Dr X. Yang, Institute of Meteorological Sciences, Zhejiang Meteorological Bureau, 73 West Genshan Road, Hangzhou 310017, China. E-mail: firstname.lastname@example.org
Dense meteorological station network-derived data on daily surface air temperatures over the period 1961–2007 were used to investigate the changes in the thermal growing season (GS) indicators for east China. The 394 stations are classified into six categories: metropolises, large cities, medium-sized cities, small cities, suburbs, and rural area using satellite-measured night-time light imagery and census data. Only the temperature data on 258 small cities and rural stations were used to calculate the GS indicators to reflect more ‘natural’ changes in thermal GS parameters. During the studied period, the regional mean length of the GS significantly extended by 3.05 and 2.61 d decade−1 for base temperatures of 5 and 10 °C, respectively. This extension is attributed primarily to the GS initiating at an earlier time (2.49 and 2.10 d decade−1 for base temperatures of 5 and 10 °C, respectively), rather than to the delayed end of the GS (0.55 and 0.51 d decade−1 for base temperatures of 5 and 10 °C, respectively). The mean growing degree days (GDD) has increased by 51.84 and 35.89 degree days decade−1 on average at temperatures higher than 5 and 10 °C. When the temperature data from all the 394 stations(including metropolis, large city, medium city, and suburban) were used to calculate the GS indicators, urban heat island (UHI) effects were evident, especially in highly urbanized Yangtze River Delta. The GS extension and GDD increase in metropolises increased by more than onefold over those observed for rural areas. This result indicates significant UHI effects on climatic GS changes. On the basis of the GDD changes, we find that UHI effects contributed to more than 10% in the GDD increase at temperatures higher than 10 °C. Therefore, excluding the urbanization effects from station observational data in evaluating changes in GS indices is necessary, especially for regions characterized by rapid urbanization.
In the past decade, climate change has inspired an increasing interest in changes in the growing season (GS) of vegetation. The effect of climate change on the GS of terrestrial vegetation can be analysed through phenological observations (Menzel, 2000; Chmielewski and Rotzer, 2002), remote sensing-based indicators (Tucker et al., 2001; Zhang et al., 2006), and climatic measurements (Menzel et al., 2003; Walther and Linderholm 2006; Linderholm et al., 2008). An increasing number of studies have reported shifts in timing and length of the GS (GSL) in the Northern Hemisphere (Walther et al., 2002; Root et al., 2003; Schwartz et al., 2006), particularly in Europe (Menzel and Fabian, 1999; Menzel, 2000; Chmielewski and Rötzer, 2001; Ahas et al., 2002; Menzel et al., 2006; Gordo and Sanz, 2009) and North America (Schwartz and Reiter, 2000; Richardson et al., 2006; Morin et al., 2009; Qian et al., 2012). Although phenological changes may vary with geographic location and species, abundant observations show similar trends: the earlier arrival of spring—an ecological response to climate warming (Walther et al., 2002; Parmesan and Yohe, 2003).
In recent years, several researchers have investigated the changes in GS indicators for China. Schwartz and Chen (2002) analysed Chinese phenological and climatological data from 1963 to 1993 and found that the onset of the spring phenophase showed no apparent change. However, the last spring frost began earlier and the first autumn frost occurred at a later date. Using plant phenology data from 26 stations of the Chinese Phenology Observation Network, Zheng et al. (2002) analysed the changes in plant phenophase in spring and the effect of climate warming since the 1980s on the plant phenophase in China. Since the 1980s, the phenophase in north and northeastern China, as well as in the lower reaches of the Yangtze River, has advanced, whereas that in the eastern part of southwestern China and the middle reaches of the Yangtze River has been delayed. In 16 stations in east China, statistically significant correlations were found between the changes in the spring phenophase and the temperature of one or several months before phenophase onset (Zheng et al., 2006). Along with the changes in mean monthly temperatures, the woody plants in Beijing flowered at a period earlier by an average of 5.4 d from 1990 to 2007 compared with flowering onset from 1963 to 1989 (Bai et al., 2011).
