Extreme climate events in China: IPCC-AR4 model …cas.nuist.edu.cn/TeacherFiles/file/20150603/... · Zhihong Jiang·Jie Song·Laurent Li ... Laboratoire de Météorologie Dynamique,

Climatic Change (2012) 110:385–401DOI 10.1007/s10584-011-0090-0

Extreme climate events in China: IPCC-AR4 modelevaluation and projection

Zhihong Jiang · Jie Song · Laurent Li · Weilin Chen ·Zhifu Wang · Ji Wang

Received: 25 November 2009 / Accepted: 25 February 2011 / Published online: 10 May 2011© Springer Science+Business Media B.V. 2011

Abstract Observations from 550 surface stations in China during 1961–2000 areused to evaluate the skill of seven global coupled climate models in simulatingextreme temperature and precipitation indices. It is found that the models havecertain abilities to simulate both the spatial distributions of extreme climate indicesand their trends in the observed period. The models’ abilities are higher overallfor extreme temperature indices than for extreme precipitation indices. The well-simulated temperature indices are frost days (Fd), heat wave duration index (HWDI)and annual extreme temperature range (ETR). The well-simulated precipitationindices are the fraction of annual precipitation total due to events exceeding the95th percentile (R95T) and simple daily intensity index (SDII). In a general manner,the multi-model ensemble has the best skill. For the projections of the extremetemperature indices, trends over the twenty-first century and changes at the endof the twenty-first century go into the same direction. Both frost days and annualextreme temperature range show decreasing trends, while growing season length,heat wave duration and warm nights show increasing trends. The increases areespecially manifested in the Tibetan Plateau and in Southwest China. For extremeprecipitation indices, the end of the twenty-first century is expected to have morefrequent and more intense extreme precipitation. This is particularly visible in themiddle and lower reaches of the Yangtze River, in the Southeast coastal region, inthe west part of Northwest China, and in the Tibetan Plateau. In the meanwhile,

Z. Jiang (B) · J. Song · L. Li · W. Chen · Z. Wang · J. WangKey Laboratory of Meteorological Disaster, Ministry of Education,Nanjing University of Information Science and Technology,Nanjing, 210044, Chinae-mail: [email protected]

J. SongNorthern Illinois University, DeKalb, IL, USA

L. LiLaboratoire de Météorologie Dynamique, IPSL, CNRS, UPMC, Paris, France

386 Climatic Change (2012) 110:385–401

accompanying the decrease in the maximum number of consecutive dry days inNortheast and Northwest, drought situation will reduce in these regions.

1 Introduction

Impacts of climate change are felt most strongly through changes in extreme climateevents, which are responsible for a major part of climate-related economic losses(Kunkel et al. 1999; Easterling et al. 2000; Meehl et al. 2000). China, strongly affectedby the East Asian Monsoon, is especially vulnerable to frequent weather and climatedisasters such as droughts, floods and heat waves. Meteorological disasters areestimated to cause a loss of 3 to 6% for the Chinese gross domestic product each yearsince 1990. Variations in extreme weather and climate events under the backgroundof global warming are of great concern in China and many studies can be found onthis issue (e.g. Song 2000; Zhai and Pan 2003; Zhai et al. 2005; Qian and Lin 2005).While these studies have provided observational benchmarks for evaluating climatemodels, most of them are diagnostic analyses based on the observed data.

Historically, climatologists looked to the past as a guide to the future. But withclimate change, the future will no longer resemble the past. A report by the NationalIntelligence Council (2009) assessed the national security implications of climatechange in China by the year 2030 and found that with only 7% of the world’s arableland available to feed 22% of the world’s population, China is suffering a scarcity ofnatural water resources, fast-growing urbanization and industrialization. The adverseimpacts of an increase in frequency of extreme climate events on agriculture, whichis highly water-dependent, will bring high risk for China’s food security and will addextra pressure to existing social and resource (such as energy) stresses. Although theexpected South-to-North Water Diversion Project may alleviate the water stress ofsome northern regions, whether it will be sufficient remains a question.

Global ocean–atmosphere coupled climate models are the most important toolsfor the simulation of the current climate and projection of future change underscenarios of greenhouse gas emissions. In the last 10 years, numerous studies havefocused on simulation and projection of climate changes in China. Studies presentedin Jiang et al. (2005) and Luo et al. (2005), based on climate models from the ThirdAssessment Report of the Intergovernmental Panel on Climate Change (IPCC-TAR), showed that global general circulation models (GCMs) had certain abilitiesin simulating annual and seasonal mean climate for East Asia. In a general manner,simulated surface air temperature is more realistic than precipitation. Their studiesrevealed, however, that most of the models presented a spurious precipitation centerin central and western China. Large biases are also found over the Tibetan Plateau.Based on a combination of 14 climate models and four scenarios from IPCC-TAR,Luo et al. (2005) concluded that projected increase in the mean regional surface airtemperature over China is in the range of 1.2◦C ∼ 9.2◦C and the mean regionalprecipitation change over China is in the range of −122 ∼ 298 mm by the endof the twenty-first century. These large ranges reflect not only various emissionscenarios but also important dispersions among different models, which imply a largeuncertainty in climate projections over China for the late twenty-first century.

