Simulated and projected climate extremes in the Zhujiang River Basin, South China, using the regional climate model COSMO-CLM

Authors


ABSTRACT

This study presents a detailed analysis of simulated and projected climate extremes in the Zhujiang River Basin (ZRB). Daily output from the regional climate model COSMO-CLM (CCLM), driven by the ECHAM5 general circulation model, is used. The hindcast simulation covers the period from 1961 to 2000 while the projection concentrates on the near future period from 2011 to 2050. Spatio-temporal statistical characteristics are investigated for three temperature and three precipitation indicators. The six simulated annual and monthly indicators are statistically compared with synoptic observations. The analysis is based on daily values of 195 grid points and 192 meteorological stations.

The findings are presented and interpreted in terms of the model's capability. Compared to observations, slightly higher values for temperature indicators and slightly lower values for precipitation indicators are simulated. With the resulting good similarities in the spatial variation and trends, we conclude that CCLM is able to satisfyingly reproduce climate extreme for the simulated period. Therefore, our analyses show that CCLM can be used to project climate extremes in the ZRB for the period from 2011 to 2050. The projected changes indicate warmer and wetter conditions in the northern and southern regions, especially in winter and spring. This includes more intense rainfall events, which might potentially increase the risk of flooding in the central parts of the basin in these seasons. Warmer and dryer conditions can be expected in the western and eastern parts of the region, especially in summer and fall. These lower precipitation amounts but warmer temperatures will probably increase the evapotranspiration, which potentially leads to a higher risk of drought. Regarding these findings in climate extremes, adverse consequences in various sectors, such as agriculture, water, and energy should be anticipated.

1. Introduction

Climatic changes have huge potential impacts on the socio-economic welfare and peoples living conditions. Especially, changes in the frequency and magnitude of extreme weather events are an important driver for economic and society changes. Changes in precipitation pattern can lead to both higher drought and higher flood risk, which are potentially linked to high socio-economic costs. In recent years, climate change and its implications were investigated in numerous studies (Adger et al., 2007; Klein Tank et al., 2009).

In China, the fast growing population and industrialization have increased the potential for economic damages from severe events due to climate change (Feng et al., 2007). According to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC AR4) and China's National Assessment Report on Climate Change increases in annual temperature, total precipitation, and heavy precipitation events have been observed in China and are very likely to increase in the future (Qian and Lin, 2005; Trenberth et al., 2007; Ding et al., 2007; Liu et al., 2009; Fischer et al., 2011a, 2011b; Gemmer et al., 2011; Jiang et al., 2011; Xu et al., 2011).

Future climate projections give important estimations on potential changes in magnitudes and frequencies of extreme events (from now on defined as climate extremes) and their related hazardous consequences. The analysis of changes in climate conditions (e.g. variability and extremes) at different spatial and temporal scales is of great benefit to national and regional adaptation strategies. Global circulation models (GCM's) are known to be in good agreement with observed magnitudes and tendencies of temperature while it is believed that the coarse resolution results in poor agreement especially with the precipitation patterns (Kharin et al., 2007). High resolution regional climate models (RCM's) represent a huge improvement in modelling small scale and local effects, especially precipitation (Meehl et al., 2000; Pal et al., 2007; Rockel and Geyer, 2008; Kunkel et al., 2010). A number of studies utilize RCM's to investigate climate change impacts at high resolution for European, American, and Asian regions (Trenberth et al., 2007; Sillmann and Roeckner, 2008; Alexander and Arblaster, 2009).

Changes in the intensity and frequency of climate extremes are of major concern for socio-economic welfare. Hence, it is of high importance for regional institutions of various sectors (e.g. water, industry, and agriculture) to be aware of current and future climate extremes related risks within their area. The water sector bears climate change related risks such as floods and droughts and the river basin scale emphasizes the logical spatial extent for a study region. Focusing on climate characteristics and extremes in China, RCM's were mostly applied to analyze future changes in climate conditions on a larger scale (Wang et al., 2003; Ding et al., 2006; Gao et al., 2008; Xu et al., 2009; Chen et al., 2011). To our knowledge, there is no study published dealing with the ability of RCM's to simulate climate extremes in the Zhujiang River Basin (ZRB) comprehensively.

The ZRB in South China covers an area of approximately 450 000 km2 with a population of more than 166 million. The region is currently one of the most economically prosperous areas of China, with very high development rates, and one of China's highest gross domestic product per capita of more than 40 000 Chinese Yuan/year (National Bureau of Statistics of China: www.stats.gov.cn). As large numbers of the population and economic facilities are exposed to certain climate risks (Fischer et al., 2011a, 2011b; Gemmer et al., 2011) and hence highly vulnerable, knowledge on and adaptation to potential climate extremes are needed to lower the vulnerability. It is presumed that simulations and projections with the COSMO-CLM (CCLM) will potentially alter the knowledge on climate conditions and extremes.

Therefore, the objective of this study is to evaluate and analyze the simulated and projected climate parameters of the CCLM for the ZRB. We will investigate various error estimations, trends and frequencies of observed and simulated climate indicators and analyze the projected indicators for the period 2011–2050. Based on the results, current climate characteristics can be substantiated and the performance of projections with CCLM can be specified.

In Section 'Data and methodology', we describe the applied data and methodologies. Thereafter, we show the results of the validation of CCLM in Section 'Validation of CCLM'., and perform an analysis of the projections in Section 'Projection of climate extremes with CCLM (2011–2050)'.. In the final Section 'Conclusions and discussion'., we discuss and conclude all findings.

