J. Climate, 29, 1109–1125. DOI:10.1175/JCLI-D-15-0524.1|
|Chowdary, J. S., S. P. Xie, J. J. Luo, et al., 2011: Predictability of Northwest Pacific climate during summer and the role of the tropical Indian Ocean. Climate Dyn., 36, 607–621. DOI:10.1007/s00382-009-0686-5|
|Deser, C., and M. S. Timlin, 1997: Atmosphere–ocean interaction on weekly timescales in the North Atlantic and Pacific. J. Climate, 10, 393–408. DOI:10.1175/1520-0442(1997)010<0393:AOIOWT>2.0.CO;2|
|Ding, Y. H., Q. Q. Li, W. J. Li, et al., 2004: Advance in seasonal dynamical prediction operation in China. Acta Meteor. Sinica, 62, 598–612. DOI:10.11676/qxxb2004.059|
|Doblas-Reyes, F. J., R. Hagedorn, and T. N. Palmer, 2005: The rationale behind the success of multi-model ensembles in seasonal forecasting—II. Calibration and combination. Tellus A, 57, 234–252. DOI:10.1111/j.1600-0870.2005.00104.x|
|Doblas-Reyes, F. J., A. Weisheimer, T. N. Palmer, et al., 2010: Forecast Quality Assessment of the ENSEMBLES Seasonal-to-Decadal Stream 2 Hindcasts. ECMWF Technical Memorandum 621, ECMWF, Reading UK,1–45, doi: 10.21957/l0x9tmf12.|
|Du, Y., S. P. Xie, G. Huang, et al., 2009: Role of air–sea interaction in the long persistence of El Niño-induced north Indian Ocean warming. J. Climate, 22, 2023–2038. DOI:10.1175/2008JCLI2590.1|
|Fan, Y., K. Fan, and B. Q. Tian, 2016: Has the prediction of the South China Sea summer monsoon improved since the late 1970s?. J. Meteor. Res., 30, 833–852. DOI:10.1007/s13351-016-6052-8|
|Feng, J., and J. P. Li, 2011: Influence of El Niño Modoki on spring rainfall over South China. J. Geophys. Res. Atmos., 116, D13102. DOI:10.1029/2010JD015160|
|Feng, J. Q., L. J. Yu, and D. X. Hu, 2014: Influence of Indian Ocean subtropical dipole on spring rainfall over China. Int. J. Climatol., 34, 954–963. DOI:10.1002/joc.3732|
|Feng, L., T. J. Zhou, B. Wu, et al., 2011: Projection of future precipitation change over China with a high-resolution global atmospheric model. Adv. Atmos. Sci., 28, 464–476. DOI:10.1007/s00376-010-0016-1|
|Huang, B. Y., V. F. Banzon, E. Freeman, et al., 2014: Extended reconstructed sea surface temperature version 4 (ERSST. v4). Part I: Upgrades and intercomparisons. J. Climate, 28, 911–930. DOI:10.1175/JCLI-D-14-00006.1|
|Huang, D. Q., J. Zhu, Y. C. Zhang, et al., 2013: Uncertainties on the simulated summer precipitation over Eastern China from the CMIP5 models. J. Geophys. Res. Atmos., 118, 9035–9047. DOI:10.1002/jgrd.50695|
|Huang, G., K. M. Hu, and S. P. Xie, 2010: Strengthening of tropical Indian Ocean teleconnection to the Northwest Pacific since the mid–1970s: An atmospheric GCM study. J. Climate, 23, 5294–5304. DOI:10.1175/2010JCLI3577.1|
|Jia, X. J., J. Y. Lee, H. Lin, et al., 2014: Interdecadal change in the Northern Hemisphere seasonal climate prediction skill: Part I. The leading forced mode of atmospheric circulation. Climate Dyn., 43, 1595–1609. DOI:10.1007/s00382-013-1988-1|
|Jiang, D. B., H. J. Wang, and X. M. Lang, 2005: Evaluation of East Asian climatology as simulated by seven coupled models. Adv. Atmos. Sci., 22, 479–495. DOI:10.1007/BF02918482|
|Kanamitsu, M., W. Ebisuzaki, J. Woollen, et al., 2002: NCEP-DOE AMIP-II reanalysis (R-2). Bull. Amer. Meteor. Soc., 83, 1631–1644. DOI:10.1175/BAMS-83-11-1631|