Phenological data are comparatively scarce; hence, satellite measurements are used to assess the main phenological stages of the vegetation in China. Chen and Pan (2002) analysed the relationship amongst phenological, meteorological, and normalized difference vegetation index (NDVI) data. The authors suggested that the annual mean air temperature, annual number of growing degree days (GDD), mean air temperature in late winter and spring, and GS time-integrated NDVI are the most important influencing factors for GS length. Using phenological and NDVI data, Chen et al. (2005) found that the GS from 1982 to 1993 in temperate east China significantly extended, an effect that is attributed primarily to delayed end dates and not advanced onset. Combining species-level observations, NDVI, meta-analysis, and phenological modelling, Ma and Zhou (2012) revealed that spring in China started at an average of 2.88 d earlier per decade in response to spring warming from 1980 to 2006.
Similar findings were observed in the studies on the climatological GS in China. On the basis of meteorological data, Xu and Ren (2004) determined an extended GS of 6.6 d from 1961 to 2000, during which the largest GS extension occurred in the Tibetan Plateau. Liu et al. (2006) observed a prolonged GS of 17 d over the eastern and central Tibetan Plateau from 1961 to 2003. Song et al. (2010) revealed that the mean length of the thermal GS extended at a rate of 2.3 d decade−1 in north China from 1951 to 2007, at which most of the changes are due to earlier GS onset in spring. Liu et al. (2010) reported that the national average of GS onset has shifted from 4.6 to 5.5 d earlier, whereas the average GS end has shifted from 1.8 to 3.7 d later, thereby increasing GSL by 6.9 to 8.7 d, depending on the chosen base temperature. By contrast, Jiang et al. (2011) revealed that the lengthening of the GS in Xinjiang, northwest China is attributed mainly to the delay in GS end (1.5 d decade−1) in autumn, rather than to the advance in GS onset (1.0 d decade−1) in spring.
An important issue in estimating thermal GS with station observational surface air temperature, particularly in China, is that few weather stations are located in completely or perfectly rural locations (Ren et al., 2008). Urban areas generally have a warm bias relative to surrounding rural areas. This phenomenon is called the urban heat island (UHI) effect. Most of the previous investigations on the thermal GS in China are based on the records of national reference climate stations and basic weather stations. Most of these stations are located near cities or towns; additionally, the temperature records are compromised by the UHI effects. A significant UHI effect is observed in spring and autumn (Ren et al., 2008; Yang et al., 2011), resulting in considerable bias in the calculations of thermal GS indicators. Remotely sensed data (White et al., 2002; Zhang et al., 2004; Han et al., 2008) and phenological observations (Roetzer et al., 2000) show considerable phenological differences amongst the urbanized and rural areas in east China, eastern United States, and central Europe. However, most studies on climatic GS did not consider the influence of the UHI, as evidenced by recent studies on the thermal GS changes in China. A preliminary study by Liu et al. (2010) indicated no significant difference between the temporal changes observed in two time series (with and without urban station data) because they used only the dataset from national stations. These stations are primarily located near cities and have already been affected, to some extent, by the increased UHI effect. Song et al. (2010) also indicated that the UHI effect on GS trends should not be strong, but no detailed investigation has been performed to clarify this issue.
East China is densely populated, with dramatic economic development and growth since China's reform and opening up in the late 1970s. This area has been experiencing rapid urbanization over the past three decades. With the continuous expansion in population and urban land-use area, regional UHI problems have become increasingly serious. A recent study has shown that urbanization is responsible for roughly one-fourth of the regional average warming trends observed in east China from 1981 to 2007, and for up to 44% of the temperature increase (almost 0.4 °C decade−1) in metropolises with populations of more than 1 million (Yang et al., 2011). Therefore, intensive UHI effects may account for a large proportion of bias in estimating the thermal resource in this region.
This article aims to investigate the ‘natural’ changes in the thermal GS in east China using dense meteorological station network-derived observational temperatures from 1961 to 2007. The changes in GS onset, end, and length, as well as the GDD, are analysed. In particular, the urbanization effect on assessing the trend of GS thermal resource is discussed. The rest of the article is organized as follows. Section '2. Description of the study area, data, and analysis methods' provides a detailed explanation of the study area, data, and analysis approaches; Section '3. Results' presents the results obtained; and Section '4. Conclusion and discussion' includes a brief discussion of the results and concluding remarks.