Recent development of global climate models has brought considerable progressto the simulation of the current climate and the projection of future change. Using

Climatic Change (2012) 110:385–401 387

recent data available in the framework of the Fourth Assessment Report of IPCC(IPCC-AR4), Zhou and Yu (2006) evaluated the performance of 19 state-of-the-artclimate models in simulating surface air temperature over China and the globe. Theyreported that the models’ reproducibility of the surface air temperature, averagedover China, situated at an acceptable level; however, it was lower than that of theglobal and hemispheric averages. They also found that the simulated warming trendin the second half of the twentieth century was generally weaker than the observedtrend. Jiang et al. (2008) and Liu and Jiang (2009) analyzed the projected futureclimate changes from a sub-group of 13 state-of-the-art climate models included inthe IPCC-AR4 with B1, A1B, A2 scenarios. The CO2 concentration associated withthe B1, A1B and A2 scenarios reaches 550 ppm, 720 ppm and 850 ppm, respectively,by 2100 in the special report on emissions scenarios (SRES) (Nakicenovic et al. 2000).They found that the warming of the mean surface air temperature is in the rangeof 1.6◦C ∼ 5◦C, and the annual precipitation increase is in the range of 14 mm ∼155 mm over China by the end of the twenty-first century. Compared to the resultspresented in Luo et al. (2005) with IPCC-TAR data, climate models used in IPCC-AR4 narrowed significantly the dispersion for the projected changes of mean climatein China. This progress is mainly due to the improvement of models in IPCC-AR4.However, improvement in the simulation of mean climate does not necessarily implyimprovement in the simulation of climate extremes. Few studies have been reportedon the assessment of climate extremes in China. How good are the IPCC-AR4models in reproducing the observed climate extremes in China? What can we expectfor their future projection? These are the main questions that we want to address inthe present paper.

2 Extreme climate indices and data used

Katz and Brown (1992) already suggested that the sensitivity of extremes to changesin mean climate may be greater than one would assume from simply shifting the lo-cation of the climatological distributions. A few other studies based on observationsor simulations confirmed that changes in extreme climate events are greater than inmean climate (Kunkel et al. 2003; Klein and Konnen 2003; Groisman et al. 2005; Zhaiand Pan 2003; Zhai et al. 1999, 2005; Qian and Lin 2005; Jiang et al. 2009).

In order to identify observed and projected future changes in extreme events,a number of studies have attempted to develop a set of indices that are able toeffectively extract climate change information and are highly sensitive to globalwarming. These indices have to be weakly correlated, contain independent infor-mation and be able to measure changes in climate variability on a variety of spaceand time scales. Frich et al. (2002) developed and selected ten indicators for theabove-mentioned purposes. These indices do not go to the very extreme part of thestatistical distribution, but they are less noisy and more robust in practice. When suchindicators were applied to global land data from 1949 to 1999, it was revealed that asignificant proportion of the global land area was increasingly affected by changes inclimatic extremes during that period (Frich et al. 2002).

These indices were also considered as a high priority field in the World ClimateResearch Program (WCRP) Coupled Model Intercomparison Project#3 (CMIP3)multi-model data set (Meehl et al. 2007), which also was used as core indicators

388 Climatic Change (2012) 110:385–401

to identify and monitor extreme climate for the IPCC-AR4. Tebaldi et al. (2006)analyzed the trends in these indices based on a nine-model ensemble dataset coveringthe entire twentieth and twenty-first centuries at a global scale. They pointed outthat the future climate is characterized not only by intensified precipitation butalso by substantial geographic variability in the frequency of heavy precipitation.However, they did not validate the simulations against any observations during thelast decades where observed data did exist. Moreover, their study was at global scalewithout focusing on any particular region. Based on these indices, Alexander andArblaster (2009) evaluated the simulations of nine global coupled climate modelsover Australia and discussed their future projections. Nevertheless, it was difficult tohave a precise image of future changes of climate extremes in China from their work.

China is a region with complex topography where a strong monsoon system op-erates. Important spatial variability in both mean climate and extremes is expected.This large spatial variation has been confirmed in observational data during the last50 years in China, in particular with precipitation-related indices (Qian and Lin 2005;Wang et al. 1999; Zhai et al. 2005; Jiang et al. 2009). The aim of this study is to extendthe work of Tebaldi et al. (2006) by examining the modeling ability of IPCC-AR4models to regional variations of climate extremes in China and the future projectionsunder different greenhouse gas emission scenarios.

Among the ten indices defined by Frich et al. (2002) and used in this study, fiveof them are temperature related: frost days (Fd), intra-annual extreme temperaturerange (ETR), growing season length (GSL), heat wave duration index (HWDI) and

Table 1 The ten indicators defined by Frich et al. (2002) for monitoring changes in climate extremes

Indicator Acronym Definition Unit

Frost days Fd Total number of frost days Days(days with absolute minimumtemperature <0◦C) days

Growing season length GSL Period between when Tday > 5◦C Daysfor >5 d and Tday < 5◦C for >5 d

Intra-annual extreme ETR Difference between the highest ◦Ctemperature range temperature observation of any

given calendar year and the lowesttemperature reading of the samecalendar year

Warm night index Tn90 Percent of time Tmin > 90th percentile %of daily minimum temperature

Heat wave duration index HWDI Maximum period > 5 consecutive days Dayswith Tmax > 5◦C above the 1961–1990daily Tmax normal

Rain days R10 Number of days with precipitation Days≥10 mm/d

Consecutive dry days CDD Maximum number of consecutivedry days (Rday < 1 mm)

Maximum 5 days rainfall R5d Maximum precipitation total in 5 days mmSimple daily intensity index SDII Annual total/number of Rday ≥ 1 mm/day mm/dayHeavy precipitation fraction R95T Fraction of annual total precipitation %

events exceeding the 1961–199095th percentile

Climatic Change (2012) 110:385–401 389

Table 2 Main characteristics of the seven global ocean–atmosphere coupled general circulationmodels used in this study

Model Research center Resolution Reference

GFDL-CM2.0 USA, GFDL 144 × 90 L24 Delworth et al. (2006)GFDL-CM2.1 USA, GFDLINM-CM3.0 Russia, INM 72 × 45 L21 Volodin and Diansky (2004)IPSL-CM4 France, IPSL 96 × 72 L19 Marti et al. (2005)MIROC3.2-MEDRES Japan, CCSR T42 L20 Hasumi et al. (2004)

NIES FRCGCCNRM-CM3 France, CNRM T63 L45 Salas-Mélia et al. (2005)NCAR-PCM1 USA, NCAR T63 L45 Washington et al. (2000)

Additional information is available through http://www-pcmdi.llnl.gov/ipcc/model_documentation/ipcc_model_documentation.php

warm night index (Tn90). The other five indices are related to precipitation: simpledaily intensity index (SDII), maximum number of consecutive dry days (CDD),number of days with precipitation greater than 10 mm (R10), maximum 5-dayprecipitation total (R5d), and fraction of annual total precipitation due to eventsexceeding the 95th percentile (R95T) defined from 1961–1990. Table 1 summarizesthe definitions of the different indices. These indices incorporate both intensity andduration of the extreme events and have been considered as the common indicatorsfor studying extreme climate variations.

