1. Introduction
Skillful seasonal streamflow prediction can help water managers to effectively manage or mitigate the hazards of droughts or floods. The skill of seasonal streamflow prediction is affected by two main factors: the predictive skill of seasonal climate forecasts (precipitation and temperature) and the initial land surface conditions (Wood and Lettenmaier 2008; Mahanama et al. 2008, 2012; Li et al. 2009; Koster et al. 2010).
A number of analyses (e.g., Wood and Lettenmaier 2008; Li et al. 2009; Pagano et al. 2009; Mahanama et al. 2008, 2012; Koster et al. 2010) focused on the seasonal streamflow forecast skill derived from accurate estimates of initial land surface conditions in a region. Other studies used multiple regression approaches (Maurer and Lettenmaier 2003; Bierkens and van Beek 2009) or seasonal climate models (Wood et al. 2005) to justify seasonal predictable climate signals, including sea surface temperature (SST) and related large-scale atmospheric circulation patterns, spring snowpack anomalies, meteorological forcing, and initial land surface states. All these signals can improve seasonal climate forecasts and then improve seasonal streamflow prediction further.
The scientific underpinnings of seasonal climate models (Shukla et al. 2000) are based on observed seasonal variability. The observed seasonal variability is often idealized as a linear combination of climate noise and climate signal. The climate noise is presumed to consist only of variability due to weather systems that are unpredictable beyond 2 weeks (Lorenz 1963). The climate signal (Palmer and Anderson 1994) is presumed to arise from changing external conditions, including changing SSTs, volcanic activity, solar activity, or boundary conditions, for example, soil moisture, sea ice, and snow cover. The amount by which the actual variance exceeds the noise variance is considered a measure of the potential predictability (Madden and Shea 1978).
Atmospheric general circulation models (AGCMs) forced by observed SSTs are the tools used to understand climate variability and seasonal predictability (e.g., Kumar et al. 1996; Zwiers 1987; Rowell 1998; Zheng and Frederiksen 1999). A comparison of the model-simulated and observed spatial distributions and magnitudes of the two components, climate noise and climate signal, may provide useful diagnostics to identify AGCM deficiencies (Zheng et al. 2004). In terms of simulating rainfall variability and evaluating seasonal predictability, the superiority of coupled GCM (CGCM) simulations over AGCMs simulations has been shown by Kug et al. (2008), among others. CGCMs account for interactions between the atmosphere and ocean, while AGCMs simulations are performed using only atmospheric physics laws with prescribed SST boundary conditions.
Remarkable improvements in seasonal precipitation and temperature forecasts have been achieved by a variety of CGCMs, including ENSEMBLES from the European Centre for Medium-Range Weather Forecasts (Weisheimer et al. 2009), and the second version of the Climate Forecast System (CFSv2) from the National Centers for Environmental Prediction (NCEP; Saha et al. 2014). Yuan et al. (2011) analyzed the capability of CFSv2 for global predictions of surface air temperature and precipitation and compared the predictive skill of the CFSv1, CFSv2, and ENSEMBLES models. They found that CFSv2 shows a significant skill enhancement compared to CFSv1 and comparable performance to the ENSEMBLES global climate models. Luo et al. (2013) also evaluated summer temperature and precipitation predictive skill of CFSv2 over China using deterministic and probabilistic metrics. The deterministic predictive skill for summer temperature is high while that for summer precipitation is low. The probabilistic predictive skill is limited for both summer temperature and precipitation. Yuan et al. (2013) also evaluated predictive skill of CFSv2 and CFSv1 basin-averaged precipitation over the conterminous United States and found that CFSv1 and CFSv2 have higher squared correlation and smaller error than ensemble streamflow prediction for monthly precipitation. They also found that the forecasts that are conditional on ENSO have further improvements over southern basins out to 4 months. Furthermore, CFSv2 improves predictive skill over many regions compared with CFSv1, especially during winter and spring. Both CFSv1 and CFSv2 are limited by the coarse resolution that can be improved through spatial downscaling (Yuan and Liang 2011; Yuan et al. 2012).
We are interested in developing seasonal streamflow prediction capability in China using climate predictions from dynamic climate models such as CFSv2. The objective of this paper is to evaluate the predictive skill of temperature and precipitation from CFSv2 over China so that we have a better understanding of how useful these climate predictions are for hydrological predictions. In this paper, we assess the predictive skill of precipitation and surface air temperature in 17 large regions of China using the 26-yr reforecast dataset from CFSv2. The paper is organized as follows. A brief description of the CFSv2 and observations is given in section 2. Section 3 presents the evaluation method and metrics. The results and discussion are presented in section 4, followed by conclusions in section 5.