2. Data and methodology

2.1. Observational data

Daily temperature and precipitation records of 192 meteorological stations in the ZRB (Figure 1) for the period 1961–2007 are provided by the National Meteorological Information Center (NMIC) of the China Meteorological Administration (CMA). Most of the stations are located in hilly areas at an average altitude of under 700 m, except in the mountainous western part of the basin. The datasets passed the internal temporal homogeneity check of the China National Meteorological Center (CNMC), which controlled the quality by using the departure accumulating method (Buishand, 1982) and the Pettit's test (Schönwiese, 2006). Gaps in precipitation data account for less than 0.005%, and were reconstructed by the median precipitation from at least three neighboring stations.

Figure 1.

Overview map of the Zhujiang River Basin (elevation and main rivers), with the outline of the four regions (W = West-, N = North-, S = South-, and E = East-region), the meteorological observation stations, and the 0.5° grid points.

2.2. COSMO-CLM

The RCM CCLM is based on the weather prediction Lokal Modell (Steppeler et al., 2003), and adapted and extended for the use in climate projections (Böhm et al., 2006). It is a dynamical RCM based on primitive thermo-hydro-dynamical equations to describe the atmospheric circulation at resolutions between 1 and 50 km. Currently, CCLM is used in various studies (Bachner et al., 2008; Ebell et al., 2008; Hollweg et al., 2008; Rockel and Geyer, 2008; Rockel et al., 2008; Nikulin et al., 2011). CCLM uses a nesting technique to downscale a coarse resolution dataset/model (usually a GCM or reanalysis). The calculations are carried out on a rotated geographical grid with a terrain following altitudinal coordinates.

The calculation of CCLM is carried out on a 0.44° × 0.44° rotated grid covering the CORDEX-East-Asia domain (http://wcrp.ipsl.jussieu.fr/SF_RCD_CORDEX.html) with 32 atmosphere levels in vertical direction and nine soil layers. The parameterization of CCLM was optimized with a hindcast simulation of the years 1959–2000 driven by ERA40 reanalysis (Uppala et al., 2005). The first year (1959) was not considered in our analysis as it was used for the model spinoff. In the following, the optimized parameter setup is used for a GCM ECHAM5 driven run covering the same period. Here, we used the historical 20th century reconstruction run of the ECHAM5, realization 1, with greenhouse gas concentrations based on observed values. The projection starts in the year 2001, initiated with the last year of the hindcast simulation. For our analysis, we will concentrate on the period 2011–2050. We use the SRES A1B emission scenario for our projections, which assumes a rapid economic growth and a quick spread of new and efficient technologies with a balanced emphasis on all energy sources (IPCC SRES, 2000). The driving model for the projection was ECHAM5, respectively (ECHAM5_A1B run 1). A full validation of the used CCLM run including future projections for larger regions is in progress and will be published soon. For analytical purposes, CCLM is interpolated to a 0.5° × 0.5° regular geographical grid resulting in 195 grid points within the ZRB. For comparison purposes, daily data is also used from the ECHAM5_A1B run for 24 grid points based on the T63 Gaussian grid (approximately 2.8° × 2.8°).

2.3. Indicators

Six indicators are calculated from observed, simulated, and projected daily data in order to analyze and describe annual and monthly climate characteristics with particular focus on climate extremes. These are the annual and monthly averaged daily mean temperature (TMEAN), the annual and monthly maximum and minimum daily mean temperature (TMAX and TMIN). The annual and monthly total precipitation sum of wet days (≥1 mm/d, PRCPTOT), the maximum annual and monthly total precipitation sum of five consecutive days (RX5DAY), and the annual and monthly sum of dry days, i.e. days with precipitation below 1 mm (DRY DAYS).

TMEAN and PRCPTOT were chosen as they represent the two most recognized climate characteristics. These two indicators represent no real climate extremes, nevertheless, they are the most basic climatic variables, and their sufficient modelling is a precondition for further analyses. Furthermore, they are analyzed in numerous studies of the region (cf Fischer et al., 2011a; Gemmer et al., 2011) or in studies on RCM projections (cf. Bachner et al., 2008; Sillmann and Roeckner, 2008; Liu et al., 2009), and improve the comparability to other studies. TMAX and TMIN are single-day events and are depicted as the most extreme temperature events (hottest or coldest day) of each year or month. RX5DAY and DRY DAYS represent heavy precipitation or drying events which potentially lead to flood or drought events. The indicators were defined on fixed terms (Table 1) predetermined by the CMA and as recommended by the CCl/CLIVAR Expert Team for Climate Change Detection Monitoring and Indices (ETCCDMI) (Su et al., 2008; Alexander and Arblaster, 2009; Klein Tank et al., 2009; Fischer et al., 2011a, 2011b).

The comparison of the simulated and observed indicators is conducted at two spatial scales. On the one hand, we investigate the spatial relation and value for each grid point and meteorological stations within the entire basin. On the other hand, we focus on the region-averaged indicators of four specified regions. Thus, the regionalized indicators are the average of the annual and monthly indicators at all grid points or stations within these regions. The regions are subjectively chosen based on results in past research (Fischer et al., 2011a, 2011b; Gemmer et al., 2011). Each region covers 45–50 stations, i.e. approximately 25% of the available stations. According to their location (Figure 1), the four regions are named region West (W), North (N), South (S), and East (E).

