RECENT ADVANCES IN THE STUDY OF EXTRA-SEASONAL TO ANNUAL CLIMATE PREDICTION AT IAP/CAS
LIN Zhaohui, WANG Huijun, ZHOU Guangqing, CHEN Hong,
LANG Xianmei, ZHAO Yan and ZENG Qingcun
ICCES, Institute of Atmospheric Physics, Chinese Academy of Sciences,P. O. Box, 9804, Beijing 100080, China
Recent advances on the dynamical climate prediction at the Institute of Atmospheric Physics, Chinese Academy of Sciences (IAP/CAS) during the last five years have been briefly described in this paper. Firstly, the IAP dynamical climate prediction system has been introduced, and the real-time prediction results by IAP DCP during 1998-2002 and their verification have also been summarized next. Then, the investigation of the importance of atmospheric initial condition on the seasonal climate prediction has been demonstrated, along with the studies on the influences from surface boundary conditions (e.g., land surface characteristics, sea surface temperature). Several techniques (e.g., ensemble technique, correction method, etc.) have also been illustrated which lead to the increase of the prediction skill for summer rainfall anomalies over China. Finally, the paper concludes with a list of critical requirements needed for the further improvement of dynamical seasonal climate prediction.Key words: seasonal prediction, Ensemble technique, ENSO prediction, soil moisture, correction method
As one of the main goals of the World Climate Research Programme study CLIVAR (Study of Climate Variability and Predictability) (WCRP, 1995), international research activities on the exploitation of predictability and the making of predictions of seasonal-to-interannual climate variation have been carried out in a number of research institutions, and there are also many national and international programs, such as, SMIP (Dynamical Seasonal Prediction Model Intercomparison Project), DEMETER (Development of a European Multimodel Ensemble system for Seasonal to inTERannual prediction), have been implemented. Especially during the last several years, rapid progress has been achieved in the dynamical seasonal to interannual prediction of climate anomalies.
China is located in the East Asian summer monsoon regions, where the monsoon related to rainfall anomalies could result in severe flood and drought conditions, and then induce great losses for the local economy and even human lives. So the seasonal climate prediction has become more and more important in the planning of economical development and protection of the people from disasters.
Early in 1989, the experiment of seasonal and extraseasonal predictions of summer monsoon rainfall anomaly by using Atmospheric-Oceanic coupled GCM had been carried out in the Institute of Atmospheric Physics, Chinese Academy of Sciences (IAP/CAS) (Zeng et al., 1990), after that the first short term climate prediction system (IAP PSSCA, hereafter referred to as IAP DCP-I) has been established and applied to the semi-operational predictions. All these efforts were summarized in Yuan et al. (1996) and Zeng et al. (1997).
In this paper, we would like to briefly summarize the research activities on the dynamical climate prediction since 1998, including the development of second generation of IAP dynamical climate prediction system and its application to the real-time prediction, the study of the climate predictability and some other issues arising from the practical climate prediction.
II. DEVELOPEMNT OF DYNAMICAL CLIMATE PREDICTION SYSTEM
1. ENSO Prediction System
By using the Global Atmosphere-Tropical Pacific Ocean Coupled Model, incorporated with the initialization technique for the coupled model (Zhou et al., 1999), an ENSO prediction system has been established in the Institute of Atmospheric Physics, Chinese Academy of Sciences (IAP/CAS) (Zhou et al., 1998). Here, the Tropical-Pacific Ocean model is a 14-layer OGCM ( 121°E to 69°W between 30°S and 30°N ) with horizontal resolution of 2° in longitude by 1° in latitude and 14 layers in the vertical (Zhang and Endoh, 1994), and the global atmosphere model is IAP 2L AGCM (Zeng et al., 1986；Lin and Zeng，1997).
In order to assess the forecast skills, the hindcast experiments for the period of November 1981 to December 1997 were performed. The anomaly correlations and root mean square (rms) errors of the predicted SST anomalies averaged over Nino3 region are shown in Fig. 1.
Fig.1. Correlations (a) and root-mean-square errors (b) with observation for forecast (solid line) and persistence (dashed line), respectively.
