Spatiotemporal change in China's frost days and frost-free season, 1955–2000

Authors


Abstract

[1] From 1955 to 2000, China has experienced a decrease in the number of frost days, while the length of the frost-free season between the last spring freeze and the first fall frost has increased. Three distinct regimes can be detected in the time series: up to about 1973, the annual number of frost days was about 2 d higher than the 1961–1990 average; from 1973 to 1985, the annual number of frost days held close to that average with remarkably little interannual variability; and after 1985, the annual number of frost days decreased rapidly with distinct reversal around 1992. The dates of first and last frost show two patterns: before 1980, these dates fluctuated around the 1961–1990 average, but after 1980 (and especially from 1993) the frost-free season was rapidly lengthened. The numbers of frost days are highly correlated with minimum temperature (Tmin) in north China in spring and fall; while in south China frost dates correlate with minimum temperatures in winter. Generally, the seasonal relationships between Tmin and frost days are significant in both the temporal and spatial domains when seasonal average Tmin falls within a range of ±10°C. Analyzing annual and seasonal influences on the number of frost days, we find that water vapor plays a significant role. Regionally, the greater influence on the length of the frost-free season in south China has been the delayed onset of the autumn frost, while in north China the spring and autumn dates each have a comparable influence on the length of the frost-free season. The initial lengthening of the frost-free season lagged about 10 years behind the rapid increase in daily minimum temperatures, while the decrease in the annual number of frost days lagged by about 15 years.

1. Introduction

[2] Global warming in the late twentieth century is the most significant widespread climatic trend of the past 1200 years [Osborn and Briffa, 2006]. Observed trends over this period are consistent with models showing that temperatures increase with anthropogenic emissions of greenhouse gases, offset by cooling from tropospheric sulfate aerosols [Crowley, 2000; Karl and Trenberth, 2003; Stott et al., 2000; Karoly et al., 2003; Jones et al., 2003]. An important characteristic of late twentieth century warming is the differential change in daily maximum and minimum temperatures (Tmax and Tmin), most obviously from 1950 to 1980, with several factors contributing to the difference [Dai et al., 1997, 1999; Easterling et al., 1997; Liu et al., 2004b; Vose et al., 2005].

[3] For decades, most analyses of change in long-term temperature and precipitation observations have focused on changes in mean values, while variations and trends in extreme climate events have only recently received much attention [Easterling et al., 2000; Alexander et al., 2006]. Extreme events affect a wide variety of natural and human systems. Changes in their frequency or magnitude could have dramatic ecological, economic, and sociological consequences. In climate simulations, extreme temperature and precipitation events respond substantially to increased emissions of greenhouse gasses from human activities [Diffenbaugh et al., 2005]. Human-induced greenhouse forcing has recently played an important role in extreme climate events over the past half-century [Kiktev et al., 2003]. An analysis of one global data set reveals coherent spatial patterns of change in some extreme events, notably increases in warm summer nights, decreases in freezing nights, and decreases in intra-annual extreme temperature ranges [Frich et al., 2002]. Updating those findings with a new global data set, a subsequent analysis finds that over 70% of the global land area experienced an increase in warm nights and a decrease in cold nights [Alexander et al., 2006].

[4] Frost days, defined as days when the daily minimum temperature falls below 0°C, are decreasing in every country and region where the phenomenon has been studied (Plummer et al. [1999] (Australia and New Zealand), Heino et al. [1999] (northern and central Europe), Easterling [2002] (United States), Zhai and Pan [2003] (China), and Qian and Lin [2004] (China)). The occurrence of frost has a considerable impact on many human activities, including the agricultural and construction industries [Heino et al., 1999]. Most published research on this topic has focused on global- and national-scale trends over multidecade time spans. Global coupled climate model simulations highlight the importance of multiple inter-related physical processes, such as changes in soil moisture, clouds, diurnal temperature range, and sea level pressure, to the general pattern of decreasing frost days [Meehl et al., 2004]. These may have varying regional implications. Simulations with the RegCM3 model reveal that fine-scale processes are critical to local- and regional-scale vulnerability to climate change and extreme temperature response was regulated by land surface heterogeneity, including complexity of land cover and topography [Diffenbaugh et al., 2005]. We can infer from such simulations that the change in frost patterns across space and time may differ from the changes in Tmin, but further research at a finer scale is required.