2. Description of the study area, data, and analysis methods
2.1. Study area
East China is a geographical and loosely defined cultural region that covers the eastern coastal areas of China, including the provinces of Anhui, Fujian, Jiangsu, Jiangxi, Shandong, and Zhejiang, as well as the municipality of Shanghai (Figure 1). This region comprises part of the North China Plain on the north, the Yangtze Delta Plain in the middle, and mountainous areas on the south. The urban agglomeration in the Yangtze River Delta (YRD) has the highest city density and urbanization level in China. It is composed of Shanghai and 14 prosperous cities in the Jiangsu and Zhejiang Provinces, forming a zigzag-shaped belt.
According to the climate regionalization conducted by Zheng et al. (2010), the study area was divided into three climatic zones: the northern plain (NP), lower valley of the Yangtze River (LVYR), and southern hilly region (SHR). The boundaries between the three climatic zones are 34°N and 29.5°N.
The daily mean air temperature observations over 1961–2007 from 478 weather and climate observation stations were used. These stations include the national reference climate stations, national basic weather stations, and ordinary weather stations. Missing data are inevitable in long-term monitoring at most stations. In total, the missing data account for 1.78% of the total records for daily temperature from 1961 to 2007. Calculating the GDD necessitates a complete time series. Stations with records that have missing data for more than 1 month in succession were excluded. A total of 84 stations were excluded, leaving 394 for the study. Amongst these, 179 have missing data on less than 31 d of observations, whereas 103 stations have data gaps of only 1–2 d. The missing data of the stations included in this study were completed using conventional statistical methods: (1) if missing data amount to only 1 d of observation, these are replaced by the average value of its neighbouring days; (2) if missing data account for more than two consecutive days of observations, a simple linear interpolation algorithm is used to fill the data gaps on the basis of the temperature data on the missing year from neighbouring stations that have no missing values.
He et al. (2006) showed that the Defense Meteorological Satellite Program (DMSP) Operational Linescan System (OLS) night-time light imagery can represent the actual urban development in China. To reflect the urbanization level in east China, the authors used version 4 stable night-time light products in 2007 with 1-km spatial resolution; these products were downloaded from the National Geophysical Data Center (http://www.ngdc.noaa.gov/dmsp/download.html). The DMSP/OLS image has digital number values ranged from 0 to 63.
2.3. Classification of stations
Yang et al. (2011) developed an objective and fast method for classifying urban and rural stations on the basis of DMSP/OLS night-time lights and statistical data from the administrative, unit-based urban land area for Shanghai and the other six provinces. Within a particular province/municipality, the number of grids with area of 1 km2 was accumulated according to the DMSP/OLS digital number at the grid in descending order beginning from the maximum value. Once the area (i.e. total number of grids) is approximately equal to the statistical urban land area, the DMSP/OLS digital value at the last grid accumulated is defined as the threshold of an urban area; that is, a grid with a night-time light value of no less than the threshold. The mean of DMSP/OLS values over a circular area with a 7-km radius around the station was used to identify station type. If the mean DMSP/OLS value for a station exceeds the threshold of its province/municipality, the station is classified as an urban station; otherwise, it is categorized as rural.
Urban stations that were categorized by the above-mentioned method were grouped into four categories based on the non-agricultural population data from the China City Statistical Yearbook: metropolis stations (with a non-agricultural population of more than 1.0 million), large city stations (with a non-agricultural population of 0.5–1.0 million), medium city stations (with a non-agricultural population of 0.1–0.5 million), and small city stations (with a non-agricultural population of 0.01–0.1 million). Suburban stations were classified on the basis of their proximity to a large city or metropolis station within a radius of 30-km and a continuous DMSP/OLS night-time light patch.
The 394 stations were classified into six groups (i.e. metropolis, large city, medium city, small city, suburb, and rural area) according to DMSP/OLS night-time light imagery and census data in 2007. The temperature data from 258 stations with relatively low UHI effects, including small cities and rural stations, were used to assess the ‘natural’ thermal GS changes from 1961 to 2007.