2. Data used in this study
The first version of the NCEP Climate Forecast System (CFSv1) was released for operational use at NCEP in August 2004 and was the first quasi-global, fully coupled atmosphere–ocean–land model used at NCEP for seasonal predictions (Saha et al. 2006). The second version, CFSv2, became operational at NCEP in March 2011. The reforecast configuration for CFSv2 (T126L64) includes four 9-month hindcast runs, a single (0000 UTC cycle) one-season (~123 day) hindcast run, and three (0600, 1200, and 1800 UTC cycles) 45-day hindcast runs. The four 9-month hindcasts are used in this paper. The 9-month hindcasts are initiated from every fifth day and run from all four cycles (0000, 0600, 1200, and 1800 UTC cycles) of that day, beginning 1 January each year, over a 29-yr period from 1982 to 2010. The CFSv2 used in the reforecast consists of the NCEP Global Forecast System at T126 (~0.937°) resolution, the Geophysical Fluid Dynamics Laboratory Modular Ocean Model, version 4.0, at 0.25°–0.5° grid spacing coupled with an interactive three-layer sea ice model, the four-layer Noah land surface model, and historical prescribed (i.e., rising) CO2 concentrations (Saha et al. 2014). In this study, data (total precipitation rate and temperature at 2 m saved in 6-hourly time series format) were retrieved from CFSv2 to get daily precipitation and mean, maximum, and minimum temperature at 2 m from 1982 to 2009.
Gauge-based daily precipitation gridded on a 0.5° latitude–longitude grid over East Asia (5°–60°N, 65°–155°E) for the period 1978–2007 (Xie et al. 2007) and station-based daily mean, minimum, and maximum temperature on a 0.5° latitude–longitude grid over mainland China for the period 1961–2009 prepared at the National Climate Center of the China Meteorological Administration (Xu et al. 2009) are used to verify CFSv2 forecasts. The common period of 26 years (i.e., 1982–2007) is used in the evaluation, and all datasets are also regridded to a common 1° grid over mainland China.
In this study, China was divided into 17 large hydroclimatic regions (Fig. 1) based on the watershed divisions standard. The climate classifications, for example, Köppen–Geiger climate types (Peel et al. 2007), were also considered in shaping the boundaries of these regions to ensure nearly uniform regional climatic characteristics. A summary of these 17 hydroclimatic regions is provided in Table 1, including a list of climate type, annual precipitation, and the regional areas. In the evaluation, both forecast and observations of temperature and precipitation are spatially averaged for each hydroclimatic region; therefore, the evaluation really focuses on the regional scales.
The 17 large hydroclimatic regions in China.
Citation: Journal of Hydrometeorology 15, 4; 10.1175/JHM-D-13-0208.1
A summary of 17 hydroclimatic regions in China.
3. Method
4. Results and discussion
The objective of this study is to provide a general overview of the predictive skill of the CFSv2 precipitation and temperature forecasts for different lead times throughout the seasonal cycle for all 17 hydroclimatic regions. Seasonal cycle here is defined as 12 overlapping seasons each spanning 3 months. For example, early winter refers to the months of November–January (NDJ; i.e., Julian days from 305 to 31), winter refers to December–February (DJF; from 335 to 59), and late winter refers to January–March (JFM; from 1 to 90). Other seasons are defined in a similar fashion.
The correlation coefficient between the CFSv2 forecasts and observed variables is a measure of the predictive pattern similarity. The 17 correlation coefficient arrays between CFSv2 forecasted and observed precipitation at all 17 hydroclimatic regions are shown as 17 plots in Fig. 2. The correlation coefficient between CFSv2 forecasts and observations was computed on the basis of CFSv2 9-month hindcasts and corresponding observations over the 26-yr period from 1982 to 2007. The vertical axis in Fig. 2 denotes the target date of the CFSv2 forecasts, while the horizontal axis denotes different lead times. The lead times are defined for seven overlapping forecast events each spanning 3 months, similar to how the seasons are defined. For example, lead time 1 corresponds to the period from day 1 to day 90, lead time 2 from day 31 to day 120, lead time 3 from day 61 to day 150, and so on. We note that the correlation coefficients range from insignificant zero values to values as high as 0.6–0.7 for some regions in certain seasons. For example, in region 9 (upper Yangtze River), the correlation coefficient reaches a value close to 0.7. We can see that there are two seasonal predictive pattern maxima (yellow slanted bar): one maximum is from late summer [July–September (JAS)] to late autumn [October–December (OND)] and the other is from winter (DJF) to spring [March–May (MAM)]. These two seasonal predictive pattern maxima are less dependent on prediction lead time. That means we can predict seasonal precipitation from late summer (JAS) to late autumn (OND) and from January and during winter (DJF) to spring (MAM) from last year’s July regardless of lead times. Almost all of the 17 plots demonstrate two seasonal predictive pattern maxima: one maximum is from late summer (JAS) to late autumn (OND) and the other is from winter (DJF) to spring (MAM). The correlation coefficients in the northern and western regions (except region 1, i.e., inland rivers in Xinjiang) are higher than in the southeastern regions. Moreover, the correlation coefficients for winter (DJF) and spring (MAM) are generally less than those for late summer (JAS) and late autumn (OND), while in only one region, region 13 (Yangtze River), the correlation coefficient for winter (DJF) and spring (MAM) are higher than for late summer (JAS) and late autumn (OND). Apparently, central regions 4 (Yellow River), 6 (Hai River), and 10 (Huai River) and southern region 16 (Pearl River) have no predictive capability. We conclude that these regions, located in the middle of the Yellow River, Huai River, Hai River, and Pearl River, are difficult for CFSv2 to predict for all seasons.