[5] The frost-free season is defined as the days between the last spring frost and first fall frost. The length and timing of the season can affect various human activities such as navigation, hydroelectric power generation, forestry, and especially agriculture [JAWF, 1994]. Many crops are likely to suffer some degree of damage when ground surface temperatures fall below 0°C [Chen et al., 1993]. From an economic perspective, the length of the frost-free season is more important than the number of frost days, but it has attracted little attention from researchers. Among the exceptions, two analyses of records for Canada and the United States reveal that the frost-free season has lengthened significantly, with regional differences [Bonsal et al., 2001; Easterling, 2002], and another has documented temporal variations of the change in the frost-free season in the United States [Kunkel et al., 2003].

[6] Several researchers have examined changes in temperature extremes in China using different types of indices to define extreme events [Qian and Lin, 2004; Yan et al., 2002; Zhai et al., 1999; Zhai and Pan, 2003]. The annual number of frost days has decreased significantly across most of mainland China, as has occurred in most other part of world [Qian and Lin, 2004; Zhai and Pan, 2003]. Research to date has concentrated on the overall trends of changes in annual number of frost days. As of yet, little effort has been made to characterize the nature of these changes over space and time, especially in relation to other variables. Our objective in this study is to examine the spatial and temporal changes in the number of days of frost and the timing and length of the frost-free season, as well as the processes that could contribute to those changes, for the latter half of the twentieth century.

2. Data and Analysis Method

2.1. Data Sources

[7] Data for this study consist of daily records provided by the China Meteorological Administration (CMA) through a bilateral agreement of joint research between CMA and the U.S. Department of Energy (DOE) on global and regional climate change [Riches et al., 2000]. Measurements include daily mean temperature (Tmean), daily maximum temperature (Tmax), daily minimum temperature (Tmin), air pressure, relative humidity, total cloud cover (cc), surface wind speed (V), and precipitation (p).

[8] We took several steps to ensure the quality and consistency of the data used in this study. According to the CMA, identical standards and instrumentation were used at all 305 stations. However, because of malfunctioning instruments in the early years when most of China's weather stations were first established, there are more gaps in the time series during the period of 1951 to 1954. We excluded those years from this study and relied only on data reported from 1955 to 2000.

[9] According to Li et al. [2004b], abrupt changes in the time series of a single station are most likely caused by the relocation of the station. For each station, the time series were tested visually in different plots to exclude larger breaks in homogeneity. As a result of this inspection, we excluded one station near Shanghai after detecting an abrupt change in the time series around 1960. Furthermore, we calculated and mapped the trends of Tmin, Tmax and Tmean for each station for comparison with the trends of other stations in close geographical proximity; no obvious spatial discontinuities in the sign or magnitude of trends were noted. We recognize that there may be some undetected inhomogeneities resulting from unreported relocations of stations and, still more likely, changes in the local environments around stations, including urbanization.

[10] Missing data are inevitable for long-term monitoring at most stations. Where data were missing for up to seven consecutive days for Tmean, Tmin, and Tmax and up to 3 d for other variables examined, we used a simple linear interpolation algorithm to fill the data gaps. We used a stepwise regression to fill the gaps when the data were missing for more than seven consecutive days for Tmean, Tmin, and Tmax, and more than 3 d for other variables. The stepwise regression was performed every 5 years, with the missing station as the dependent variable and all of the other stations that had no missing values for the variable as independent variables. The stepwise regression gave a minimum coefficient of determination R2 of 0.992, 0.998, and 0.996 for Tmean, Tmin, and Tmax, respectively. The methods we used to fill the missing data do not substantially affect our results in analyzing China's climate dynamics. Missing data accounted for less than 0.38% of the total records from 1955 to 2000 for all of the variables examined. We compared the gap-filled data set with the original data set (with missing values) and found that the differences in means and trends between the data sets were not statistically significant (p = 0.05). We used the gap-filled data set in this study considering the fact that the missing data were not randomly distributed along the time series but rather had more data missing in the early years [Liu et al., 2004a, 2004b].

2.2. Data Analysis

[11] In this paper we focus on days of frost and length of frost-free-season. We define frost day as days when the daily minimum temperature is below 0°C. For this purpose we used the daily observation records from 268 stations (excluding the stations along China's southern coast that averaged less than two frost days per year). The distribution of the stations used in this analysis and the division of China into eight climatic regions, as defined by Liu et al. [2005], are shown in Figure 1. The use of these eight regions, defined by latitude and longitude, allows for comparison with previous studies of other climatic phenomena. Furthermore, they coincide roughly with China's socioeconomic macroregions [Skinner et al., 2000], facilitating the analysis of implications of climatic change on characteristic regional agricultural and economic activities.

Figure 1.