2.4. Definitions of growing season parameters
The GS varies, depending on geographical location and associated climate regime; hence, no universal definition of thermal season exists (Linderholm, 2006). The base temperatures and other parameters used for calculating GS indices differ amongst studies because most focus on regional GS changes. The temperature thresholds of thermal GS typically used in previous studies on China were 0, 5, and 10 °C (Liu et al., 2006; Zheng et al., 2006; Liu et al., 2010; Song et al., 2010; Jiang et al., 2011). Given the climate conditions in east China, we used two temperature thresholds—5 and 10 °C—to calculate the yearly GS onset and end for each station. We define thermal GS as the period from the onset of the GS (specifically, the first five consecutive days during which a daily average temperature higher than 5 and 10 °C occurred; denoted as GSS) to the end of the GS (defined as the last five consecutive during which a temperature higher than 5 and 10 °C occurred; denoted as GSE) (Zheng et al., 2006; Jiang et al., 2011). The GSL is the difference between the GSS and GSE. The annual total GDD is the accumulation of daily mean temperatures higher than the threshold base temperatures of 5 and 10 °C during the GS (Liu et al., 2010).
All trends of the GS parameters were calculated by simple linear regression, and the degree of significance was assessed using related P values. To determine spatial patterns, the trends were spatially interpolated by inverse distance weighting in ArcGIS software. A time series for each climatic region was created based on the arithmetic average of all stations in each subregion.
3.1. Average GS indices from 1971 to 2000
To elucidate the spatial distribution of the GS indices for east China, the 30-year mean value of GS onset, end, length, and degree day in 1971–2000 were calculated as reference. Figure 2 presents the GS indices (with 10 °C as base temperature), which exhibited a zonal orientation. The GSS occurred at an earlier period, whereas the GSE occurred at a later period in the southern regions than in the north. The GSL and GDD were greater in the southern part of east China. The GS started, on average, after 91 Julian days (April) in the NP, 61–90 Julian days (March) in the LVYR, and 0–60 Julian days (January–February) in the SHR (Figure 2(a)). The GS ends, on average, after 276–314 Julian days (early October–early November) in the NP, 315–326 Julian days (middle November) in the LVYR, and > 327 Julian days (after later November) in the SHR (Figure 2(b)). The GSL was shortest in the NP (150–227 d), longer in the LVYR (228–259 d), and longest in the SHR (>260 d) (Figure 2(c)). Most of the stations in east China registered a GDD higher than 10 °C for more than 2000 degree days, with the largest observed in the southeast coastal area, at more than 3500 degree days (Figure 2(d)). The distribution of GS indices for east China shows that the zonal orientation agrees with the boundaries of the three climatic zones, classified by Zheng et al. (2010).
3.2. Trends of GS indices
3.2.1. GS onset
The statistics for GS onset, end, and length in the three climatic zones in east China are shown in Table 1. The regional average GSS was earlier by 11.7 d for a base temperature of 5 °C and by 9.9 d for a base temperature of 10 °C over the 47-year study period. These values are larger than that derived by Liu et al. (2010), indicating a possible relationship to the significant warming after 2000. The largest GSS trends for both base temperatures were exhibited by the NP, followed by the LVYR and SHR (Table 1). The spatial distribution of the linear trends of the GSS for the base temperatures of 5 and 10 °C are shown in Figure 3. Out of 258 stations, 244 and 238 observed earlier spring onset in 1961–2007 for 5 and 10 °C, respectively. The most significant negative trends occurred in the NP and north LVYR. In the south LVYR and SHR, most of the stations registered non-significant negative trends and some stations (6 stations for 5 °C and 20 stations for 10 °C) showed non-significant positive trends (delayed spring).
Table 1. Linear trends in GS parameters from 1961 to 2007 in three climatic zones
The mean GSE in east China was generally delayed by an average of 2.58 d for 5 °C and by 2.38 d for 10 °C from 1961 to 2007. For 5 °C, the largest positive trend in GSE was found in the LVYR (1.18 d decade−1). The largest GSE delay occurred in the SHR (0.74 d decade−1) for 10 °C (Table 1). Figure 4 shows the spatial pattern of linear GSE trends in east China from 1961 to 2007. Most stations registered positive GSE trends. Out of 258 stations, 196 and 200 showed increasing trends for the 5 and 10 °C base temperatures, respectively but most of the stations did not pass significance tests at 0.05.