Correlation coefficients of CFSv2 forecasts as a function of lead (horizontal) and target date (vertical) for precipitation over the 17 regions of China.
Citation: Journal of Hydrometeorology 15, 4; 10.1175/JHM-D-13-0208.1
Studies have also shown that the primary source of predictability of global precipitation is still the ENSO variability (Wang et al. 2009). However, CFSv2 show low predictive skill in all 17 hydroclimatic regions in China during summer. The notion that climate can be modeled and predicted by prescribing the lower boundary conditions is inadequate for validating models and predicting summer monsoon rainfall (Wang et al. 2005) because the East Asian summer monsoon (EASM) and related seasonal rain belts assume significant variability at intraseasonal, interannual, and interdecadal time scales (Ding and Chan 2005). Two external forcings, Pacific and Indian Ocean SSTs and the snow cover in Eurasia and the Tibetan Plateau, are primary contributing factors to physical processes and mechanisms related to the EASM. However, the internal variability of the atmospheric circulation [intraseasonal variability, such as 10–20 days and 30–60 days (i.e., Madden–Julian oscillation)] also affects the activity of EASM (Ding and Chan 2005). Thus, the potential predictability of the Asian monsoon climate, especially for precipitation, is low in seasonal time scales as the contribution of the boundary forcing is relatively low and that of the internal dynamics is relatively large (Ying et al. 2013). The CFSv2 captures the general features of EASM flows in both the lower and upper troposphere (Jiang et al. 2013a). However, it predicts a weaker-than-observed western Pacific subtropical high (WPSH) and monsoon trough over the Indian subcontinent, as well as a stronger-than-observed Somali jet and westerlies over the equatorial central and eastern Indian Ocean in the lower troposphere. Compared with observations, the CFSv2 has biases in the southerly flow over South Asia in the lower troposphere and the anticyclonic circulation over the Asian continent and the western North Pacific in the upper troposphere. Their results also show that the bias of CFSv2 in predicting monsoon flow is dynamically consistent with the deficiency of the model in predicting monsoon precipitation (Jiang et al. 2013a).
CFSv2 also shows low predictive skill in all 17 hydroclimatic regions in China, except the middle and lower Yangtze River and southeast rivers during winter. This is because the East Asian winter monsoon (EAWM) exerts significant impacts on the weather and climate patterns in and outside East Asia (Jiang et al. 2013b). Zhou and Wu (2010) investigate respective impacts of the EAWM and El Niño–Southern Oscillation (ENSO) on winter (November–March) rainfall in China by removing the interdependence between the EAWM and ENSO. They found that circulation and temperature anomalies over the tropics and midlatitude North Pacific are mainly induced by ENSO, and those over midlatitude Asia are closely linked to the EAWM. The warm, ENSO-induced, lower-level southwesterly winds deflect from the southeast coast of China, and thus the influence of ENSO on winter rainfall is mainly in southern China. The lower-level, southerly winds associated with a weak EAWM penetrate northward over eastern China, and thus the EAWM influences winter rainfall in eastern China. The predictability of EAWM shows large differences between the southern portion and the northern portion of East Asia (Jiang et al. 2013b). The southern EAWM component, whose variability is mainly affected by ENSO, exhibits larger predictability. However, smaller predictability is found for the northern EAWM component, which is mostly governed by the extratropical atmospheric circulation such as the Arctic Oscillation, which has been poorly predicted by CFSv2. Moreover, the prediction of EAWM by CFSv2 over land is worse than that over oceans.
Correlation coefficients between CFSv2 forecasted and observed temperatures at the target date of the CFSv2 forecasts and different lead times for the 17 hydroclimatic regions are shown in Fig. 3. The patterns for CFSv2 forecasted maximum and minimum temperatures are similar to that of the mean temperature and are therefore not shown here. Figure 3 shows that the correlation patterns of temperature forecasts are highly similar to that of precipitation forecasts, even though the correlation coefficients of temperature forecasts are visibly higher than that of precipitation forecasts, with the highest value above 0.9.
As in Fig. 2, but for mean temperature.