The eight climatic regions of China and the spatial distribution of the 268 weather stations used in this study.

[12] Surface specific humidity (q) was calculated based on the measured relative humidity. The estimated saturation vapor pressure was determined by air temperature. Sea level pressure (SLP) was calculated based on the elevation of the station, air temperature, standard gravity, and measured air pressure. The daily temperature range (DTR) is taken as the difference between Tmax and Tmin.

[13] We analyzed the spatial and temporal changes in the number of days below 0°C, and in the dates of the first autumn frost, last spring frost, and the length of the frost-free season, following the definitions for these phenomena introduced by Easterling [2002]. First, for each station we calculated a time series of anomalies (from the 1961–1990 base period average) of the number of frost days for each year. Regional values for the number of frost days were calculated as the unweighted arithmetic average of these values for all stations in each region. The results for China as a whole were derived from the average values of these eight climate regions, weighted by the land area of each region. We computed the trend of frost days and their statistical significance level with simple linear regression models for each of the regions and for the country as a whole. The autocorrelation is not a problem in this study because the regression was performed on an annual basis. We confirmed this by using the Durbin-Watson statistic to test the time series for first-order autocorrelation. We calculated seasonal and annual average SLP, q, Tmin, cc, V, and p for each region and for China as a whole. To analyze the processes that affect the change in the number of days of frost, we constructed statistical regression models to quantify the relationship with SLP, q, Tmin, cc, V, and p. For the purpose of analyzing temporal variation in the change in frost days and the frost-free season, we applied a nine-point binomial filter, a method to smooth out the year-to-year variations in a time series and show the longer-term trend.

[14] In order to see the spatial character of relationship between frost days and Tmin, we calculated correlations between time series of the annual numbers of frost days and the annual averaged Tmin at individual stations. To examine the relationship between frost days and Tmin in transitional areas, we estimated the 0°C line and ±10°C lines across mainland China. Here, the 0°C line is defined as any location where nighttime minimum temperature reaches 0°C during the year; similarly, the ±10°C line is defined as any location where nighttime minimum temperature reaches ±10°C during the year.

3. Results

3.1. Trends in the Annual Number of Days of Frost

[15] The annual number of days of frost declined significantly over the period of analysis in all climatic regions of China (Figure 2a). Nationally, we find that the annual number of frost days decreased at a rate of 2.9 d per decade during the 1955–2000 period with the maximum decrease of 3.3 d per decade on the Tibetan Plateau and minimum decrease of 0.5 d per decade in SE China during the same period. This represents a somewhat more rapid decline than the 2.4 d per decade rate found by Zhai and Pan [2003] for 1951–1999 based on records for 196 stations. By comparison, two previous studies of the United States reported declines of 0.8 d per decade nationally [Easterling, 2002] and 3.0 d per decade in the western United States over similar time periods [Feng and Hu, 2004], while still higher rates were found in parts of northern Europe [Heino et al., 1999].

Figure 2.

Significance of regional trends in the number of frost days in days per decade (a) annual, (b) winter, (c) spring, and (d) fall. The national average trends are given below each map.

[16] Frost days are in decline not only on an annual basis but also seasonally. The national trend shows similar rates of decline in all three seasons; regional differences are illustrated in Figures 2b, 2c, and 2d. High latitude areas have the least change in the number of days of frost during the winter, but high rates in fall and spring, while the reverse holds true for low latitude areas. Thus, fall and spring decreases predominate in north China, but winter decreases are most important in south China. Northeast China has the only exception to this pattern, with a low rate of decline in number of frost days for autumn.

[17] It has been assumed that changes in frost days are not simply linked to comparable changes in nighttime minimum temperature [Meehl et al., 2004]. Tmin increased fastest at higher latitudes, with the greatest increase in northeast China and the lowest in the southwest [Liu et al., 2004b]. Table 1 shows regional trends of annual and seasonal average Tmin, as former analyses, climate regions in the north with higher rate of increase rate in annual average of Tmin than south. From Figure 2a we can see that the spatial patterns of change in annual number of frost days differ from the spatial patterns of change of annual average Tmin. By contrast, days of frost declined most in east China, the Tibetan Plateau, and the North China Plain, with the least decline in the southeast. These differences indicate that some other factors besides average Tmin are influencing the change in the number of days of frost.

Table 1. National and Regional Trends (°C Per Decade) of Tmin in China, 1955–2000
Climatic RegionWinter (December–February)Spring (March–May)Fall (September–November)Annual
  • a

    Statistically significant at the 0.01 level.

  • b

    Statistically significant at the 0.05 level.