The ten extreme climate indices from global climate models are publicly availablein the Program for Climate Model Diagnosis and Intercomparison (PCMDI) archive.Seven of these models have provided complete datasets for the historical period aswell as for three future emission scenarios: A2 (higher), A1B (mid-range) and B1(lower). In the present study, we used all the seven available models with their basicinformation shown in Table 2. Before doing any analysis and inter-comparison, wefirst used a bilinear scheme to interpolate all data onto 2◦ × 2◦ grid cells. We alsocalculated the multi-model ensemble (MME) by using the arithmetic average of allindividual models.

Observed daily precipitation data along with daily mean, maximum and mini-mum surface air temperature data at 550 stations for the period 1961–2000 wereobtained from the Information Center, China Meteorological Administration. Theten extreme indices were calculated for each station as in Frich et al. (2002) and theninterpolated to the 2◦ × 2◦ grid cells with the Cressman (1959) interpolation scheme.

3 Models evaluation

3.1 All-China-mean indices

We will first examine the annual mean values of the ten extreme climate indicesaveraged for the whole of China during the period 1961–2000. Table 3 provides thecomparison between observed and MME simulated extreme indices averaged forthe 1961–2000 period over the region in China. In addition, the mean precipitationduring the same period is 850.1 mm from MME and 555.6 mm from observation.To make more quantitative comparisons, relative errors of each model, with respect

390 Climatic Change (2012) 110:385–401

Table 3 Annual mean values of extreme climate indices from multi-model ensemble (MME) andfrom observation (Obs.) during 1961–2000 in China

Fd ETR GSL HWDI Tn90 R10 CDD R5d SDII R95T(days) (◦C) (days) (days) (%) (days) (days) (mm) (mm/day) (%)

MME 138.8 52.3 184.1 5.4 10.0 16.0 69.7 84.2 7.9 21.6Obs. 188.0 56.6 151.0 6.9 10.4 21.9 41.8 78.3 5.8 20.1

to observations, were calculated and shown in Fig. 1 either by the color shadingor the number above each box. It is revealed that biases are of same sign for Fd(positive) and GSL (negative) for all models. This indicates that all models have coldbiases for both daily mean temperature and daily minimum temperature, which isconsistent with Jiang et al. (2009) and Zhou and Yu (2006) in evaluating monthlymean temperature in China for IPCC-AR4 models. Warm night indices (Tn90) alsopresent positive bias except for MIROC. Heat wave indices (HWDI) are mostlybiased to positive values except in NCAR and CNRM models. For precipitationindices (lower half of Fig. 1), all models overestimate R10 with biases from 6% to68%, but underestimate most other indices, especially CDD with biases from −32%to −57%, and SDII with biases from −18% to −32%. Further examination withannual-mean total precipitation also revealed an overestimation of 36% to 73%among models. The overestimation in the annual precipitation accompanied by an

Fig. 1 Relative errors of simulated extreme climate indices in China ((modeled – observed)/observed × 100%) and the mean absolute error (MAE) for the period 1961–2000 for each model.Numbers above the shaded boxes indicate the corresponding values

Climatic Change (2012) 110:385–401 391

underestimation in SDII implies that simulated precipitation events are too frequent,which is consistent with biases found in earlier models (Sun et al. 2005). Models withtoo-frequent precipitation events give lower values in CDD. The overestimations inboth R10 and precipitation frequency contribute to the overall overestimation of theannual precipitation.

From Fig. 1 showing both temperature and precipitation indices, it can be seenthat most models have better performance for the extreme temperature indicesthan for the extreme precipitation indices. It is expected that the characteristicsof the models, such as the processes in the water cycle, play an important role inthe discrepancies for simulated precipitation. For example, the bias in CDD maybe due to the models’ shortcoming with producing too many weak rainfall events.In addition, coarse spatial resolution of the GCMs may limit the models’ abilityto simulate intense precipitation events, which are much less spatially coherentthan extreme temperature events. Indices with multi-model mean absolute errors(MAE) of less than 10% are ETR, Tn90, R5d, and R95T. Comparison of the MAEsamong the seven models indicates that CNRM_CM3, MIROC3.2_MEDRES andNCAR_PCM1 are the top three models with relatively small biases, in addition toa smaller MAE in the MME as well.

3.2 Spatial variations

We now examine the models’ ability to reproduce the spatial distribution of extremeindices over China. The models are able to simulate the basic characteristics ofobserved patterns (figure not shown). To quantify the resemblance between theobservations and the models, spatial-pattern correlation coefficients were calculatedbetween simulated and observed indices. A temporal average from 1961 to 2000was performed before doing the correlation coefficient over all grid points fallinginside the Chinese territory (Fig. 2). Most of the models perform quite well for allextreme climate indices except Tn90 and CDD. The GFDL_CM2.0 model, whichshows the highest correlation coefficient for Tn90 and the second highest for ETR,R10, CDD, SDII, and R95T, has the highest averaged coefficient of all the indicesamong different models, and the MME ranks the second in the averaged coefficient.