|Kosaka, Y., S. P. Xie, N. C. Lau, et al., 2013: Origin of seasonal predictability for summer climate over the Northwestern Pacific. Proc. Natl. Acad. Sci. USA, 110, 7574–7579. DOI:10.1073/pnas.1215582110|
|Krishnamurti, T. N., C. M. Kishtawal, T. E. LaRow, et al., 1999: Improved weather and seasonal climate forecasts from multimodel superensemble. Science, 285, 1548–1550. DOI:10.1126/science.285.5433.1548|
|Kumar, V., and T. N. Krishnamurti, 2012: Improved seasonal precipitation forecasts for the Asian monsoon using 16 atmosphere–ocean coupled models. Part I: Climatology. J. Climate, 25, 39–64. DOI:10.1175/2011JCLI4125.1|
|Li, C. F., R. Y. Lu, and B. W. Dong, 2012: Predictability of the western North Pacific summer climate demonstrated by the coupled models of ENSEMBLES. Climate Dyn., 39, 329–346. DOI:10.1007/s00382-011-1274-z|
|Li, C. F., R. Y. Lu, and B. W. Dong, 2014: Predictability of the western North Pacific summer climate associated with different ENSO phases by ENSEMBLES multi-model seasonal forecasts. Climate Dyn., 43, 1829–1845. DOI:10.1007/s00382-013-2010-7|
|Liu, W., B. Y. Huang, P. W. Thorne, et al., 2015: Extended reconstructed sea surface temperature version 4 (ERSST. v4): Part II. Parametric and structural uncertainty estimations. J. Climate, 28, 931–951. DOI:10.1175/JCLI-D-14-00007.1|
|Ma, J., S. P. Xie, and H. M. Xu, 2017a: Intermember variability of the summer Northwest Pacific subtropical anticyclone in the ensemble forecast. J. Climate, 30, 3927–3941. DOI:10.1175/JCLI-D-16-0638.1|
|Ma, J., S. P. Xie, and H. M. Xu, 2017b: Contributions of the north pacific meridional mode to ensemble spread of ENSO prediction. J. Climate, 30, 9167–9181. DOI:10.1175/JCLI-D-17-0182.1|
|Miao, C. Y., Q. Y. Duan, L. Yang, et al., 2012: On the applicability of temperature and precipitation data from CMIP3 for China. PLoS One, 7, e44659. DOI:10.1371/journal.pone.0044659|
|Qiu, Y., W. J. Cai, X. G. Guo, et al., 2009: Dynamics of late spring rainfall reduction in recent decades over southeastern China. J. Climate, 22, 2240–2247. DOI:10.1175/2008JCLI2809.1|
|Rajeevan, M., C. K. Unnikrishnan, and B. Preethi, 2012: Evaluation of the ENSEMBLES multi-model seasonal forecasts of Indian summer monsoon variability. Climate Dyn., 38, 2257–2274. DOI:10.1007/s00382-011-1061-x|
|Si, D., Y. H. Ding, and Y. J. Liu, 2009: Evaluation of Meiyu prediction in the Yangtze–Huaihe region by coupled ocean-atmosphere general circulation model (BCC_CM1.0). Acta Meteor. Sinica, 67, 947–960. DOI:10.11676/qxxb2009.092|
|Stockdale, T. N., O. Alves, G. Boer, et al., 2010: Understanding and predicting seasonal-to-interannual climate variability — The producer perspective. Proced. Environ. Sci., 1, 55–80. DOI:10.1016/j.proenv.2010.09.006|
|Taylor, K. E., 2001: Summarizing multiple aspects of model performance in a single diagram. J. Geophys. Res. Atmos., 106, 7183–7192. DOI:10.1029/2000JD900719|
|van der Linden, P., and J. F. B. Mitchell, 2009: ENSEMBLES: Climate Change and Its Impacts: Summary of Research and Results from the ENSEMBLES Project. Met Office Hadley Centre, FitzRoy Road, Exeter EX1 3PB, UK, 160 pp.|