Northeast China (NE)0.73a0.66a0.26b0.51a
North China Plain (NCP)0.70a0.46a0.32a0.43a
East China (E)0.40a0.19a0.24a0.22a
Southeast China (SE)0.36a0.020.23b0.19a
North Central China (NC)0.69a0.32a0.35a0.41a
Southwest China (SW)0.26a0.080.16a0.15a
Northwest China (NW)0.77a0.32a0.27a0.41a
Tibetan Plateau (TP)0.56a0.32a0.34a0.36a
Nationwide0.57a0.32a0.28a0.33a

[18] Nationally, the rate of increase in Tmin in winter is 0.57°C per decade, much faster than spring and fall rates (0.32 and 0.28°C per decade, respectively). Winter rates are also faster for each of the eight climate regions taken separately. While Tmin has increased fastest in the winter, the number of frost days has shown its smallest change in that season. On the other hand, the two phenomena show similar spatial patterns in spring and fall. This suggests that the spatial relationship between Tmin and the number of frost days differ among the seasons.

[19] Simulations by Meehl et al. [2004] suggest that the relationship between Tmin and frost days will be more obvious along the margin of the 0°C line. In order to test that relationship, we calculated correlations between time series of the annual numbers of frost days and the averaged Tmin at individual stations. Figure 3 shows the results of those calculations in relation to the 0°C line. Tmin is significantly correlated with the numbers of frost days for most sites, but the higher correlation between Tmin and frost days is not restricted to the area along the 0°C line.

Figure 3.

Correlation between Tmin (during spring, fall, and winter) and the annual number of frost days at the 268 weather stations. Dashes represent the 0°C and 10°C lines.

[20] We also calculated the seasonal correlation between time series of the numbers of frost days and Tmin at individual stations (Figure 4). We find that, in spring and fall, Tmin is highly correlated with the number of frost days in northern China. In winter, Tmin is highly correlated with the number of frost days in southern China. Even though the greatest increase in Tmin is found in northeast China, the minimum temperatures in that part of China remain below 0°C, so that region sees less change in the number of frost days. Examining the area of significant correlations between frost days and Tmin, we see that it falls between the −10°C line and the +10°C line, straddling the 0°C line. Generally, the correlations are higher along the 0°C line. Our results support the findings of Meehl et al. [2004] that the relationship between Tmin and frost days is more obvious along the margin of the 0°C line if we analyze the relationship on seasonal basis.

Figure 4.

Seasonal correlation between Tmin and the number of frost days at the 268 weather stations. (a) Winter (December–February), (b) spring (March–May), and (c) fall (September–November). Dashes represent the ±10°C and 0°C lines.

[21] Another way of looking at the results is to plot scattergrams of Tmin against frost days using the 1955–2000 average values of the variables at the stations (Figure 5). The plots of fall, winter, and spring values clearly show strong relationships between Tmin and frost days (r = −0.98, −0.96, and −0.98, for fall, winter, and spring respectively, and r = −0.96 for all seasons together). We can also see from the plot that when station seasonal or annual average Tmin > 10°C, the counts of frost days are near zero, so the numbers of frost days do not change as Tmin increases. In winter, when seasonal average Tmin < −10°C, essentially all days are frost days, so the numbers of frost days do not change as Tmin decreases. If we calculate the relationship between Tmin and frost days only for sites with 1955–2000 average Tmin between −10°C and 10°C, the strength of the relationship increases (r = −0.99, −0.98, and −0.99, for fall, winter, and spring respectively, and r = −0.98 for all seasons together). We conclude that the relationship between Tmin and frost days is high in both the temporal and spatial domains when seasonal or annual average Tmin falls between −10°C and 10°C. This implies that, under such circumstances, Tmin can be used as a proxy for the number of frost days when daily data are not available. These results have implications for studies concerned with the impacts on frost days from future climate change.

Figure 5.

Annual and seasonal scatterplot of the 1955–2000 values of Tmin versus frost days: (a) annual, (b) winter, (c) spring, and (d) fall.

3.2. Temporal Variations in the Number of Days of Frost

[22] In order to analyze the temporal variation in the changing number of frost days in relation with Tmin, we first consider the temporal change in Tmin. In Figure 6, we see that Tmin begins an increasing trend around 1970, both annually and seasonally. Figure 7 gives the time series of national average values of annual and seasonal number of frost days. Here we see three distinct regimes. During the early part of the period (prior to 1973), the annual number of frost days was higher than the base period (1961–1990) average by about 2 d. From 1973 to 1985, the annual number of frost days was near the base period average and had remarkably little interannual variability. After 1985, the annual number of frost days decreased rapidly with a distinct reversal in 1992. The three seasonal time series of frost days are similar in character to the annual time series.