Citation: Journal of Hydrometeorology 15, 4; 10.1175/JHM-D-13-0208.1
The seasonal predictive performance indices of CFSv2 forecasts for 17 hydroclimatic regions of China are plotted using the Taylor diagrams in Fig. 4 (precipitation) and Fig. 5 (mean temperature). These figures summarize both aspects of model performance, that is, the correlation coefficients and normalized standard deviation. The correlation between the prediction and observed reference variable is given by the azimuthal position. The radial distance from the origin is proportional to the normalized standard deviation. To display the different seasonal predictive performance, four different colors are used here to denote four different seasons, that is, the colors blue, green, red, and brown represent winter, spring, summer, and autumn, respectively. Three shapes are used to represent three periods in one season, for example, triangles, circles, and rectangles in green stand for early spring, spring, and late spring, respectively. For any given correlation, the score should increase as the forecast variance approaches the observed variance. The coordinate position marked by “REF” on the x-axes implies the perfect match between prediction and observation. As seen in Fig. 4, the correlation coefficients of precipitation forecasts are almost always below 0.7. The normalized standard deviations also diverge from reference deviations. The relatively high normalized standard deviations in region 5 (upper Yellow River), 7 (Songhua River), 12 (southwest rivers in Yunnan), and 13 (Yangtze River) indicate that the forecasts tend to exaggerate the dispersion from the mean and vice versa in region 16. The northern and western regions 2 (inland rivers in northern Tibet), 3 (inland rivers in Inner Mongolia), 5 (upper Yellow River), 7 (Songhua River), 8 (Liao River), 9 (upper Yangtze River), 11 (southwest rivers in southern Tibet), 12 (southwest rivers in Yunnan), and 13 (Yangtze River) have relatively good seasonal predictive performance for precipitation, especially from late summer [August–October (ASO)] to early autumn [September–November (SON)]. The humid regions—regions 13 (Yangtze River), 14 (middle Yangtze River), 15 (lower Yangtze River), and 17 (southeast rivers)—tend to have good seasonal predictive performance for precipitation during winter. Seasonal predictive performance for spring is poor in all regions. Since mean temperature has higher correlation coefficients than maximum and minimum temperature, we again only use mean temperature to represent predictive skill of temperature. As shown in Fig. 5, the seasonal predictive performance of the mean temperature is better than that of precipitation in all 17 regions because the mean temperature has high correlation coefficients and deviations close to the reference (i.e., 1). However, the seasonal predictive performance of temperature for summer (June–August), winter (DJF), and early winter (NDJ) are notably worse than for other seasons.
Seasonal predictive performance of CFSv2 precipitation forecasts over the 17 regions of China.
Citation: Journal of Hydrometeorology 15, 4; 10.1175/JHM-D-13-0208.1
As in Fig. 4, but for mean temperature.
Citation: Journal of Hydrometeorology 15, 4; 10.1175/JHM-D-13-0208.1
As the Taylor diagrams show detailed seasonal predictive performance of precipitation and temperature forecasts, it is desirable that an integrated skill score can characterize the predictive skill with a score ranging from zero to one (based on both the pattern similarity and magnitude difference). The seasonal predictive skill score as defined in Eq. (2) for the 17 regions in different seasons are summarized in Fig. 6.
Seasonal predictive skill score of CFSv2 forecasts over the 17 regions of China: (top) precipitation and (bottom) mean temperature.
Citation: Journal of Hydrometeorology 15, 4; 10.1175/JHM-D-13-0208.1
In Fig. 6 (top; precipitation), the predictive skill is notable during early autumn (ASO) and autumn (SON) in only a few regions, that is, regions 2 (inland river in northern Tibet), 5 (upper Yellow River), 7 (Songhua River), 9 (upper Yangtze River), 11 (southwest rivers in southern Tibet), and 12 (southwest rivers in Yunnan). Regions 9 and 13 (Yangtze River) also have some predictive skill during early spring [February–April (FMA)] and spring (MAM). This predictive skill score of CFSv2 precipitation can be verified by comparing it with previous results related to seasonal potential predictability (Zhao et al. 2008; Feng et al. 2011; Ying et al. 2013). As pointed out by Madden and Shea (1978), the actual climatic variance beyond the climatic noise variance in one season indicates this region having potential predictability. To some extent, the small climatic noise variance can also indicate the potential predictability. Zhao et al. (2008) found that the climatic noise variance of seasonal precipitation decreases gradually from south to north and from the coast to inland. Figure 6 (top) also shows that skill score is low during summer, mainly in southeastern regions, that is, regions 16 (Pearl River), 17 (southeast rivers), 15 (lower Yangtze River), 14 (middle Yangtze River), and 10 (Huai River), because the climatic noise variance is largest in these regions during summer. The climatic noise variance during spring is less than during summer mostly in the aforementioned regions. The climatic noise variance during autumn is less than during spring. The climatic noise variance is the smallest during winter. As the climatic noise decreases, the potential predictability increases. Because potential predictability of observed precipitation and predictive skill of CFSv2 precipitation are both low during the summer, this phenomenon means that it is difficult to find the sources of predictability and mechanism for climate forecast models to predict summer precipitation well. Although potential predictability of precipitation is high during winter, the CFSv2 precipitation prediction did not have a comparable predictive skill in all regions. Hence, winter precipitation forecasts of CFSv2 need to be further improved. During spring, there is no predictive skill of CFSv2 predictions in northern and western regions, that is, regions 1 (inland rivers in Xinjiang), 3 (inland rivers in Inner Mongolia), 4 (Yellow River), 8 (Songhua River), and 11 (southwest rivers in southern Tibet). During autumn, the predictive skill of CFSv2 forecasts is very low in region 1 (inland rivers in Xinjiang) and region 10 (Huai River). The above regions during spring and autumn could be improved according to their potential predictability.