Fig. 2 Spatial patterncorrelation coefficientsbetween simulated andobserved values for differentextreme climate indices during1961–2000. Solid lines indicatethe 95% significance level

Fd ETR GSL HWDI Tn90 R10 CDD R5d SDII R95T Mean-0.8

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

GFDL CM2.0 GFDL CM2.1 INM CM3 IPSL CM4 MIROC3.2 MEDRES CNRM CM3 NCAR PCM1 MME

Cor

rela

tion

Coe

ffici

ents

Indices

392 Climatic Change (2012) 110:385–401

Figure 3 displays the difference between simulated indices (MME) and observedones. For temperature indices, the models’ biases are distributed with almost thesame sign in most regions of China although the values vary geographically. Forfrost days (Fd), the multi-model ensemble presents an overestimation with smallererrors to the east of 110◦E and the largest errors over the Tibetan Plateau (Fig. 3a).For annual extreme temperature range (ETR), areas of large overestimation, about10◦C, appear in Southwest China and over the Tibetan Plateau, smaller errors about−5◦C to 5◦C occur in other areas (Fig. 3b). For growing season length (GSL),the underestimation in models occurs for most regions with larger biases (−60 ∼−100 days) to the west of 110◦E. Tibet is not considered for GSL. For heat waveduration index (HWDI), there are overestimations of 8–12 days in most of easternand southern China, but underestimations of 4–8 days in parts of Northwest China(Fig. 3d). For warm night index (Tn90), models present a very low spatial correlationcoefficient when correlated to the observation. The observed Tn90 field shows adistinct low-value region located over the Tibetan Plateau, to the west of 110◦E, andsmaller variations to the east of 110◦E, however, the models just show a monotonicincrease from north to south. This is believed to be related to the poor represen-tation of the Tibetan Plateau and the associated complex topography where mostbiases reach the maximum. Furthermore, these regions also have maximum biasesin terms of mean climate for surface air temperature as revealed in Zhou and Yu(2006).

Before evaluating the spatial biases of simulated extreme precipitation indices,total annual precipitation is examined because any bias in the annual precipitationsimulation is expected to have an effect on the bias of the indices. The multi-modelensemble is generally capable of simulating the spatial characteristics of rainfall inChina with drier conditions in the northwest and wetter conditions in the southeast.The spatial pattern correlation coefficient between the models and the observationsis 0.70 which is significant at the 95% confidence level. We can, however, observesome systematic errors. Simulated annual precipitation is lower by 200 to 600 mmto the south of the Yangtze River, but higher in the north (figure not shown).Such systematic errors have been observed in the earlier generation of models inIPCC-TAR (e.g. Jiang et al. 2005). The new generation of climate models stillsimulates a spurious precipitation center in the eastern Tibetan Plateau (100◦E–105◦E, 25◦N–35◦N). In addition, simulated annual precipitation is significantly higherin the southern Tibetan Plateau. This is certainly related to the coarse resolution ofmodels with an unsatisfactory representation of the sharp topographic gradient (Gaoet al. 2003, 2004).

For extreme precipitation indices, the spatial patterns of model biases varysubstantially. The common bias appears to be the reduction in spatial contrast inextreme precipitation. Models tend to underestimate R10, R5d and R95T in southernChina where heavy precipitation is frequent and intense, but overestimate them innorthern and western China where heavy precipitation events are rare (Fig. 3f, h, j).Similarly, for CDD, models tend to have an underestimation in northern and westernChina where drought occurs more frequently, but an overestimation in the southwhere drought occurs less frequently (Fig. 3g). These biases seem to resemble thesystematic error in the simulated annual mean precipitation.

Furthermore, the simulated R10 (days with rainfall rate larger than 10 mm/day)are too high in the eastern Tibetan Plateau around 110◦E–105◦E, 25◦N–35◦N. This

Climatic Change (2012) 110:385–401 393

80E 90E 100E 110E 120E 130E

20N

30N

40N

50N

106 108 110 112 114 116 118 120

4

6

8

10

12

14

16

18

20

22

unit :Day -40 ~ 0 0 ~ 40 40 ~ 80 80 ~ 120 120 ~ 160 160 ~ 200

(a) Fd

80E 90E 100E 110E 120E 130E

20N

30N

40N

50N

-15~0 0~10 10~20 20~30 30~40 40~50

106 108 110 112 114 116 118 120

4

6

8

10

12

14

16

18

20

22

Unit:day

(f) R10

80E 90E 100E 110E 120E 130E

20N

30N

40N

50N

106 108 110 112 114 116 118 120

4

6

8

10

12

14

16

18

20

22

-8 ~ -4 -4 ~ 0 0 ~ 4 4 ~ 8 8 ~ 12

Unit:Degree

(b) ETR

80E 90E 100E 110E 120E 130E

20N

30N

40N

50N

-150~-100 -100~-50 -50~0 0~20

106 108 110 112 114 116 118 120

4

6

8

10

12

14

16

18

20

22

Unit:day

(g) CDD

80E 90E 100E 110E 120E 130E

20N

30N

40N

50N

106 108 110 112 114 116 118 120

4

6

8

10

12

14

16

18

20

22

unit:Day

-120 ~ -90 -90 ~ -60 -60 ~ -30 -30 ~ 0

(c) GSL

80E 90E 100E 110E 120E 130E

20N

30N

40N

50N

-150~-100 -100~-50 -50~0 0~50 50~100

106 108 110 112 114 116 118 120

4

6

8

10

12

14

16

18

20

22

Unit:mm

(h) R5d

80E 90E 100E 110E 120E 130E

20N

30N

40N

50N

106 108 110 112 114 116 118 120

4

6

8

10

12

14

16

18

20

22

unit:day

-8 ~ -4 -4 ~ 0 0 ~ 4 4 ~ 8 8 ~ 12

(d) HWDI

80E 90E 100E 110E 120E 130E

20N

30N

40N

50N

-15~-10 -10~-5 -5~0 0~5

Unit:mm/day

106 108 110 112 114 116 118 120

4

6

8

10

12

14

16

18

20

22

(i) SDII

80E 90E 100E 110E 120E 130E

20N

30N

40N

50N

106 108 110 112 114 116 118 120

4

6

8

10

12

14

16

18

20

22

unit:%

-0.5 ~ 0 0 ~ 0.5 0.5 ~ 1 1 ~ 1.5 1.5 ~ 2

(e) Tn90

80E 90E 100E 110E 120E 130E

20N

30N

40N

50N

-15~-10 -10~-5 -5~0 0~5

Unit:%

106 108 110 112 114 116 118 120

4

6

8

10

12

14

16

18

20

22

(j) R95T

Fig. 3 Differences between simulated and observed values for different extreme temperature andprecipitation indices (simulation minus observation) for the period 1961–2000. a Fd, b ETR, c GSL,d HWDI, e Tn90, f R10, g CDD, h R5d, i SDII and j R95T