2.4. Methods

2.4.1. Mean error and root mean squared error

The difference between observed and simulated indicators for both scales are estimated and analyzed with different methods. To characterize the average difference between the two datasets at a grid point or region, the mean error (BIAS) is estimated, which is defined as the positive or negative amount between the observed (o) and simulated (c) mean of the indicators (for 40 years: n = 40). Similar to the BIAS, the root mean squared error (RMSE) is estimated for each point and region to make additional statements about the magnitude of the inter-annual differences as it combines positive and negative discrepancies of the annual amounts. The BIAS is also used to identify the averaged monthly differences of each indicator in the entire basin and partially in the four regions. The equations for BIAS and RMSE are given here:

display math

For identifying similarities in the probability distribution of each annual indicator, the BIAS were tested on significant similarity using the p-values of a t-test above the 0.05 significance level. Furthermore, we visualize the sorted simulated annual data points to the observed for the entire ZRB (Schönwiese, 2006; Alexander and Arblaster, 2009).

Table 1. List of climate indicators
IDIndicator nameDefinitions
TMEAN (°C)Average temperatureAnnual and monthly mean value of daily mean temperature
TMAX (°C)Maximum TemperatureAnnual and monthly maximum value of daily mean temperature
TMIN (°C)Minimum TemperatureAnnual and monthly minimum value of daily mean temperature
PRCPTOT (mm)Total wet-day precipitationAnnual and monthly total precipitation in wet days (rainfall ≥ 1 mm/d)
RX5DAY (mm)Max 5-day precipitationAnnual and monthly maximum consecutive 5-d precipitation
DRY DAYS (d)Dry daysAnnual and monthly number of dry days (rainfall < 1 mm/d)

2.4.2. Spatial correlation and spatial variance

To quantify the differences in the spatial distribution of the observed and simulated indicators for the ZRB, the Pearson product moment correlation coefficient (PCOR) is calculated. Therefore, the unequal distributed station observations as well as the simulation results are interpolated to a regular geographical grid with 0.5° × 0.5° resolution. The PCOR is here defined as the covariance of observed (o) and simulated (c) means of each grid point divided by the product of their standard deviations (σ) (Sheskin, 2004; Wilks, 2006). The equation for PCOR is as follows:

display math

PCOR values between 0.5 and 1.0 indicate a strong spatial correlation. However, the dissimilarities in resolution of the three studied data may cause systematic differences in the spatial correlation coefficients (Corder and Foreman, 2009).

To describe the basin-wide spatial variance in the simulated and observed indicators, the principal component analysis (PCA) is applied to the annual indicators. The spatial variance gives deeper insight in the similarities or dissimilarities of the respective time series' (Bordi et al., 2004; Leung and Wu, 2005). A set of linearly independent spatial patterns (loadings) are generated, which describe the correlations with the specific principal components (PC). We concentrate here on the first four principle components (PC1–PC4), as it is assumed that these explain most of the variation from the mean, i.e. the spatial variance (Bordi et al., 2004; Leung and Wu, 2005; Schönwiese, 2006; Wilks, 2006). The assumption is subject to be proven by the results in the validation section. The higher the percentage of explained variance the stronger the station-/grid-based time series are correlated to the mean. The differences in spatial variance of simulated and observed indicators are estimated according to the findings in the PCA and the visual pattern of PC1.

2.4.3. Trend estimation

The linear regression, based on ordinary least squares, was calculated to determine the absolute decadal trend (unit/10a) of increase/decrease at each point and region. The Mann–Kendall trend test is further applied to each point and region average for the time period 1961–2000 and 2011–2050, respectively, to determine the significance of estimated trends. A comprehensive description on the application and definition of the Mann–Kendall trend test can be found in Gemmer et al. (2004), Liu et al. (2008), and Yang et al. (2010). The threshold for positive or negative trends (Kendall's Tau) is based on the 0.05 significance level. For all time series', the one-order autocorrelation coefficients have been determined to identify possible overestimations of significant trends (Hamed and Rao, 1998).

3. Validation of CCLM

In the following section, we will calculate and analyze the above described statistical values of the observed and simulated time series. The comparison considers the time period from 1961 to 2000. This is done to evaluate the capability of CCLM to simulate the climate indicators for the ZRB, representatively. CCLM was driven by the first 20th century historical reconstruction run of ECHAM5 based on observed carbon dioxide emissions. After evaluating the hindcast simulation, we will subsequently analyze the projected climate indicators for the time period 2011–2050.

3.1. Annual area-averaged BIAS and RMSE

In Figure 2, the spatial distribution of the average TMEAN, TMAX, and TMIN (1961–2000) of the observations and the simulations with CCLM and ECHAM5 are presented. In all maps, an even distribution with a Northwest–Southeast disparity in the temperature indicators can be observed. The distribution patterns of the CCLM simulations are visually more similar with the observations than the ECHAM5 simulations. This is mainly a result of the higher resolution of CCLM resulting in a more accurate representation of the orography. However, the simulated temperatures of CCLM tend to be higher than the observed ones, especially in the south-eastern parts of the basin, where the BIAS is up to 2.7 °C.

Figure 2.

Observed, CCLM-simulated, and ECHAM5-simulated averaged annual temperature indicators: (a) TMEAN, (b) TMAX, and (c) TMIN for the Zhujiang River Basin, 1961–2000.