Also shown are the ones by persisting “predictions” for the same times. Although the forecast skills decrease with time, the forecasts are skillful at least up to more than 15 months in advance with the correlations above 0.54. In particular, for the predictions exceeding 4 months, the forecast skills are very much better than the persistent ones. In the first four months the skill of our “predictions” is slightly worse than the persistent ones. This is mainly attributed to the lack of ocean thermocline data in initial conditions. The mean rms errors are less than 0.89°C, much smaller than the persistent ones after 4 months from initial time (Zhou et al., 1998) .
After the establishment of IAP ENSO prediction, it has been used for the prediction of sea surface temperature (SST) anomaly evolution in the tropical Pacific Ocean (Zhou and Zeng, 2001). Real time prediction results show the decay of SST anomalies over Nino3 region from 1998, then the SST anomalies would turn to be negative in August 1998, and then kept negative during the whole year of 1999 and the first half-year of 2000, after the onset of La Nina event from the summer of 1998. The ensemble prediction for year 2002 which is initiated from 1st of January, February and March respectively, indicates that 4-year long lasting negative SST anomalies over Nino3 region will end and turn to be positive during the spring of 2002, although the magnitude of SST anomaly will not be large. Generally speaking, all these predictions agree quite well with the observation.
2. Dynamical Climate Prediction System
Since the establishment of the first generation of IAP dynamical climate prediction system (IAP DCP-I, also referred to as IAP Prediction System for Short-term Climate Anomaly－PSSCA), which is described in detail by Li (1992) and Zeng et al. (1997), this system had been applied to the semi-operational real time climate prediction from 1989 to 1997 (e.g., Zeng et al., 1990; Li et al., 1992; Zeng, 1994; Yuan et al., 1996; Zeng et al., 1997). Generally speaking, IAP DCP-I can quite well predict the large positive and negative anomalies of summer rainfall resulting in disastrous climate events, however, the prediction skill for IAP DCP-I is relative low over North China and Northwest China compared with over Eastern part of China and Southern China.
In order to further increase the predictive skill of IAP dynamical climate prediction system, many efforts have continually been taken. These efforts include the improvement of atmospheric and oceanic general circulation model, land surface model, the incorporation of the IAP ENSO prediction system into the prediction system, the improvement of the ensemble technique and correction system, etc. , and the second generation of IAP dynamical climate prediction system (IAP DCP-II) has been established in 1998 (Lin et al., 1998). Generally, IAP DCP-II consists of five components, i.e., IAP ENSO Prediction System; Prediction Integrations and Anomaly Coupling Technique; Ensemble Prediction Technique; Correction System and Prediction Products and Analyses, more detailed description of IAP DCP-II can also be found in Lin et al. (1999).
Hindcast experiment for the summer rainfall anomalies during 1980-1994 had been carried out by using IAP DCP-I and IAP DCP-II respectively. Generally speaking, the predictive skill of IAP DCP-II is higher than IAP DCP-I, especially over North China, Northeast China and Hetao region, and there is also slight increase in the predictive skill over South China and Yangtze River valley (Lin et al., 1998). Figure 2 gives the anomaly correlation coefficient (ACC) between the predicted and observed summer rainfall over Eastern China (105°E East) and North China (105－120°E, 34－42°N) , and we can find from Fig.2a that, the predictive skill for IAP DCP-II is higher than for IAP DCP-I, the averaged ACC during 1980-1994 is increased from 0.06 for IAP DCP-I to 0.25 for IAP DCP-II over Eastern China (Fig.2a). As for North China (Fig.2b), there is almost no skill for IAP DCP-I in the prediction of summer rainfall anomaly after 1985, because the corresponding correlation coefficients over North China are all negative during these years. However, for IAP DCP-II, the correlation coefficients are all positive except for year 1985, 1990 and 1994, and the correlation coefficient can exceed 0.50 for many years.
Fig.2. Anomaly Correlation Coefficient (ACC) between the prediction and observation for IAP DCP-I (dashed line) and IAP DCP-II (solid line) over (a) Eastern China (105°E East) and (b) North China (105－120°E, 34－42°N).