394 Climatic Change (2012) 110:385–401

problem was also found for all the earlier models, in which a false heavy precipitationregion was always simulated in the eastern Plateau (Gao et al. 2003, 2004; Jiang et al.2005). The MME can represent the main characteristics of SDII distribution with val-ues decreasing from southeast to northwest. However, systematic underestimationexists in the Southeast, the Northwest interior and the Northeast of China (Fig. 3i).For example, the simulated rainfall intensity index (SDII) is lower (higher) thanobserved value by 5 ∼ 10 mm/day in the Southeast coastal area (southern TibetanPlateau and its eastern slope around 100◦E–105◦E, about 35◦N). In Northwest, Northand Northeast China, a weaker SDII together with a higher annual precipitationtotal indicates that there are too many precipitation days in the models. In contrast,the weaker SDII may be the direct cause for a lower annual precipitation total inSoutheast and South China.

For the maximum number of consecutive dry days (CDD), which is inverselyrelated to the precipitation days, neither individual models nor MME can simulate acorrect spatial pattern. The spatial correlation coefficient with the observations is just0.25 for MME. Simulated consecutive dry days are very low in Northeast, Northwest,North and central China. For arid and semi-arid areas in Northwest China, the simu-lated CDD are only 40 ∼ 70 days, lower than the observed values by about 120 days(Fig. 3g). As discussed previously, a higher frequency in simulated precipitation dayshas led to shorter precipitation events and a low CDD. In Southeast China, simulatedCDD are very close to observed values, only about 10 days larger than observedvalues.

In summary, the models do have a good ability to simulate the spatial distributionsof the main extreme climate indices except for the warm night (Tn90) and theconsecutive dry days (CDD). For temperature indices, the largest biases are foundover the Tibetan Plateau. For precipitation indices, there is a general reduction ofspatial contrast over China.

3.3 Trends during the observed period

Linear trends of the extreme indices averaged over all of China were calculatedwith the least square method for 1961–2000. Results of both simulated and observedtrends are shown in Fig. 4.

For observed temperature indices, Fd (frost days) and ETR (annual extremetemperature range) show decreasing trends by −2.7 days and −0.6◦C per decade,respectively, but GSL (growing season length), HWDI (heat wave duration) andTn90 (warm night) show increasing trends by 1.1 days, 1.1 days, and 1.8% per decade,respectively. The observed trends, except for ETR, are statistically significant atthe 95% confidence level. Comparison between simulated and observed trendsreveals that most models and MME show consistent signs with the observation.Only MIROC3.2_MEDRES shows a significant decrease of 1 day in GSL againstan observed increase of 1.1 day. Overall, MME shows improved skill in simulatingtemperature indices trends in comparison to individual models alone.

For the observed precipitation indices, R10, R5d, SDII and R95T show positivechanges in China by 0.11 days, 0.21 mm, 0.06 mm/day and 0.30% per decade,respectively. On the other hand, the observed CDD shows a decreasing trend,which is −0.99 days per decade and the trend is significant at the 95% confidence

Climatic Change (2012) 110:385–401 395

Fig. 4 Linear trends of all-China-mean observed and simulated climate extreme indices for theperiod 1961–2000 (/decade). Shaded boxes together with the above numbers indicate the lineartrends. Values that are at 95% significant confidence level are marked in bold

level. When compared to those observation-based results, INM-CM3.0, IPSL_CM4and NCAR_PCM1 models have consistent linear trends, however, with significantdifferences in absolute values. Again, MME is revealed to provide improved perfor-mance in overall evaluation in comparison with each individual model.

In summary, the models do have certain abilities to simulate both the spatialdistributions of extreme climate indices and their trends in the observed period.The models’ abilities for the extreme temperature indices are generally higher thanthat for the extreme precipitation indices. The best-simulated extreme temperatureindices are Fd, HWDI and ETR, while the best-simulated precipitation indices areR95T and SDII. In comparison of all the models’ simulation capability, MME standsout as the best in reproducing most of the extremes for temperature and precipitationeither in regional means, spatial patterns, or temporal trends. In a general manner,the multi-model ensemble shows the best skills in simulating both temperature andprecipitation indices, when being compared to individual models.

4 Projection of the extreme climate indices for the twenty-first century

The above analyses confirm that GCMs, especially their ensemble average, MME,do have a certain skill to simulate the extreme climate statistics in China. This gives

396 Climatic Change (2012) 110:385–401

Table 4 Linear trend for the twenty-first century and changes for the end of the twenty-first century(in comparison to the period 1961–1990) for various extreme temperature indices (Table 1) averagedover China. Results are given for the three emission scenarios B1, A1B and A2 separately. Numbersin parenthesis are relative changes

Scenario Fd (day) ETR (◦C) GSL (day) HWDI (day) Tn90 (%)

Trend/100 year A2 −42.0 −1.7 28.1 90.2 44.3A1B −37.0 −1.2 23.5 74.2 37.6B1 −21.2 −0.8 12.4 35.2 25.7

2071–2100 A2 −39.8 (−21%) −1.3 (−2.2%) 36.4 (19%) 72.8 (810%) 38.0 (365%)deviation A1B −36.8 (−19%) −1.1 (−1.8%) 33.8 (18%) 59.2 (659%) 33.2 (319%)(% difference)

B1 −25.8 (−13%) −0.8 (−1.4%) 23.6 (12%) 32.6 (363%) 25.3 (243%)

us some confidence in using such models to project extreme climate indices for thetwenty-first century under different emission scenarios. For our analysis, we use theperiod of 2071–2100 relative to 1961–1990 to represent the end of the twenty-firstcentury.