In Figure 3, the spatial distribution of the averaged means (1961–2000) of PRCPTOT, RX5DAY, and DRY DAYS of the observations and the historical simulations with CCLM and ECHAM5 are presented. The precipitation indicators exhibit more unevenly regionalized differences. Comparing the CCLM simulations with the observations, PRCPTOT shows slightly higher positive biases for the central-northern area and the delta region of the Zhujiang River, while an underestimation is found for the central-south area. Especially, high positive biases are simulated for RX5DAY in the delta region, while strong negative biases are apparent in the western and central areas. For DRY DAYS, very high positive differences are simulated in region South, but moderate to high negative biases in the other three regions. The simulated spatial distribution of RX5DAY and PRCPTOT of CCLM, especially the difference between the mountainous western and the low lying south-eastern parts, is similar to the observations. However, the spatial distribution patterns of DRY DAYS differ significantly. This implies that on one hand CCLM is able to capture the spatial structure, the magnitude of maximum 5-d rainfall events, and also the regional patterns of the total annual distribution, but on the other hand is not able to reproduce the correct number of dry days significantly. Nevertheless, the distribution patterns of the CCLM simulations are visually much more similar with the observations than the ECHAM5 simulations. ECHAM5 simulates PRCPTOT and RX5DAY with much lower and more uniformly distributed amounts than CCLM, and also with less DRY DAYS than the observations, which is probably due to the much lower resolution of ECHAM5. Hence, CCLM produces more reliable regional distribution pattern and probabilities in precipitation indicators.

Figure 3.

Observed, CCLM-simulated, and ECHAM5-simulated averaged annual precipitation indicators: (a) PRCPTOT, (b) RX5DAY, and (c) DRY DAYS for the Zhujiang River Basin, 1961–2000.

The average annual BIAS and RMSE of the CCLM and ECHAM5 simulations of the three temperature and three precipitation indicators are listed in Tables 2 and 3. For CCLM, we find a warm BIAS for TMEAN and TMAX in each region, while for TMIN only in region South a warm BIAS is found, while a cold BIAS is found in region West and North. Similar for both, CCLM and ECHAM5, only in TMIN significant similarities occur (in ZRB, region South and East). The resulting basin-wide BIAS shows a general over-/underestimation of temperature extremes (TMAX too high and TMIN too low), potentially resulting from an overestimation of the diurnal temperature range in CCLM. Looking at the RMSE in the temperature indicators of CCLM the differences in the BIAS can be supported, as the values in RMSE are relatively close to the absolute values in BIAS. The largest RMSE in the temperature indicators are generally found in region North and East. Compared to the BIAS and RMSE of CCLM, ECHAM5 simulates TMAX closer to the observations at basin and regional scale, except for region West (Table 3).

Table 2. Average annual BIAS and RMSE of CCLM-simulated indicators in four regions of the Zhujiang River Basin (ZRB), South China, 1961–2000
IndicatorBIASRMSE
RegionZRBWNSEZRBWNSE
  1. Numbers in bold indicate a significant similarity of the observed and simulated indicators above the 0.05 significance level based on the p-value of a two-sided t-test.

TMEAN (°C)1.10.41.21.81.11.30.91.51.91.2
TMAX (°C)2.61.83.22.72.32.82.23.53.02.6
TMIN (°C)0.5−1.2−0.90.70.02.02.02.22.52.6
PRCPTOT (mm)−138−13644−50960323226353689440
RX5DAY (mm)17.7−18.31.083.921.749.429.339.2149.381.9
DRY DAYS (d)0.6−7.0−16.354.5−19.016.517.627.056.628.5
Table 3. Average annual BIAS and RMSE of ECHAM5-simulated indicators in four regions of the Zhujiang River Basin (ZRB), South China, 1961–2000
IndicatorBIASRMSE
RegionZRBWNSEZRBWNSE
  1. Numbers in bold indicate a significant similarity of the observed and simulated indicators above the 0.05 significance level based on the p-value of a two-sided t-test.

TMEAN (°C)1.31.4−0.32.30.71.51.60.92.51.0
TMAX (°C)1.02.2−0.31.8−0.51.32.60.92.11.1
TMIN (°C)0.10.2−2.92.00.12.22.13.73.42.5
PRCPTOT (mm)−415−255−447−408−563471317515490650
RX5DAY (mm)−84.8−52.2−89.7−92.2−11087.158.997.5100.5113.4
DRY DAYS (d)−23.2−17.9−22.2−21.6−27.928.222.827.428.035.4

The BIAS in PRCPTOT and DRY DAYS of CCLM are negative in all regions with the exception of small positive amounts for region North and region South, respectively. This means that less total precipitation and fewer dry days are generally simulated by CCLM. In contrary, the basin-averaged BIAS of RX5DAY is positive, with only region West exhibiting lower maximum 5-d rainfall amounts. The RMSE of the precipitation indicators are very high in region South and region East. Most obviously for CCLM are the strong BIAS and RMSE in all three precipitation indicators for region South (Table 2) which is in accordance to Figure 3. At basin scale, PRCPTOT and DRY DAYS simulated with CCLM are mainly underestimated, while RX5DAY is overestimated. This implies that CCLM generates fewer rain days which are more intense, especially in region South. The precipitation related indicators show for ECHAM5 high absolute values (Table 3). Large amounts in BIAS and RMSE indicate strong differences between the probability distribution of observed and simulated time series. According to the p-values of the t-test, the distributions of the CCLM-simulated precipitation indicators show several significant similarities to the observation (Table 2), whereas the distributions of ECHAM5 show none (Table 3). This is underlined with the visualization of basin-averaged and sorted observed and simulated annual indicators in scatter plots (Figure 4). Here, we can see very similar distributions of the simulated temperature indicators, except for TMAX, where ECHAM5 is closer to the observed distribution. An interesting visualized feature is the sharper angle of the simulated distributions (compared to the diagonal), which indicates a higher variability in annual values.