In Fig. 6 (bottom), the predictive skill of the mean temperature is high in all regions from late summer (JAS) to late autumn (OND) and from late winter (JFM) to late spring [April–June (AMJ)]. The predictive skill of mean temperature is low in summer and winter. From a spatial perspective, warm and humid regions such as regions 9 (upper Yangtze River), 12 (southwest rivers in Yunnan), 16 (Pearl River), and 17 (southeast rivers) have relatively low predictive skill compared to other regions. According to the previous research on climatic noise and potential predictability of observed atmospheric temperature (Ma and Cao 1999; Le et al. 1999; Zheng et al. 2000), CFSv2 is capable of predicting atmospheric temperature, which means that CFSv2 can simulate the temperature well. The climatic noise of temperature increases with latitude and altitude; the effects are larger in winter than in summer (Ma and Cao 1999). The continental air from Siberia and Mongolia plays a significant role, and the ocean acts as an adjustor and a reducer in the noise, except for the tropical Pacific Ocean in the transitional season months (Ma and Cao 1999). Compared with winter, the temperature noise decrease in spring and autumn is evident. The climate noise in southeastern regions, regions 16 (Pearl River) and 17 (southeast rivers), surpass that in the adjacent inland regions because of the influence of the outskirts of the subtropical Pacific high. In summer, the climatic noise derived from the modified dry, cold air from Siberia and Mongolia often confronts with the wet, warm air from the tropical ocean.
Magnitude differences between CFSv2 forecasts and observations, that is, biases, are presented in Figs. 7 and 8 using the mean bias and the normalized RMSE. Figure 7 (top) shows that seasonal precipitation biases are more intense in humid regions than arid and semiarid regions. From early spring (FMA) to early summer [May–July (MJJ)], the mean seasonal predictions in southwestern regions, that is, regions 11 (southwest rivers in southern Tibet), 12 (southwest rivers in Yunnan), and 13 (Yangtze River), are 2 mm day−1 more than the mean seasonal observation. The mean seasonal precipitation predictions in southeastern regions, that is, regions 10 (Huai River), 15 (lower Yangtze River), 16 (Pearl River), and 17 (southeast rivers) are less than the mean seasonal observation by approximately 1 mm day−1 during early summer (MJJ) to early autumn (ASO). The mean bias is small in arid or semiarid regions, where annual precipitation is less than 600 mm, including regions 1 (inland rivers in Xinjiang), 2 (inland rivers in northern Tibet), 3 (inland rivers in Inner Mongolia), 4 (Yellow River), 6 (Hai River), and 8 (Songhua River). The small biases do not imply better predictive skill in Fig. 6 (top).
As in Fig. 6, but for mean bias.
Citation: Journal of Hydrometeorology 15, 4; 10.1175/JHM-D-13-0208.1
As in Fig. 6, but for normalized RMSE.
Citation: Journal of Hydrometeorology 15, 4; 10.1175/JHM-D-13-0208.1
The cold biases of mean temperature shown in Fig. 7 (bottom) are larger in region 5 (upper Yellow River) and region 11 (southwest rivers in southern Tibet). CFSv2 underpredicts the temperature there because of their high elevation. The warm biases of mean temperature in region 4 (Yellow River) maybe due to inappropriate land cover represented in CFSv2. The summer temperature generally tends to be homogeneous, as found by Luo et al. (2013). CFSv2 underpredicts the temperature in warmer and humid regions and overpredicts the temperature over colder and arid regions. However, the winter temperature does not follow this homogenous trend, as it inversely overpredicts the temperature in warmer and humid regions and underpredicts the temperature over colder and arid regions.
The normalized RMSE in Fig. 8 also measures the differences between CFSv2 forecasts and observations. Twelve arid and semiarid regions show more variation in precipitation than four humid regions during early winter (NDJ) to early spring (FMA; Fig. 8, top). For temperature, in Fig. 8 (bottom), the differences depend more on elevation, that is, regions 5 (upper Yellow River) and 11 (southwest rivers in southern Tibet).