Linear trends of temperature indices from MME for the period 2001 to 2100 andthe anomalous values of 2071–2100 (relative to 1961–1990) are presented in Table 4.Tn90, HWDI and GSL increase while Fd and ETR decrease. These are consistent insigns with the observed trends during the last 40 years (Fig. 4). Further observationshows that the changes in the extreme indices are generally larger for higher emissionscenarios (A2 > A1B >B 1). The largest changes are found in the heat wave durationindex (HWDI) and the warm night index (Tn90) under all the three scenarios. Forexample, under the medium emission scenario (A1B), HWDI and Tn90 increase by659% and 319%, respectively. On the other hand, Fd decreases by 19% and ETR alsohas small decreases under A1B. That means, future climate will experience a drasticincrease in heat waves and warm nights even for the medium emission scenario.These huge relative changes are related to their very small values in the present-dayclimate

All precipitation indices (Table 5) except the consecutive dry days show increasingtrends for all scenarios. The indices R10, R5d, SDII and R95T under A1B scenarioincrease by 22%, 16%, 11% and 28%, respectively, for the late twenty-first century.Higher emission scenarios give larger changes. In terms of relative changes, R95Tobserves the most important variation for most scenarios. This implies that future

Table 5 Same as in Table 4, but for the mean extreme precipitation indices (Table 1)

Scenario R10 (day) CDD (day) R5d (mm) SDII (mm/day) R95T (%)

Trend/100 year A2 3.4 1.4 18.8 0.9 6.3A1B 3.2 1.0 15.0 0.7 5.9B1 1.9 −0.4 8.0 0.4 2.9

2071–2100 A2 2.3 (22%) 1.4 (4%) 15.0 (18%) 0.7 (12%) 6.0 (31%)deviation A1B 2.3 (22%) 0.7 (2%) 12.8 (16%) 0.6 (11%) 5.4 (28%)(% difference) B1 1.8 (17%) −0.7 (−2%) 8.7 (11%) 0.4 (7%) 3.9 (21%)

Climatic Change (2012) 110:385–401 397

rainfall extremes will increase, especially in the late twenty-first century with en-hanced intensity of heavy precipitation. The consecutive dry days CDD averagedover whole of China (Table 5) have small and inconsistent variation. They decreaseunder the lower emission (B1) scenario, but increase under mid-range (A1B) andhigher (A2) emission scenarios. The small change in CDD is in fact the result ofspatial cancellation from different regions of China.

When projected trends of the extreme indices averaged over China for the twenty-first century (2001–2100) by MME (Tables 4 and 5) are compared with their observedtrends for the period 1961–2000 (Fig. 4), it is found that although the projectedtrends are consistent in signs with the observed trends, except for CDD (Fig. 4), theyclearly show enhanced strength, especially for the indices that are closely related withextreme events (HWDI, TN90, R5D, R95T), with larger trends for higher emissionscenarios. For example, HWDI may increase from 11.0 day/100-year (1961–2000)to 90.2, 74.2 and 35.2 day/100-year in the twenty-first century for the A2, A1B andB1 scenarios, respectively; R5D may increase from 2.1 mm/100-year (1961–2000) to18.8, 15.0 and 8.0 mm/100-year in the twenty-first century for the A2, A1B and B1scenarios, respectively; while CDD showed a weak decreasing trend −9.9 day/100-year in the past (1961–2000), its trends in the twenty-first century, 1.4, 1.0 and−0.4 day/100-year for the A2, A1B and B1 scenarios, respectively, are not consistentin signs. All these show that most extreme events will have increasing trends for thetwenty-first century under the global warming scenarios.

We now examine the spatial distributions of the projected changes of extremeindices. The relative differences between projected indices near the end of thetwenty-first century (2071–2100) under the SRES A1B and the simulated historicclimate (1961–1990) are shown in Fig. 5. Large values of decrease in Fd are observedin southern and central China, also in Northwest China. Because frost days in lowlatitudes are small in the current climate, projected reduction in Fd therefore showslarge values in relative terms for southern China. Although the relative changes inETR are small, most areas in China experience decreases in the annual extremetemperature range except in southern China, which implies larger increases inextreme minimum temperatures than in extreme maximum temperatures. Increasesin growing season length GSL (Fig. 5c) are observed in the whole domain with amaximum over the eastern Tibetan Plateau. For heat wave duration HWDI (Fig. 5d)and warm night Tn90 (Fig. 5e), large increases in relative values are also observed inthe whole domain, with a maximum over the western Tibetan Plateau.

For precipitation indices, a remarkable feature is the general increase, in mostregions, of R10 (Fig. 5f) representing number of days with precipitation greater than10 mm. Increases are also obtained in indices R5d (Fig. 5h) and R95T (Fig. 5j)describing heavy precipitation, and precipitation intensity SDII (Fig. 5i). Theirlargest increases mainly occur in the south Tibetan Plateau, in the west part ofNorthwest China and in the regions along the Yangtze River (about 30◦N) and theHuai River (about 35◦N). An increasing risk of flooding is thus expected for mostregions of China. In contrast to the increasing heavy precipitation, the maximumnumbers of consecutive dry days CDD (Fig. 5g) decrease in a long zone extendingfrom Northeast China to the north flank of the Tibetan Plateau. This arid and semi-arid region will have a reduced drought situation. CDD increases, however, in thewest of Northwest China and in South China, augmenting the drought risks of theseregions.