3. Real Time Climate Prediction
Since 1998, IAP DCP-II, instead of IAP DCP-I, has been applied to the real-time prediction of summer rainfall anomalies. The prediction is conducted in two-tiered fashion (e.g., Bengtsson et al., 1993; Hunt et al., 1994), in the first step, the tropical Pacific sea surface temperature anomaly (SSTA) is predicted by IAP ENSO prediction system, and in the second step, an ensemble of AGCM runs is preformed by combining the observed February SSTA and that predicted by IAP ENSO prediction, and final prediction product is obtained by averaging the total ensemble after correction, and the real-time ensemble prediction is usually carried out in March every year, with 28 different atmospheric conditions from NCEP real-time analysis ranged from February 1st to February 28th ( Lin et al.,1998).
Verification with the observed summer rainfall anomalies shows that, IAP DCP-II can quite well predict the large-scale patterns of summer flood and drought conditions over China. For example, the severe flood over Yangtze River Valley and Northeast China for year 1998, the positive summer rainfall anomalies over southern part of China for year 1999, and the rainfall maximum over lower reach of Yangtze River valley for year 2001, all these are quite well predicted by IAP DCP-II (Lin et al., 1998, 2000; Zeng et al., 2003). On the other hand, the persisted drought conditions over North China from 1999 to 2002 have also been pretty well predicted by IAP DCP-II (Lin et al., 2000, 2002; Zeng et al., 2003). All these indicate the capability of IAP DCP-II in the prediction of summer rainfall anomalies over China.
Together with the ensemble mean, the standard deviation of the ensemble prediction, and the probability distribution of positive rainfall anomalies are also been presented as another two prediction products, with small standard deviation and large possibility distribution for more reliable prediction ( Lin et al., 1998, 2000; Zeng et al., 2003).
Figure 3 gives the predicted summer (JJA mean) rainfall anomalies over China for year 2002, together with the observation result. Comparison between observation (Fig.3a) and prediction (Fig.3b) shows that, IAP DCP-II can quite reasonably predict the observed drought conditions over North and Northeast China, and the positive rainfall anomalies over the lower reach of Yangtze River valley and most part of southern China. On the other hand, the positive rainfall anomalies over western Hetao region and western and northern part of Xinjiang have also been quite correctly predicted. However, IAP DCP-II has not captured the precipitation maximum over southern China, and the magnitude of the predicted rainfall anomaly is also relatively weak compared with the observation.
Fig.3. Percentage summer (JJA mean) rainfall anomalies (%) over China for year 2002
(a) Observation; (b) Prediction by IAP DCP-II.
4. Experimental Prediction with Higher Resolution Model
Although the current version IAP DCP can quite well predict the large scale patterns of summer rainfall anomalies, its capability in predicting the detailed spatial distribution of the rainfall anomalies is limited by its relative coarse resolution. With the increasing development of computing environment, different version of IAP climate models with higher resolution has been adopted for the experimental seasonal climate prediction.
One of these efforts is the application of IAP 2L AGCM with 2° ´ 2.5° horizontal resolution. Twenty one-year hindcast experiment during 1980-2000 has been carried out, and hindcast results have been verified with the observation in order for the generation of corresponding correction system. Generally speaking, there does have some increase in the skill for the prediction of summer rainfall anomalies over China.
Another one is the use of IAP 9L AGCM, instead of IAP 2L AGCM. After the assessment of the predictive skill of IAP 9L AGCM from the 30-year hindcast experiments (Lang et al., 2003a), IAP 9L AGCM has been applied to the experimental seasonal and extra-seasonal prediction of summer rainfall anomalies for 2002, along with the climate anomalies for winter of 2002/2003, and spring of year 2003, and the prediction results show to be quite encouraging (Wang et al., 2003；Lang et al., 2003a).
III. RESEARCHES ON THE CLIMATE PREDICABLILITY
1. Impact of Atmospheric Initial Conditions
There are several studies suggesting the important role of initial atmospheric conditions on the climate anomalies in the subsequent seasons (e.g., Sun and Sun, 1995; Hoerling and Ting, 1995). By using IAP DCP-I, four sets of integrations have been performed in order to assess the impact of initial conditions on the prediction of summer climate anomalies, and experiment results show that, the impact of the initial conditions is quite significant in middle and high latitude, however, they find that its impact is relative local compared with the impact from SST (Zhao and Guo, 2000).