Figure 4.

Scatter plots of six basin averaged and sorted annual observed and simulated indicators (CCLM = blue; ECHAM5 = red), (a) TMEAN (°C), (b) TMAX (°C), (c) TMIN (°C), (d) PRCPTOT (mm), (e) RX5DAY (mm), and (f) DRY DAYS (days) for the Zhujiang River Basin, 1961–2000.

3.2. Monthly distribution

The monthly distributions of observed and simulated area-averaged PRCPTOT and TMEAN for the ZRB are shown in Figure 5. Additionally, the differences between CCLM simulations and observations, i.e. monthly mean differences, for the four regions are displayed in Figure 6. For the entire basin, the simulated PRCPTOT and TMEAN follow the course of the year satisfyingly. Monthly PRCPTOT is mostly underestimated, which can be found for all regions, except region West from February to May and region South from October to December. TMEAN shows a slight underestimation from April to July and a strong overestimation (by more than two degrees) from September to November. Overall, CCLM simulates monthly PRCPTOT and TMEAN more realistic than ECHAM5, which shows strong deficiencies especially in spring.

Figure 5.

Basin-averaged observed (yellow bar and solid line), CCLM-simulated (blue bar and dashed line), and ECHAM5-simulated (red bar and dotted line) monthly PRCPTOT (bars) and TMEAN (lines) for the Zhujiang River Basin, 1961–2000.

Figure 6.

Area-averaged monthly differences in (a) PRCPTOT and (b) TMEAN, i.e. the BIAS of monthly averaged observation and CCLM simulation for Region West (W), North (N), South (S), and East (E) for the Zhujiang River Basin 1961–2000.

The basin-wide averaged observed and simulated monthly extreme indicators (RX5DAY and TMAX, DRY DAYS and TMIN) show a similar course of the year and monthly differences as in PRCPTOT and TMEAN for the ZRB (Figure 7). Higher amounts in RX5DAY and DRY DAYS are simulated with CCLM for February to April but lower maximum 5-d rainfall amounts and fewer dry days are simulated for the summer months. Simulated with CCLM, TMAX exhibits stronger differences as TMEAN year-round, with a high overestimation of the summer and fall months, while TMIN shows a distinct underestimation during summer. This is also due to a larger diurnal temperature range in CCLM. Compared to ECHAM5, CCLM is closer to the observations considering the inner-annual variability of the precipitation extremes. Here it is remarkable, that CCLM underestimates RX5DAY most of the year, while it simulates a positive BIAS for the whole year. This might be explained by the incorrect simulation of a higher variability in monthly RX5DAY and the averaging of annual RX5DAY for the entire ZRB. Nevertheless, both models seem to be able to capture the monthly precipitation variability of the East Asian Monsoon. Conclusively, CCLM shows a more realistic annual course than ECHAM5, with an underestimation of the precipitation indicators and TMIN and an overestimation of TMAX all during summer. The performance of CCLM can be explained by the higher resolution capturing the regional precipitation pattern better than ECHAM5.

Figure 7.

Same as in Figure 5, but for basin-averaged monthly observed, CCLM-simulated, and ECHAM5-simulated (a) RX5DAY precipitation (bars) and TMAX (lines); (b) number of DRY DAYS (bars) and TMIN (lines) for the Zhujiang River Basin, 1961–2000.

3.3. Spatial correlation and spatial variance

On the basis of average annual means of observed and simulated grid points, the PCOR of each indicator is estimated for the entire basin. The spatial correlation coefficients for the CCLM-simulated PRCPTOT, RX5DAY, and DRY DAYS are moderately high at 0.45, 0.63, and 0.59, respectively. All CCLM-simulated temperature indicators show high positive spatial correlation coefficients with TMEAN at 0.87, TMAX at 0.65, and TMIN at 0.88. Compared with the CCLM results, ECHAM5-simulated indicators show similar spatial correlation coefficients with the observed indicators, except for DRY DAYS (Figure 3) even though the region-wide BIAS differs broadly. This similarity might be related to the fact that CCLM is dynamically downscaled from ECHAM5. However, the CCLM-simulated indicators correspond temporally and spatially well with the observations.

The PC1 of the observed TMEAN describes 70.3% of the spatial variation in the station data (Figure 8(a)), while the PC1 of the CCLM-simulated TMEAN describes 85.2% of the spatial variation in the gridded data (Figure 8(b)). For PRCPTOT, the observed PC1 describes 33.2% and the CCLM-simulated PC1 describes 45.1% of the spatial variation (Figure 8(c) and (d)). It can be seen that the loading patterns of PC1 for observed and CCLM-simulated TMEAN and PRCPTOT show similar spatial patterns (Figure 8(a)–(d)), while more uniform patterns are apparent for the CCLM-simulated loadings. The first four components (PC1–PC4) describe 87.3 and 97.5% of the spatial variation of the observed and CCLM-simulated TMEAN, respectively. The spatial variation of the observed and CCLM-simulated PRCPTOT can be described by 57.1 and 75.7% with the PC1–PC4, respectively. The higher percentages in spatial variation of CCLM imply a lower variability within the simulated time series. Based on this, the CCLM-simulated spatial variation of PRCPTOT and TMEAN of all 195 grid points are less diverse than the observed spatial variation of all 192 stations within the ZRB. Taking the similar spatial distribution into account, we conclude that the spatial variation in TMEAN and PRCPTOT are satisfyingly well represented by CCLM.