3.2.3. Length of growing season (GSL)
For 5 °C, the most significant extension of the GSL occurred in the NP, with 19.79 d on average, followed by the LVYR and SHR with increases of 18.57 and 8.27 d, respectively. For 10 °C, GSL extension over the three climatic zones was uniform, with increases of 11.56–12.83 d. The regional average GSL extended by 14.34 d for the 5 °C base temperature and 12.27 d for the 10 °C base temperature. The spatial pattern of the linear trends of the GSL in east China from 1961 to 2007 is shown in Figure 5. Most of the stations showed positive GSL trends for the two base temperatures from 1961 to 2007. Statistically significant GSL extensions were found at 108 stations for 5 °C and at 105 stations for 10 °C. Most of these stations are located in the northern part of east China. Only 6 and 15 stations showed a non-significant shortening of the GSL for 5 and 10 °C, respectively.
3.2.4. Growing degree days (GDD)
On the whole, the mean GDD of east China increased by an average of 243.67 degree days at temperatures higher than 5 °C and 168.68 degree days at temperatures exceeding 10 °C from 1961 to 2007. The largest trend in GDD increase at temperatures higher than 5 °C was observed in the LVYR, whereas that for 10 °C was observed in the SHR (Table 1). Figure 6 illustrates the spatial patterns of the linear trends of the GDD for both base temperatures. Statistically significant positive trends in the GDD were found in 214 stations for temperatures exceeding 5 °C, and in 168 stations for temperatures higher than 10 °C. Only 5 and 12 stations showed non-significant decreases in GDD at temperatures higher than 5 and 10 °C, respectively. For both base temperatures, the YRD and the southeast coastal area, with the highest urbanization level shown in Figure 1, showed the most significant increases in GDD. This phenomenon indicates that some UHI effects remain in these regions.
3.3. Urbanization effects on thermal GS changes
Table 2 presents the trends of the GS indices for the 5 and 10 °C base temperatures for each station group in east China. During the past 47 years, the largest changes in GS indices all occurred at the metropolitan stations, with early GSSs of 20.77 and 15.93 d, delayed GSEs of 5.12 and 4.56 d, and GSL lengthening of 25.90 and 20.49 d. GDD increases of 480.72 and 346.53 degree days for 5 and 10 °C, respectively, were also observed. The rural sites displayed the weakest changes, with early GSSs of 9.31 and 8.74 d, delayed GSEs of 2.07 and 3.06 d, GSL lengthening of 11.42 and 11.80 d, and GDD increases of 228.98 and 152.14 degree days for 5 and 10 °C, respectively. The GSL extension and GDD increase in the metropolitan stations increased by more than onefold over those in the rural stations, indicating significant UHI effects on the GS changes. In the sites located in large to small cities, the trends for early GSS, delayed GSE, GSL lengthening, and GDD increase showed a monotonic descent. Compared with the rural stations, the suburban stations close to large cities also exhibited larger changes in GS indices.
Table 2. Trends of GS indices for 5 and 10 °C for each station group
Statistical significance at 0.05.
Statistical significance at 0.01. Unit for GSS, GSE, and GSL is d decade−1; unit for GDD is degree day decade−1.
The linear trends of GS parameters from 1961 to 2007, derived by all the 394 stations, were calculated and interpolated to determine the spatial patterns of the UHI effects on the GS changes. Compared with the trends of the GS parameters from 258 stations with low UHI effects, those calculated by all the 394 stations have similar spatial patterns (figures not shown). Figure 7 shows the differences between the trends of GS indices for 10 °C base temperature calculated by all the 394 stations and the 258 stations with relatively low UHI effects. A striking correlation was found between the differences in GS trends (i.e. the UHI effect; Figure 7) and the urbanization level determined by DMSP/OLS night-time light imagery (Figure 1). The UHI effects in the YRD significantly delayed the GSS, extended the GSL, and increased the GDD. Therefore, the significant UHI effects in east China, especially in the YRD, affect thermal resource assessment.