Seasonal ENSO influences on precipitation shown by correlation coefficients between seasonal precipitation and the Oceanic Niño index in the 17 regions for different seasons can be seen in Fig. 9. The Oceanic Niño index is defined as the 3-month running mean of Extended Reconstruction Sea Surface Temperature, version 3 (ERSST.v3b), SST anomalies in the Niño-3.4 region (5°N–5°S, 120°–170°W; Smith et al. 2008), which are derived from the National Oceanic and Atmospheric Administration (NOAA) Climate Prediction Center (CPC). Correlation coefficients between CFSv2 precipitation and the Oceanic Niño index (Fig. 9, bottom) are analogous to the correlation coefficients between observational precipitation and the Oceanic Niño index (Fig. 9, top), which means that CFSv2 has the ability to predict seasonal ENSO influence on precipitation. From early spring (FMA) to late spring (AMJ), seasonal ENSO influences on CFSv2 precipitation are weaker than on observed precipitation, primarily in southeastern humid regions, such as regions 14 (middle Yangtze River), 15 (lower Yangtze River), 16 (Pearl River), and 17 (southeast rivers), while seasonal ENSO influences on CFSv2 precipitation are stronger in southwestern regions, such as regions 11 (southwest rivers in southern Tibet) and 12 (southwest rivers in Yunnan). During late winter (JFM), seasonal ENSO influence is reversed between CFSv2 and observed precipitation in regions 1 (inland rivers in Xinjiang), 14 (middle Yangtze River), 15 (lower Yangtze River), 16 (Pearl River), and 17 (southeast rivers). We infer that improved seasonal ENSO predictions from late winter (JFM) to late spring (AMJ) in CFSv2 precipitation forecasts will improve the precipitation predictive skill score in these regions.
Correlation coefficients between seasonal precipitation and Oceanic Niño index over the 17 regions of China: (top) observations and (bottom) CFSv2 forecasts.
Citation: Journal of Hydrometeorology 15, 4; 10.1175/JHM-D-13-0208.1
5. Conclusions
This study has investigated where and when the precipitation and temperature forecasts possess significant or poor predictive skill in 17 large hydroclimatic regions of China by using the 26-yr reforecasts of the Climate Forecast System, version 2 (CFSv2).
The seasonal predictive skill is quantified with skill scores that relate to correlation coefficient and normalized modeled standard deviation for spatially averaged seasonal precipitation and temperature of each hydroclimatic region. The results show that the predictive skill of precipitation and temperature forecasts from CFSv2 has a stronger dependence on seasons and regions than on lead times. Both precipitation and temperature forecasts show higher predictive skill from late summer (JAS) to late autumn (OND) and from winter (DJF) to spring (MAM). Reasonably good predictive skill of precipitation is primarily found in eight northern and southwestern regions. As expected, the temperature predictive skill is generally much higher than the precipitation predictive skill in all regions. The predictive skill of spring and autumn temperature forecasts is slightly lower in warm and humid southern and southeastern regions. CFSv2 forecasts have the ability to predict influence of ENSO on seasonal precipitation amount, but CFSv2 precipitation forecasts need to be improved during summer and winter. Because of low potential predictability during the EASM and EAWM periods, significant forecast biases are observed in CFSv2 precipitation and temperature forecasts in certain regions. Hence, a better representation of corresponding physical mechanism in CFSv2 will improve the summer and winter precipitation and temperature prediction in China.
The goal of seasonal precipitation and temperature forecast is to provide essential input data in hydrological models for long-term flood and drought early warning. CFSv2 products have some unique advantages over many other similar climate forecast products in that they have a long 30-yr historical reforecast archive that allows hydrologists and other users to make use of them to correct the biases inherent in these reforecasts. Furthermore, CFSv2 is also the operational climate forecast model in NCEP whose real-time predictions are available online in near–real time for all global users to use for real-time applications. The fact that the CFSv2 seasonal precipitation and temperature forecasts have meaningful skill in certain seasons and regions in China is very significant. The skill can be potentially utilized for predicting seasonal water supply outlook and seasonal drought prediction in China and can therefore contribute to China’s water resources management. The next logical step for us is to conduct research in making use of the CFSv2 forecasts in seasonal streamflow predictions.
Acknowledgments
This study was supported by the National Science and Technology Support Plan Program (2013BAB05B04) and National Basic Research Program of China (973Program: 2010CB428402, 2010CB428403).
REFERENCES
Bierkens, M. F. P., and van Beek L. P. H. , 2009: Seasonal predictability of European discharge: NAO and hydrological response time. J. Hydrometeor., 10, 953–968, doi:10.1175/2009JHM1034.1.
Ding, Y., and Chan J. C. L. , 2005: The East Asian summer monsoon: An overview. Meteor. Atmos. Phys., 89, 117–142, doi:10.1007/s00703-005-0125-z.
Feng, X., DelSole T. , and Houser P. , 2011: Bootstrap estimated seasonal potential predictability of global temperature and precipitation. Geophys. Res. Lett., 38, L07702, doi:10.1029/2010GL046511.