Figure 8.

Loading patterns of the PC1 of (a) observed TMEAN, (b) CCLM-simulated TMEAN, (c) observed PRCPTOT, and (d) CCLM-simulated PRCPTOT, for the Zhujiang River Basin, 1961–2000.

3.4. Trend estimation

According to the Mann–Kendall test, only a small number of time series show significant trends. The station observations show only a significant increase in TMEAN by 0.14 K/decade in region East and by 0.12 K/decade in region South (Table 4). In CCLM-simulated TMAX, a significant decrease is found in region East, while in the ECHAM5-simulated time series an increase in TMEAN and a decrease in PRCPTOT, RX5DAY, and DRY DAYS, is calculated for region West. Based on linear regression, the decadal trends and tendencies are presented in Table 4. Both simulations could not simulate the observed significant increases in regional TMEAN. CCLM and ECHAM5 simulate similar trends in TMEAN, but different values in the trends of temperature extremes (TMIN and TMAX). Nevertheless, in regard to the insignificant regression slopes CCLM produces more reliable trends than ECHAM5, if compared to the observations. It should be noted that most of the time series show no one-order autocorrelation coefficients above the 95% confidence bounds, except the CCLM-projected TMEAN and DRY DAYS time series. The existing autocorrelation can be explained by the averaging of parameters within the model structure. For both cases, these results can be neglected due to no significant trends in projected DRY DAYS, and the strong trends in projected TMEAN.

Table 4. Decadal trend statistics (unit per decade) for six observed, CCLM-simulated, and ECHAM5-simulated indicators in four regions of the Zhujiang River Basin, 1961–2000
RegionTMEANTMAXTMINPRCPTOTRX5DAYDRY DAYS
(°C/10a)(°C/10a)(°C/10a)(mm/10a)(mm/10a)(days/10a)
  1. Numbers in bold show a significant trend above the 0.05 significance level

WestOBS0.060.080.17−4.52.01.3
CCLM0.17−0.02−0.28−17.6−1.22.7
ECHAM0.220.17−0.1741.86.02.8
NorthOBS0.06−0.030.2110.41.41.0
CCLM0.00−0.26−0.11−29.82.82.9
ECHAM−0.010.03−0.23−7.5−5.61.6
SouthOBS0.120.060.213.5−0.60.7
CCLM0.01−0.21−0.26−2.88.51.5
ECHAM0.100.09−0.410.41.21.2
EastOBS0.140.090.2035.6−1.8−1.5
CCLM−0.030.37−0.02−18.414.83.1
ECHAM0.010.00−0.348.92.71.8

The RCM CCLM seems to capture the annual and monthly means, the annual course of the year, the spatial distribution patterns and variation, and the trends of all indicators satisfying and the values are in most cases closer to the observations than the respective ECHAM5 results. The temperature indicators are generally better modelled than the precipitation indicators. We conclude that CCLM can be used to project spatial distribution patterns, annual and monthly means, and temporal trends of temperature and precipitation indicators in the ZRB, for the future period of 2011–2050.

4. Projection of climate extremes with CCLM (2011–2050)

In this section, we use CCLM to project the climate for the period 2011–2050 with specific focus on the climate extreme indices based on the emission scenario SRES A1B. We will analyze the differences between the climate reconstruction (1961–2000) and the future projection (2011–2050) to assess possible future changes resulting from a higher atmospheric CO2 concentration modelled by CCLM. According to our projections, conclusions to future changes in climate extremes will be elaborated.

4.1. Spatial and temporal differences

In Figure 9, the spatial distributions of the projected differences (2011–2050 relative to 1961–2000) for the temperature and precipitation indicators are presented. All temperature indicators show an increase for the entire basin, with a moderate increase in the South and East and a larger increase in regions West and North (Figure 9(a)–(c)). Following this Southeast–Northwest disparity, TMEAN exhibits even region-averaged differences from 1.0 to 1.2 °C. TMAX has the strongest increase from 1.2 to 1.8 °C, while TMIN shows a similar distribution but with a lower increase from 0.8 to 1.4 °C leading to a higher diurnal temperature range.

Figure 9.

Difference of CCLM-simulated (1961–2000) and CCLM-projected (2011–2050) averaged annual means of (a) TMEAN, (b) TMAX, (c) TMIN, (d) PRCPTOT, (e) RX5DAY, and (f) DRY DAYS in the Zhujiang River Basin.