|Wallace, J. M., C. Smith, and C. S. Bretherton, 1992: Singular value decomposition of wintertime sea surface temperature and 500-mb height anomalies. J. Climate, 5, 561–576. DOI:10.1175/1520-0442(1992)005<0561:SVDOWS>2.0.CO;2|
|Wang, B., R. G. Wu, and X. H. Fu, 2000: Pacific–East Asian teleconnection: How does ENSO affect East Asian climate?. J. Climate, 13, 1517–1536. DOI:10.1175/1520-0442(2000)013<1517:PEATHD>2.0.CO;2|
|Wang, B., R. G. Wu, and T. Li, 2003: Atmosphere–warm ocean interaction and its impacts on Asian–Australian monsoon variation. J. Climate, 16, 1195–1211. DOI:10.1175/1520-0442(2003)16<1195:AOIAII>2.0.CO;2|
|Wang, B., Q. H. Ding, X. H. Fu, et al., 2005: Fundamental challenge in simulation and prediction of summer monsoon rainfall. Geophys. Res. Lett., 32, L15711. DOI:10.1029/2005GL022734|
|Wang, B., J. Y. Lee, I. S. Kang, et al., 2009: Advance and prospectus of seasonal prediction: Assessment of the APCC/CliPAS 14-model ensemble retrospective seasonal prediction (1980–2004). Climate Dyn., 33, 93–117. DOI:10.1007/s00382-008-0460-0|
|Weisheimer, A., F. J. Doblas-Reyes, T. N. Palmer, et al., 2009: ENSEMBLES: A new multi-model ensemble for seasonal-to-annual predictions—skill and progress beyond DEMETER in forecasting tropical Pacific SSTs. Geophys. Res. Lett., 36, L21711. DOI:10.1029/2009GL040896|
|Wilks, D. S., 1995: Statistical Methods in the Atmospheric Sciences: An Introduction. 2nd Ed., Academic Press, New York, 467 pp.|
|Wu, R. G., and B. P. Kirtman, 2007: Observed relationship of spring and summer East Asian rainfall with winter and spring Eurasian snow. J. Climate, 20, 1285–1304. DOI:10.1175/JCLI4068.1|
|Wu, X. F., and J. Y. Mao, 2016: Interdecadal modulation of ENSO-related spring rainfall over South China by the Pacific decadal oscillation. Climate Dyn., 47, 3203–3220. DOI:10.1007/s00382-016-3021-y|
|Xie, S. P., K. M. Hu, J. Hafner, et al., 2009: Indian Ocean capacitor effect on Indo-western Pacific climate during the summer following El Niño. J. Climate, 22, 730–747. DOI:10.1175/2008JCLI2544.1|
|Xu, C. H., Y. Luo, and Y. Xu, 2010: Assessment and projection for spatial–temporal distribution of precipitation in China based on global climate models. Adv. Climate Change Res., 6, 398–404. DOI:10.3969/j.issn.1673-1719.2010.06.002|
|Xu, Y., and C. H. Xu, 2012: Preliminary assessment of simulations of climate changes over china by CMIP5 multi-models. Atmos. Oceanic Sci. Lett., 5, 489–494. DOI:10.1080/16742834.2012.11447041|
|You, Y. J., and X. J. Jia, 2018: Interannual variations and prediction of spring precipitation over China. J. Climate, 31, 655–670. DOI:10.1175/JCLI-D-17-0233.1|
|Yu, R. C., W. Li, X. H. Zhang, et al., 2000: Climatic features related to eastern China summer rainfalls in the NCAR CCM3. Adv. Atmos. Sci., 17, 503–518. DOI:10.1007/s00376-000-0014-9|
|Zhang, R. H., and A. Sumi, 2002: Moisture circulation over East Asia during El Niño episode in northern winter, spring and autumn. J. Meteor. Soc. Japan, 80, 213–227. DOI:10.2151/jmsj.80.213|
|Zhou, T. J., and R. C. Yu, 2006: Twentieth-century surface air temperature over China and the globe simulated by coupled climate models. J. Climate, 19, 5843–5858. DOI:10.1175/JCLI3952.1|