Jiang, X., Yang S. , Li Y. , Kumar A. , Liu X. , Zuo Z. , and Jha B. , 2013a: Seasonal-to-interannual prediction of the Asian summer monsoon in the NCEP Climate Forecast System version 2. J. Climate, 26, 3708–3727, doi:10.1175/JCLI-D-12-00437.1.
Jiang, X., Yang S. , Li Y. , Kumar A. , Wang W. , and Gao Z. , 2013b: Dynamical prediction of the East Asian winter monsoon by the NCEP Climate Forecast System. J. Geophys. Res. Atmos., 118, 1312–1328, doi:10.1002/jgrd.50193.
Koster, R. D., and Coauthors, 2010: Contribution of land surface initialization to subseasonal forecast skill: First results from a multi-model experiment. Geophys. Res. Lett., 37, L02402, doi:10.1029/2009GL041677.
Kug, J., Kang I. , and Choi D. , 2008: Seasonal climate predictability with Tier-one and Tier-two prediction systems. Climate Dyn., 31, 403–416, doi:10.1007/s00382-007-0264-7.
Kumar, A., Hoerling M. , Ji M. , Leetmaa A. , and Sardeshmukh P. , 1996: Assessing a GCM’s suitability for making seasonal predictions. J. Climate, 9, 115–129, doi:10.1175/1520-0442(1996)009<0115:AAGSFM>2.0.CO;2.
Le, Q., Cao J. , Lin Z. , and Ma K. , 1999: Climatic noise and potential predictability of monthly mean temperature over China (in Chinese). Acta Meteor. Sin., 57, 604–612.
Li, H., Luo L. , Wood E. F. , and Schaake J. , 2009: The role of initial conditions and forcing uncertainties in seasonal hydrologic forecasting. J. Geophys. Res., 114, D04114, doi:10.1029/2008JD010969.
Lorenz, E. N., 1963: Deterministic nonperiodic flow. J. Atmos. Sci., 20, 130–141, doi:10.1175/1520-0469(1963)020<0130:DNF>2.0.CO;2.
Luo, L., Tang W. , Lin Z. , and Wood E. , 2013: Evaluation of summer temperature and precipitation predictions from NCEP CFSv2 retrospective forecast over China. Climate Dyn., 41, 2213–2230, doi:10.1007/s00382-013-1927-1.
Ma, K., and Cao J. , 1999: Climatic noise and potential predictability of monthly mean temperature over China. Meteor. Atmos. Phys., 69, 231–237, doi:10.1007/BF01030424.
Madden, R. A., and Shea D. J. , 1978: Estimates of the natural variability of time-averaged temperatures over the United States. Mon. Wea. Rev., 106, 1695–1703, doi:10.1175/1520-0493(1978)106<1695:EOTNVO>2.0.CO;2.
Mahanama, S. P. P., Koster R. D. , Reichle R. H. , and Zubair L. , 2008: The role of soil moisture initialization in subseasonal and seasonal streamflow prediction—A case study in Sri Lanka. Adv. Water Resour., 31, 1333–1343, doi:10.1016/j.advwatres.2008.06.004.
Mahanama, S. P. P., Livneh B. , Koster R. , Lettenmaier D. , and Reichle R. , 2012: Soil moisture, snow, and seasonal streamflow forecasts in the United States. J. Hydrometeor., 13, 189–203, doi:10.1175/JHM-D-11-046.1.
Maurer, E. P., and Lettenmaier D. P. , 2003: Predictability of seasonal runoff in the Mississippi River basin. J. Geophys. Res., 108, 8607, doi:10.1029/2002JD002555.
Pagano, T. C., Garen D. C. , Perkins T. R. , and Pasteris P. A. , 2009: Daily updating of operational statistical seasonal water supply forecasts for the western U.S. J. Amer. Water Resour. Assoc., 45, 767–778, doi:10.1111/j.1752-1688.2009.00321.x.
Palmer, T. N., and Anderson D. L. T. , 1994: The prospects for seasonal forecasting—A review paper. Quart. J. Roy. Meteor. Soc., 120, 755–793, doi:10.1002/qj.49712051802.
Peel, M. C., Finlayson B. L. , and McMahon T. A. , 2007: Updated world map of the Köppen–Geiger climate classification. Hydrol. Earth Syst. Sci., 11, 1633–1644, doi:10.5194/hess-11-1633-2007.
Rowell, D. P., 1998: Assessing potential seasonal predictability with an ensemble of multidecadal GCM simulations. J. Climate, 11, 109–120, doi:10.1175/1520-0442(1998)011<0109:APSPWA>2.0.CO;2.
Saha, S., and Coauthors, 2006: The NCEP Climate Forecast System. J. Climate, 19, 3483–3517, doi:10.1175/JCLI3812.1.
Saha, S., and Coauthors, 2014: The NCEP Climate Forecast System version 2. J. Climate, 27, 2185–2208, doi:10.1175/JCLI-D-12-00823.1.