The projected PRCPTOT and RX5DAY show mostly strong increases in region North and South (Figure 9(d)–(e)). Here, the differences are 95–97 mm in PRCPTOT and 14–40 mm in RX5DAY. A slight decrease can be seen in the West and coastal East. Nevertheless, differences range from 16 to 80 mm in PRCPTOT, and from −1 to 10 mm in RX5DAY from West to East, respectively. Regionally opposing to PRCPTOT and RX5DAY, the projected DRY DAYS show also mostly an increase (Figure 9(f)). Here, positive differences to more dry days in the West and East range from 0.6 to 1.1 d, while it ranges from 0.2 to 0.7 d in the North and South. Hence, the projection shows an increase in total rainfall and intensity but a decrease in number of wet days, i.e. an increase in dry days.

Accordingly we can conclude that an average increase is projected by CCLM for the 50-year interval of the period 2011–2050 relative to 1961–2000 for all indicators: TMEAN (1.1 °C), TMAX (1.6 °C), TMIN (1.1 °C), PRCPTOT (73 mm), RX5DAY (15 mm), and DRY DAYS (0.7 d).

4.2. Changes in monthly distribution

In Figure 10, the projected monthly differences (2011–2050 relative to 1961–2000) in means of the temperature and precipitation indicators are presented. All projected temperature indicators show an increase in all months, with higher values in summer, fall, and winter (Figure 10(a)). CCLM projects the highest change in TMIN for winter and in TMAX for summer, leading to much warmer winter and summer. PRCPTOT and RX5DAY show the main positive differences from January to May, while DRY DAYS show negative differences in this time frame (Figure 10(b)). The increases in precipitation in the first half of the year are followed by a sudden change in June. No significant changes are observable for the second half of the year.

Figure 10.

Difference of CCLM-simulated (1961–2000) and CCLM-projected (2011–2050) basin-averaged monthly means of (a) temperature indicators and (b) precipitation indicators in the Zhujiang River Basin.

Relative to 1961–2000, the projection shows warmer and wetter conditions for the first half of the year and even stronger warming in summer and fall. The lower temperature changes in spring might be associated with the changes in precipitation.

4.3. Trend estimation

For the period 2011–2050, significant increasing trends are detected in CCLM-projected TMEAN, TMAX, and TMIN for all four regions, while none are apparent for the precipitation indicators (Table 5). The significant trends range from 0.35 to 0.48 K/decade in regional TMEAN, from 0.62 to 0.78 K/decade in TMAX, and from 0.23 to 0.52 K/decade in TMIN. Compared to almost no significant trends in the hindcast simulation, this is a remarkable feature in all temperature indicators implying a strong projected warming (Figure 11). Such significant trends are not projected for precipitation indicators, i.e. no clear statements on projected trends to more wet or dry conditions can be given (Figure 11).

Figure 11.

CCLM-simulated (1961–2000) and CCLM-projected (2010–2050) averaged annual TMEAN (red line) and PRCPTOT (green bars), including the means of 1961–2000 and 2011–2050 (dark red/dark blue lines) and the linear trend in TMEAN for 2011–2050 (black dashed line) in the Zhujiang River Basin.

Table 5. CCLM-projected decadal trends (unit per decade) for six annual indicators in four regions of the Zhujiang River Basin for the period 2011–2050
RegionTMEANTMAXTMINPRCPTOTRX5DAYDRY DAYS
(°C/10a)(°C/10a)(°C/10a)(mm/10a)(mm/10a)(days/10a)
  1. Numbers in bold show a significant trend above the 0.05 significance level.

West0.480.620.527.51.30.2
North0.450.780.3637.55.50.0
South0.350.650.36−17.6−3.9−0.5
East0.410.730.2354.210.4−1.4

5. Conclusions and discussion

In this study, we validate the RCM CCLM on its performance to simulate observed climate characteristics and extremes in the ZRB. This is done to substantiate the provided analysis of future changes in climate extremes which are projected with CCLM.

5.1. Validation of CCLM

In the first part of this study, various error estimations and differences in distributions, variability, and trends of indicators of climate extremes from observations and simulations were presented for a 40-year period (1961–2000). The results show that CCLM has certain abilities to simulate the basic characteristics of climate extremes, in both the spatial and temporal distribution, compared to the observations. The distribution patterns of the CCLM simulations are visually much more similar to the observations than the ECHAM5 simulations. CCLM seems to improve the uniform climate patterns of the coarse driving model to more detailed regional patterns, due to the higher resolution. However, the simulated temperatures of CCLM tend to be higher than the observed temperatures. Most obviously for CCLM are the strong BIAS and RMSE in all three precipitation indicators for region South. Nevertheless, CCLM can simulate monthly PRCPTOT and TMEAN more realistic than ECHAM5, which shows high deficiencies in spring. CCLM shows also a better simulation of the seasonal cycle, especially in regard to the precipitation extremes. The CCLM-simulated annual indicators correspond temporally and spatially well with the observations and simulate PRCPTOT and TMEAN with lower temporal variability but a similar spatial variation. In its spatial distribution patterns, the observed variability in TMEAN and PRCPTOT is well represented by CCLM.