398 Climatic Change (2012) 110:385–401

(a) FD

(g) CDD(b) ETR

(h) R5d

(d) HWDI (i) SDII

(c) GSL

(f) R10

(e) Tn90 (j) R95T

Fig. 5 Spatial distribution of projected relative changes (2071–2100 relative to 1961–1990) underSRES A1B: a Fd, b ETR, c GSL, d HWDI, e Tn90, f R10, g CDD, h R5d, i SDII and j R95T. Unitsof the values are in percentage. Regions of the 95% significance level with positive and negativeprojected changes are indicated by plus and inverted triangle symbols, respectively

Climatic Change (2012) 110:385–401 399

5 Summary and conclusion

The climate modeling community in the framework of IPCC-AR4 has providedvaluable global data sets of climate evolution for the twentieth century as well asfor a range of future scenarios in the twenty-first century. We used all the sevenavailable models, which provided complete extreme climate indices. Model resultswere compared to observations from 550 stations in China for the period 1961–1990,which allowed us to evaluate the simulated present-day climate extremes.

Results show that all the seven models have certain abilities to simulate the basiccharacteristics of extreme climate indices, including both the spatial distribution andtemporal trends in the observed period. Higher scores are obtained for frost days,heat wave duration and annual extreme temperature range for the temperatureindices, and fraction of rainfall exceeding the 95th percentile and simple rainfallintensity for the precipitation indices. For temperature indices, the largest biasesare found over the Tibetan Plateau. For precipitation indices, there is a reductionof spatial contrast over China. In a general manner, the models’ simulation abilityfor extreme temperature indices is higher than for extreme precipitation indices. Themulti-model ensemble constructed from all models by simple arithmetic average isin fact extremely good in reproducing most of the extremes.

For the twenty-first century projection of extreme temperature indices, the trendsand changes at the end of the twenty-first century go into the same direction. Bothfrost days and annual extreme temperature range show decreasing trends, whilegrowing season length, heat wave duration and warm night show increasing trendsmainly in the Tibetan Plateau and Southwest China. For precipitation indices, thetwenty-first century projection shows an increase in both the extreme precipitationfrequency and intensity. This is particularly true in the middle and lower reachesof the Yangtze River, the Southeast coastal regions, the west part of NorthwestChina, and the Tibetan Plateau. In the meanwhile, accompanying the decrease in theconsecutive dry days in Northeast and Northwest, the drought situation is expectedto reduce.

Further, we need to emphasize that the validation work performed in this studyby examining different extreme climate indices in the current climate models againsttheir counterparts in observation revealed many discrepancies in the latest gener-ation of models. It is always an open issue and a challenge for the modeling com-munity to improve models performance and to correctly simulate the mean climateand extreme events. Alternative approaches, such as dynamical and/or statisticaldownscaling methods, should be explored to possibly narrow the uncertainties forthe future projection of regional climate extremes.

Furthermore, it is worth noting that the extreme indices, which are widelyregarded as useful validation test for GCMs, may be not extreme enough forimpact studies. It has been well recognized that increases in extreme climate eventsassociated with climate change have real and potentially severe costs, especiallyto China’s vulnerable economy. Current understanding, however, does not allow aclear assessment of the impact of anthropogenic climate change on China’s waterresources and agriculture (Piao et al. 2010). For example, although most of themodeling results show decreases in the projected consecutive dry days in Northeastand Northwest China for the A1B scenario, to reach a more definitive conclusion,future work must improve global and regional climate simulations—especially ofprecipitation. Most importantly, to cope with increases in extreme climate events,

400 Climatic Change (2012) 110:385–401

serious reductions in greenhouse gas emissions must be undertaken to reduce theextent of future impacts.

Acknowledgements This work is supported by the National Natural Science Foundation of China(NSFC) under the grant 40875058, the National Key Technologies R&D Program under grantNo. 2007BAC29B03, the Priority Academic Program Development of Jiangsu Higher EducationInstitutions (PAPD), and Natural Science Key Research of Jiangsu Province Higher Educationgrant 07KJA17020. We acknowledge the CMA’s Information Center for providing the observedmeteorological data at surface stations. We acknowledge the modeling groups for making theirsimulations available for analysis, the PCMDI for collecting and archiving the CMIP3 model output,and the WCRP’s Working Group on Coupled Modeling (WGCM) for organizing the model dataanalysis activity. The WCRP CMIP3 multi-model dataset is supported by the Office of Science, USDepartment of Energy.

References

Alexander LV, Arblaster JM (2009) Assessing trends in observed and modelled climate extremesover Australia in relation to future projections. Int J Climatol 29:417–435

Cressman GW (1959) An operational objective analysis system. Mon Weather Rev 87:367–374Delworth TL, Broccoli AJ, Balaji V, Beesley JA, Cooke WF, Stouffer RJ (2006) GFDL’s CM2 global

coupled climate models. Part 1: formulation and simulation characteristics. J Climate 19:643–674Easterling DR, Evans J, Groisman PY, Karl TR, Kunkel KE, Ambenje P (2000) Observed variability

and trends in extreme climate events a brief review. Bull Am Meteorol Soc 81(3):417–425Frich P, Alexander LV, Della-Marta P, Gleason B, Haylock M, Klein T, Peterson T (2002) Observed

coherent changes in climatic extremes during the second half of the twentieth century. Clim Res19:193–212

Gao X, Zhao Z, Ding Y, Huang RH, Giorgi F (2003) Climate change due to greenhouse effectsin China as simulated by a regional climate model part II: climate change. Acta Meteorol Sin61(1):29–38 (in Chinese)

Gao X, Lin W, Kucharsky F, Zhao Z (2004) Simulation of climate and short-term climate predictionin China by CCM3 driven by observed SST. Chin J Atmos Sci 28(1):78–90 (in Chinese)

Groisman PY, Knight RW, Easterling DR, Karl TR, Hegerl GC, Razuvaev VN (2005) Trends inintense precipitation in the climate record. J Climate 18:1326–1350

Hasumi H, Emori S, K-1 model developers (2004) K-1 coupled model (MIROC) description. K-1Tech. Rep. 1 Center for Climate System Research, University of Tokyo