By using the CCSR/NIES AGCM, Wang et al. (2000) investigate the impact of atmospheric initial conditions (hereafter, AIC), and they suggest that, the initial atmospheric anomalies in April may have strong impacts on the simulated Eurasian flow pattern and the precipitation anomalies in the subsequent summer at high latitude, especially over East Asian region. This conclusion has been further substantiated by the two sets of 30-year hindcast experiments using IAP 9L AGCM, where they demonstrated that the impact of AIC could be global, and this impact is particularly important for those years with severe climate anomalies (Lang et al., 2003b).
Based on the reanalysis data from NCEP/NCAR and other observational data, the relationship between summer rainfall over East Asia and the Mascarene high (MH) and Australian high (AH) is demonstrated through correlation analysis and case study (Xue et al., 2003), and they find that, when MH intensifies from boreal spring to summer, (i.e., from austral autumn to winter), there is more rainfall over regions from the Yangtze River valley to Japan, in contrast, less rainfall is found over southern China and western Pacific to the east of Taiwan, and most of regions in mid-latitudes of East Asia. Compared with MH, the effect of AH on summer rainfall in East Asia is limited to localized regions; there is more rainfall over southern China with the intensification of AH. This result is of great importance to revealing the physical mechanism of interannual variability of East Asian summer monsoon and prediction of summer precipitation in China.
2. Impact of Sea Surface Conditions
The strong relationship between ENSO phenomena and the East Asian summer monsoon has been revealed by many studies (e.g., Yang and Yuan, 1995; Tao and Zhang, 1998). Through the case study for year 1998, Wang (2001) find that the SST anomalies over equatorial middle and East Pacific Ocean play key roles in the severe flooding situations over the Yangtze river Valley for year 1998, and this is one of the reasons why the SSTA over tropical Pacific Ocean has been used as the main predictor for summer rainfall anomalies over China. Wang et al. (1999) further suggest the intermittence of this relationship from both the observation and numerical studies, and they reveal that there appears to be little connection between ENSO and monsoon during the periods of low ENSO variance, and there are also periods when the variability of monsoon seems to lead ENSO variability.
By using IAP 9L AGCM, Lin et al. (1999) found that, the sea surface temperature (SST) anomalies over northern Pacific Ocean can also have significant impact on the summer rainfall anomalies over East Asian region. Furthermore, they point out that, due to the smoothing effects of the traditional scheme of interpolating the monthly mean SST to the daily SST distribution, it would be better to make use of the high frequency (e.g., daily or weekly) SST data during the real-time seasonal prediction.
Meanwhile, Zhao and Guo (2000) found that, for the severe summer flooding events for year 1998, the influence from the sea surface temperature anomalies over Indian Ocean is of most importance. Thus, in order to further increase the predictability of East Asian summer monsoon rainfall, the oceanic model used in predictions needs to be extended to cover at least the whole of Indian Ocean region.
3. Impact of Land Surface Conditions
The impact of land surface processes on the East Asian monsoon climate has been illustrated by many studies (e.g., Lin et al., 1996; Yang and Lau, 1998). By introducing the improved climate model with modified land surface parameterization, Lin et al. (1999) find that the prediction skill can be improved in some extent. Wang (2001) conducted ensemble integrations by IAP 9L AGCM to investigate the mechanism of severe flood over Yangtze River valley for 1998, and he found that, the excessive snow cover over Tibetan Plateau during winter and spring time can lead to the weakening of the South Asia High, and then the weakening of Asian monsoon circulation and its associated monsoon precipitation over the lower reach of Yangtze River valley. This result is further proved by the similar research (Zhang and Tao, 2001).
Meanwhile, the impact of soil moisture on the East Asian summer monsoon has also been revealed by Wang (2001), he suggested that the wet soil conditions during springtime over South China can lead to the weakening of the East Asian summer monsoon, and then the increased monsoon precipitation over lower reach of Yangtze River valley. With the observational data over Huaihe River Basin from HUBEX (HUaihe river Basin EXperiment) field experiment, Lin et al. (2001) further point out that, the sensitivity of the land surface model to the initial soil moisture is quite strong in late spring and summer over Huaihe River Basin, however, this sensitivity becomes relatively weaker in Autumn.