The large BIAS of ECHAM5 in various indices might be partly responsible for the BIAS of CCLM, since the simulations were driven by ECHAM5, which implies that CCLM is not able to correct the anomalous input sufficiently (Roeckner et al., 2003; Wang et al., 2003; Hagemann et al., 2005; Ebell et al., 2008). One other reason can be accounted to the simulation of TMIN and TMAX in 1-h values, while the observation measurements are generally based on 6-h values. Further reasons for the overestimation in the means of temperature indicators and underestimation of the precipitation indicators might be the averaging into four regions, as much more local disparities and high variability are found. The fact that all observations are point based but the simulated grid points encompass a 0.5° × 0.5° regular grid, which was interpolated from a 0.44 × 0.44 rotated grid, might also play an important role in the differences to the observations (Bachner et al., 2008; Rockel and Geyer, 2008). The differences in their variability and trends can also be explained by the fact that CCLM is driven by a GCM (ECHAM5) rather than reanalysis (e.g. ERA40) or station-based observations. As the ZRB is strongly influenced by the East Asian Monsoon (Wang and Ding, 1997; Yu et al., 2009; Fischer et al., 2011; Gemmer et al., 2011), additionally to the above-mentioned reasons, the more distinct underestimation of precipitation indicators in summer and fall might be also related to an underrepresentation of the East Asian Summer Monsoon in the parameterization of the CCLM (and ECHAM5).

But considering these differences, we conclude that CCLM can be used to project spatial distribution patterns, monthly means, and temporal trends of temperature and precipitation indicators in the ZRB, for the period from 2011 to 2050.

5.2. Projection of climate extremes with CCLM

The projected temperature and precipitation indicators in the ZRB are analyzed for the period from 2011 to 2050 relative to the period from 1961 to 2000. Higher values of all temperature indicators are projected for the entire ZRB, with stronger increases in the west and moderate increases in the eastern part of the basin. Especially the region North and South will experience higher total precipitation, higher RX5DAY amounts, and slightly less DRY DAYS. The western part and the far eastern corner of the basin are expected to become dryer. Concerning the seasonal cycle, there is a moderate increase of temperature in spring followed by an enhanced increase in summer and fall. Additionally, a tendency to increasing and more intense precipitation is projected. In accordance to the projected changes, we can expect warmer and wetter conditions in region North and South, especially in winter and spring. This includes more intense rainfall events, which might potentially increase the risk of flooding in the central parts of the ZRB during winter and spring. Warmer and dryer conditions can be expected in region West and East, especially in summer and fall. The lower precipitation amounts but warmer temperatures will probably increase the evapotranspiration, which potentially leads to a higher risk of drought in these regions.

Currently, there exist only few studies discussing future climate extremes of the ZRB. On the basis of downscaled ECHAM5 outputs for the Pearl River basin, the findings of Liu et al. (2009) are similar to our findings in TMEAN and PRCPTOT, but at less detailed regional and temporal resolution. The findings in projected precipitation extremes by Xu et al. (2011) show opposed trends, i.e. increasing trends in annual consecutive dry days and decreasing trends in annual RX5, for the south china basin (comprising the Zhujiang and the Yangtze River basins). These trends have been calculated from three different GCM outputs (CSIRO_MK3_5, MPI_ECHAM5, and NCAR_CCSM3). However, this study covers a larger area with lower resolution, while in our study various more reliable significant trends are found for much smaller regions. According to our evaluation our model seems to simulate the climate extremes reasonably compared to the observations. It should be noted that we use only one RCM driven by one GCM under one emission scenario (SRES A1B), which does not allow exploring uncertainties in future projections of climate extremes satisfyingly (Chen et al., 2011). The validation of the CCLM-projected significant trends in temperature indicators shows that uncertainties in trend projection exist. The projected trends underlie the same order of magnitude as the related BIAS, whose uncertainty was partly investigated by using the p-values of a t-test. To further assess the uncertainties of future projections of climate extremes in the ZRB, different types of dynamical and statistical downscaling models and driving GCM's under multiple emission scenarios need to be investigated.

Regarding the findings in climate extremes, adverse consequences in various sectors, such as agriculture, water, and energy, should be anticipated. The projected more intense maximum 5-d rainfall events can lead to e.g. higher surface runoff (and eventually flooding), increased soil erosion, and diminished water quality. In contrary, the projected increase in dry days might lead to, e.g. water scarcity (i.e. drought), soil degradation (i.e. desertification), and lowering of the groundwater table. An increase in dry days can also lead to soil desiccation and soil sealing, which in turn increases the potential of flooding. Hence, an increase in dry days and in maximum 5-d rainfall might be regarded as a positive feedback towards more and stronger drought and flood events. All such impacts, directly and indirectly, affect the agricultural production and might lead to high losses in yield, which will have adverse consequences on the food security of the entire region. The strong increases in temperature extremes in the entire basin might affect most sectors as, e.g. the energy demand or plant growth pattern will change the current economic and agricultural systems. High temperatures in summer will quite severely affect the population and economic sectors, as heat-induced health issues and higher cooling demand of public, private, and industrial sectors will consequently appear. Nevertheless, higher temperatures in winter might e.g. effectively extend the plant growth period (i.e. longer annual agricultural production) and lessens the heating demand (i.e. less energy consumption). Hard and soft measures to adapt to these and other consequences have to be identified and implemented based on specifically issued policies and regulations e.g. integrated in water resource management plans and activities.

Acknowledgements

This study was supported by the National Basic Research Program of China (973 Program) (nos. 2010CB428401 and 2012CB955903), the Special Fund of Climate Change of the China Meteorological Administration (CCSF 2011-11), the National Natural Science Foundation of China (40910177), and the Sino-German Centre for Research Promotion, NSFC/DFG (GZ601). We are grateful to the ECMWF for providing the ERA-40 data. The position of Thomas Fischer at the National Climate Center is supported by the German Development Cooperation through the Center for international Migration and Development (www.cimonline.de).

Ancillary