Taking GDD as an example, we determined that the regional average GDD at temperatures higher than 10 °C increased by 37.03 degree days decade−1 when only the data from the stations with relatively low UHI effects were used. When the data from all the 394 stations were used, the GDD increased by 41.53 degree days decade−1 over the same period. These results indicate that the UHI effects contributed to more than 10% of the GDD increase in east China. Yang et al. (2011) found weak UHI warming even over rural stations in east China. In this study, temperature data from small cities were also used in the calculation of GDD trends. Therefore, the influence of the UHI effect on the overall increase in GDD in east China determined by this study can be regarded as conservative.
4. Conclusion and discussion
4.1. Thermal GS changes in east China from 1961 to 2007
Daily mean air temperature data from a dense observational network in east China were used to investigate the changes in the GS parameters from 1961 to 2007 in east China. As a result, this study presents more rural observational sites for determining ‘natural’ trends in GS changes than previous investigations on China. Our results show that the regional mean GSL significantly increased by 3.05 and 2.61 d decade−1 for the 5 and 10 °C base temperatures from 1961 to 2007, respectively. The GSL extension in east China is attributed to the earlier GSS (2.49 d decade−1 for 5 °C and 2.10 d decade−1 for 10 °C), rather than the delayed GSE (0.55 d decade−1 for 5 °C and 0.51 d decade−1 for 10 °C). This result is in line with the findings of Liu et al. (2010) and Song et al. (2010). Larger changes in the timing of the GSS than in that the timing of the GSE were also observed in Europe from phenological records (Menzel and Fabian, 1999; Menzel, 2000) and climate data (Walther and Linderholm 2006; Linderholm et al.2008). By contrast, Jiang et al. (2011) have recently found that the lengthening of the GS in Xinjiang Province in northwest China is attributed mainly to the delay in GSE in autumn, rather than to the advance of GSS in spring. Therefore, GS evolution in different climatic regimes is characterized by large spatial differences. Moreover, spring is more advanced in the north than in the south of east China for both base temperatures, a finding consistent with those presented by previous Chinese studies in which phenological records (Zheng et al., 2002) and climatic observations (Song et al., 2010) were used. The lengthening of the GS was most significant in the NP for 5 °C, but the largest GS extension for 10 °C occurred in the SHR. To some degree, GSL changes also depend on the chosen base temperature (Liu et al., 2010).
4.2. UHI effects on thermal GS changes
Thermal GS parameters are usually calculated using observational temperature data from weather stations. An important issue presented by the observational surface air temperature data, particularly in a developing country, such as China, is that few weather stations are located in completely or perfectly rural locations. The UHI effect may have been overlooked as a major source of error in assessing regional thermal resources. In this study, significant urbanization effects on GS changes were observed in east China, especially in the highly urbanized YRD. Amongst the non-rural station groups, the metropolitan stations exhibited the largest GSL extension and GDD increase, which are more than onefold over those exhibited by the rural stations. The trends of early GSS, delayed GSE, GSL lengthening, and GDD increase showed a monotonic descent from large-city to small-city stations. The UHI effects in the YRD significantly delay the GSS, extend the GSL, and increase the GDD. For example, the UHI effect contributed more than 10% to the total increase in GDD above 10 °C. In summary, the significant UHI effect in east China, especially in the YRD, significantly influences thermal resource assessment.
The UHI effect is only partly considered in this study; thus, our estimation of the UHI contributions to the GS changes can be regarded as conservative. We draw attention, however, to an important issue—the UHI effect—in evaluating the change in GS indices when station observational data are used. This effect is particularly important to rapidly urbanized regions, such as China. In addition, interpolating observational temperature from weather stations into a gridded dataset is extensively used in ecosystem modelling studies. Weather stations tend to be located in populated areas and, therefore, tend to overestimate interpolated temperatures across rural areas (Choi et al., 2003). In preparing gridded datasets relevant to the regional application of ecosystem models, the urbanization effect should be excluded from the interpolated temperature surfaces.
The authors are very grateful to the two anonymous reviewers for their helpful comments and constructive suggestions, which led to a significant improvement in the original manuscript. This work was supported by the National Natural Science Foundation of China (Grants 41001023 and 40921140410), the Strategic Priority Research Program ‘Climate Change: Carbon Budget and Relevant Issues’ of the Chinese Academy of Sciences (Grant XDA05090204), and the Nonprofit Technology Applied Research Project of the Zhejiang Province, China (Grants 2011C23051 and 2010C33161).