All the above studies demonstrated the importance of the initial land surface conditions for the seasonal prediction of summer rainfall anomalies over East Asia monsoon region. However, at present time, the model climatological distribution of the land surface characteristics (e.g., soil moisture, snow cover, etc.) is adopted as the initial conditions when conducting the real-time seasonal climate perditions, and this might be one of the reasons why the magnitude of the predicted summer rainfall anomalies are generally weaker than the observation (Lin et al., 1998). So, the development of the land data assimilation system, especially the development of the initialization scheme of soil moisture, is urgently needed, in order to further improve the skill of seasonal climate prediction.
IV. PRACTICAL ISSUES RELATED TO SEASONAL PREDICTION
1. Ensemble Technique
Due to the chaotic influence in the climate system, the ensemble technique is needed for the seasonal climate prediction. However, it should be further pointed out that, the superposition of “small” perturbations on the initial conditions for an “individual day”, and then taking the ensemble mean of predictions with different initial conditions mentioned above can not improve the skill of seasonal to extra-seasonal climate predictions due to the large uncertainties of present climate models. Practically, in IAP/CAS, we take ensemble mean of multiple predictions initiated from different days, so the difference between individual initial conditions could be “large enough” (Zeng et al., 2003).
Zhao (2001) provided the estimates of the how many runs are needed for an ensemble for different regions of China by using IAP DCP-I, and she suggested that, the number of the integrations might be around 10 over Southeast China where the precipitation variation is dominated by the SST anomalies over tropical ocean. However, with the increased influence from the atmospheric internal variation in higher latitude regions, the minimum number of the integrations also increases, with about 20 members in North and Northeast China.
The advantage of the ensemble prediction has been further improved by Lin et al. (2002). Figure 4 gives the correlation coefficient of predicted and observed percentage summer rainfall anomalies over China for 2000, and we can find that, the anomaly correlation coefficient is around 0.25, which is higher than most of the individual integration.
Fig.3. Correlation coefficient of predicted and observed percentage summer rainfall anomalies over China for year 2000, each bar indicates one individual forecast initiated from different date of February, 2000, the ACC for the ensemble mean is also shown as indicated in the figure.
Furthermore, Yuan et al. (2000) proposed a new ensemble technique, where the different weights have been adopted for different individual runs, according to the deviations of each ensemble member from the arithmetic ensemble mean, and to the different date of each ensemble member. And they find from the 15-year hindcast experiment that, such a new ensemble technique can practically increase the predictive skill of the extra-seasonal f summer climate anomalies over China.
2. Correction Method
Due to the improper representation of the physical processes within the current generation of climate models, there always exist some systematic errors for the model climatology, so the correction technique is needed for the real-time climate predictions, as proposed by Zeng et al. (1994).
As the quasi-biennial oscillation is a strong signal in the atmosphere, Wang et al. (2000) proposed a new correction method name “WZZM”, which can both correct the amplitude of model variability and the sign of the climate anomaly through the existing quasi-biennial oscillation. They found that, after the correction by WZZM method, the predictive skill is much higher than the uncorrected prediction, which indicate the importance of the correction technique for the short-term climate prediction.
Currently, Chen et al. (2003) proposed a new correction method based on the ENSO cycle (Hereafter referred to as “CM_ENSO”). From the 21-year hindcast experiment, they divide the total number of years into three categories, e.g., El Nino year, La Nina year and normal year, and then the new correction system has been established which distinguishes the above three categories years based on the hindcast experiment. Table 1 gives that the predictive skill after the new correction technique, also shown is the predictive skill after the original correction system as described in Lin et al. (1998) (Hereafter referred to as “CM_NORM”), where the corrections are kept the same for all the hindcast years. Comparison results show that, with the new correction method “CM_ENSO”, the predictive skill is much higher than the traditional “CM_NORM” method. For example, the anomaly correlation coefficient (ACC) is only 0.06 over the whole China with “CM_NORM” correction method; however, it can reach as high as 0.12 with the “CM_ENSO” correction system.
Table 1. Comparison of Predictive Skill for the Summer Rainfall Anomalies over China with “CM_NORM” and “CM_ENSO” Correction Method Respectively