Shukla, J., and Coauthors, 2000: Dynamical seasonal prediction. Bull. Amer. Meteor. Soc., 81, 2593–2606, doi:10.1175/1520-0477(2000)081<2593:DSP>2.3.CO;2.
Smith, T. M., Reynolds R. W. , Peterson T. C. , and Lawrimore J. , 2008: Improvements to NOAA’s historical merged land–ocean surface temperature analysis (1880–2006). J. Climate, 21, 2283–2296, doi:10.1175/2007JCLI2100.1.
Taylor, K. E., 2001: Summarizing multiple aspects of model performance in a single diagram. J. Geophys. Res., 106, 7183–7192, doi:10.1029/2000JD900719.
Wang, B., Ding Q. , Fu X. , Kang I. , Jin K. , Shukla J. , and Doblas-Reyes F. , 2005: Fundamental challenge in simulation and prediction of summer monsoon rainfall. Geophys. Res. Lett., 32, L15711, doi:10.1029/2005GL022734.
Wang, B., and Coauthors, 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., and Coauthors, 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.
Wood, A. W., and Lettenmaier D. P. , 2008: An ensemble approach for attribution of hydrologic prediction uncertainty. Geophys. Res. Lett., 35, L14401, doi:10.1029/2008GL034648.
Wood, A. W., Kumar A. , and Lettenmaier D. P. , 2005: A retrospective assessment of National Centers for Environmental Prediction climate model–based ensemble hydrologic forecasting in the western United States. J. Geophys. Res., 110, D04105, doi:10.1029/2004JD004508.
Xie, P., Chen M. , Yang S. , Yatagai A. , Hayasaka T. , Fukushima Y. , and Liu C. , 2007: A gauge-based analysis of daily precipitation over East Asia. J. Hydrometeor., 8, 607–626, doi:10.1175/JHM583.1.
Xu, Y., Gao X. , Shen Y. , Xu C. , Shi Y. , and Giorgi F. , 2009: A daily temperature dataset over China and its application in validating a RCM simulation. Adv. Atmos. Sci., 26, 763–772, doi:10.1007/s00376-009-9029-z.
Ying, K., Zheng X. , Quan X. , and Frederiksen C. S. , 2013: Predictable signals of seasonal precipitation in the Yangtze–Huaihe River Valley. Int. J. Climatol., 33, 3002–3015, doi:10.1002/joc.3644.
Yuan, X., and Liang X.-Z. , 2011: Improving cold season precipitation prediction by the nested CWRF-CFS system. Geophys. Res. Lett., 38, L02706, doi:10.1029/2010GL046104.
Yuan, X., Wood E. F. , Luo L. , and Pan M. , 2011: A first look at Climate Forecast System version 2 (CFSv2) for hydrological seasonal prediction. Geophys. Res. Lett., 38, L13402, doi:10.1029/2011GL047792.
Yuan, X., Liang X. , and Wood E. , 2012: WRF ensemble downscaling seasonal forecasts of China winter precipitation during 1982–2008. Climate Dyn., 39, 2041–2058, doi:10.1007/s00382-011-1241-8.
Yuan, X., Wood E. F. , Roundy J. K. , and Pan M. , 2013: CFSv2-based seasonal hydroclimatic forecasts over the conterminous United States. J. Climate, 26, 4828–4847, doi:10.1175/JCLI-D-12-00683.1.
Zhao, X., Wu H. , and He L. , 2008: Estimate of climatic noise and potential predictability of seasonal precipitation in China (in Chinese). J. Appl. Meteor. Sci., 31, 819–827.
Zheng, X., and Frederiksen C. S. , 1999: Validating interannual variability in an ensemble of AGCM simulations. J. Climate, 12, 2386–2396, doi:10.1175/1520-0442(1999)012<2386:VIVIAE>2.0.CO;2.
Zheng, X., Nakamura H. , and Renwick J. A. , 2000: Potential predictability of seasonal means based on monthly time series of meteorological variables. J. Climate, 13, 2591–2604, doi:10.1175/1520-0442(2000)013<2591:PPOSMB>2.0.CO;2.
Zheng, X., Sugi M. , and Frederiksen C. S. , 2004: Interannual variability and predictability in an ensemble of climate simulations with the MRI-JMA AGCM. J. Meteor. Soc. Japan,82, 1–18, doi:10.2151/jmsj.82.1.
Zhou, L.-T., and Wu R. , 2010: Respective impacts of the East Asian winter monsoon and ENSO on winter rainfall in China. J. Geophys. Res., 115, D02107, doi:10.1029/2009JD012502.
Zwiers, F. W., 1987: A potential predictability study conducted with an atmospheric general circulation model. Mon. Wea. Rev., 115, 2957–2974, doi:10.1175/1520-0493(1987)115<2957:APPSCW>2.0.CO;2.