Jiang D, Wang HJ, Lang X (2005) Evaluation of East Asian climatology as simulated by sevencoupled models. Adv Atmos Sci 22(4):479–495 (in Chinese)

Jiang ZH, Zhang X, Wang J (2008) Projection of climate change in China in the 21st century byIPCC-AR4 Models. Geogr Res 27(4):787–799 (in Chinese)

Jiang ZH, Ding Y, Cai M (2009) Monte Carlo experiments on the sensitivity of future extremerainfall to climate warming. Acta Meteorol Sin 67(2):272–279 (in Chinese)

Katz RW, Brown BG (1992) Extreme events in a changing climate: variability is more importantthan averages. Clim Change 21:289–302

Klein T, Konnen GP (2003) Trends in indices of daily temperature and precipitation extremes inEurope 1946–1999. J Climate 16:3665–3680

Kunkel KE, Pielke J, Changnon SA (1999) Temporal fluctuations in weather and climate extremesthat cause economic and human health impacts: a review. Bull Am Meteorol Soc 80:1077–1098

Kunkel KE, Easterling DR, Redmond K, Hubbard K (2003) Temporal variations of extreme precip-itation events in the United States: 1895–2000. Geophys Res Lett 30. doi:10.1029/2003GL018052

Liu M, Jiang ZH (2009) Simulation ability evaluation of surface temperature and precipitation bythirteen IPCC AR 4 coupled climate models in China during 1961–2000. J Nanjing Inst Meteorol32(2):256–268 (in Chinese)

Luo Y, Zhao Z, Xu Y, Gao X, Ding Y (2005) Projections of climate change over China for the 21stcentury. Acta Meteorol Sin 19:401–406

Marti O, Braconnot P, Bellier J, Benshila R, Bony S, Brockmann P, Cadule P, Caubel A, DenvilS, Dufresne JL, Fairhead L, Filiberti MA, Foujols MA, Fichefet T, Friedlingstein P, Gosse H,Grandpeix JY, Hourdin F, Krinner G, Lévy C, Madec G, Musat I, de Noblet N, Polcher J,

Climatic Change (2012) 110:385–401 401

Talandier C (2005) The new IPSL climate system model: IPSL-CM4. Tech Rep IPSLCM4,Institute Pierre Simon Laplace, Paris

Meehl GA, Karl TR, Easterling DR, Changnon S, Pielke R, Changnon D, Evans J, Groisman P,Knutson TR, Kunke KE, Mearns LO, Parmesan C, Pulwarty R, Root T, Sylves RT, Whetton P,Zwiers F (2000) An introduction to trends in extreme weather and climate events: observations,socioeconomic impacts, terrestrial ecological impacts, and model projections. Bull Am MeteorolSoc 81(3):413–416

Meehl GA, Covey C, Delworth T, Latif M, McAvaney B, Mitchell JB, Stouffer RJ, Taylor KF (2007)The WCRP CMIP3 multimodel dataset: a new era in climate change research. Bull Am MeteorolSoc 88:1383–1394

Nakicenovic N, Alcamo J, Davis J, de Vries B, Fenhann J, Gaffin S, Gregory K, Grübler A, JungTY, Kram T, Lebre La Rovere E, Michaelis L, Mori S, Morita T, Pepper W, Pitcher H, PriceL, Riahi K, Roehrl A, Rogner H-H, Sankovski A, Schlesinger M, Shukla P, Smith S, Swart R,van Rooijen S, Victor N, Dadi Z (2000) Special report on emissions scenarios. A special reportof working group III for the intergovernmental panel on climate change. Cambridge UniversityPress, Cambridge, p 599

National Intelligence Council (NIC) (2009) China: impact of climate change to 2030—a commis-sioned research report. Special Report, by Joint Global Change Research Institute and BattelleMemorial Institute, Pacific Northwest Division, NIC 2009-020, April 2009

Piao SL, et al (2010) The impacts of climate change on water resources and agriculture in China.Nature 467:43–51. doi:10.1038/nature09364

Qian WH, Lin X (2005) Regional trends in recent precipitation indices in China. Meteorol AtmosPhys 90(3/4):193–207

Salas-Mélia D, Chauvin F, Déqué M, Douville H, Gueremy JF, Marquet P, Planton S, Royer JF,Tyteca S (2005) Description and validation of the CNRM-CM3 global coupled model. CNRMworking note 103

Song J (2000) On the detection of climate change in the proxy records of dryness/wetness in China.Int J Climatol 20:1003–1015

Sun Y, Solomon S, Dai A, Portmann RW (2005) How often does it rain? J Climate 19:916–934Tebaldi C, Hayhoe K, Arblaster JM, Meehl GA (2006) Going to the extremes: an intercomparison

of model-simulated historical and future changes in extreme events. Clim Change 79:185–211Volodin EM, Diansky NA (2004) El-Nino reproduction in coupled general circulation model of

atmosphere and ocean. Russ Meteorol Hydrol 12:5–14Wang S, Gong D, Chen Z (1999) Serious climatic disasters of China during the past 100 years. Q J

Appl Meteorol 10:43–53 (in Chinese)Washington WM, Weatherly JW, Meehl GA, Semtner AG Jr, Bettge TW, Craig AP, Strand WG Jr,

Arblaster J, Wayland VB, James R, Zhang Y (2000) Parallel climate model (PCM) control andtransient simulations. Clim Dyn 16:755–774

Zhai PM, Pan XH (2003) Trends in temperature extremes during 1951–1999 in China. Geophys ResLett 30. doi:10.1029/2003Gl018004

Zhai PM, Sun A, Ren F, Liu X, Gao B, Zhang Q (1999) Changes of climate extremes in China. ClimChange 42:203–218

Zhai PM, Zhang XB, Pan XH (2005) Trends in total precipitation and frequency of daily precipita-tion extremes over China. J Climate 18(7):1096–1108

Zhou T, Yu R (2006) Twentieth century surface air temperature over China and the globe simulatedby coupled climate models. J Climate 19(22):5843–5858