Figure 6.

Temporal changes in national annual and seasonal average Tmin anomaly (from the 1961–1990 mean), 1955–2000. The heavy line is the result of smoothing with a 9-point binomial filter with reflected ends.

Figure 7.

Temporal changes in the national average number of frost days anomaly (from the 1961–1990 mean), 1955–2000. The values have been smoothed with a 9-year binomial filter with reflected ends.

[23] However, in terms of temporal change, the trend in the number of frost days differs from that for Tmin: the rapid decrease in frost days lags behind the initial rapid increase in Tmin by about 15 years. This indicates that the lengthening of the growing season has lagged behind the increase in mean temperatures in China over this period. The seasonal trends in Tmin in China over this period may be relevant to this observation. As Liu et al. [2004b] reported, up to the mid 1980s, Tmin experienced a higher rate of increase in winter, with increases mainly in north China. Subsequently, Tmin began to increase more rapidly, with the higher rates in spring, autumn, and summer. Given the different regional and seasonal correlations between frost days and Tmin, especially the low correlation between the number of frost days and average Tmin in north China in winter, the differing seasonal trends of Tmin may partly explain why frost-days change lags Tmin changes.

[24] Figure 8 shows the time series of regional average values of annual and seasonal numbers of frost days. Considering annual and seasonal numbers, we classify the eight climate regions into three types.

Figure 8.

Temporal changes in regional average numbers of frost days anomaly (from the 1961–1990 mean) for the eight climatic regions of China, 1955–2000. The values have been smoothed with a 9-year binomial filter with reflected ends.

[25] • Type A (northeast, north central and northwest China), where variation in spring and autumn days of frost has the greatest effect on the annual total, while the temporal variation in winter frost days is very small.

[26] • Type B (southwest, east, and southeast China), in which variation in winter days of frost has the greatest effect on the annual total, while temporal variation in spring and autumn frost is very small.

[27] • Type C (North China Plain and the Tibetan Plateau), where temporal variation in all three seasons is evident.

[28] We can see two distinct regimes in the time series of climate regions of types A and B. During the period before 1985, the annual numbers of frost days fluctuated around the base period average. During the period of 1985 to 2000, the annual numbers of frost days decreased rapidly. There is an exception, though, in the southeast: prior to 1978, the annual number of frost days there fluctuated around about 0.5 d higher than the 1961–1990 average, then subsequently fluctuated around 0.5 d lower than that average. In the regions of type C, the time series resembles the three regimes described above for the national average.

3.3. Contributing Processes

[29] As pointed out by Meehl et al. [2004], the starting point for processes relating to changes in frost days are factors that have been shown to affect Tmin and the related phenomenon of diurnal temperature range. Therefore, we constructed statistical regression models to quantify the relationship with SLP, q, Tmin, cc, V, and p. The R2 values for simple linear regression models predicting the number of frost days with seven independent variables are shown in Table 2. For the eight climatic regions, the R2 values for the relationship between annual days of frost and Tmin vary from 0.60 to 0.85, all statistically significant at the 99% confidence level. Seasonally, there are low and nonsignificant R2 values for southeast China in the spring and fall and for northeast China in the winter. The relationships with sea level pressure, daily temperature range, total cloud cover, surface wind speed and precipitation show smaller R2 values (some as low as 0) and vary in strength and significance among regions.

Table 2. Coefficients of Determination (R2) for Simple Linear Regression Models of the Effect on Frost Days in the Eight Climate Regions and Nationwide, 1955–2000a
Climatic RegionSLPDTRTminqCcVP
  • a

    Independent variables are sea level pressure (SLP), daily temperature range (DTR), daily minimum temperature (Tmin), surface specific humidity (q), total cloud cover (cc), surface wind speed (V), and precipitation (p). Bold type indicates statistical significance at the 0.01 level.

a. Annual
Northeast China (NE)0.000.220.600.310.180.460.02
North China Plain (NCP)0.000.290.850.420.140.500.00
East China (E)0.080.230.760.230.010.380.10
Southeast China (SE)0.000.370.620.280.110.130.03
North Central China (NC)0.150.240.620.280.020.240.01
Southwest China (SW)0.050.090.760.320.040.120.01
Northwest China (NW)0.020.030.620.260.060.200.01
Tibetan Plateau (TP)0.190.480.830.600.010.120.22
Nationwide0.120.490.900.560.120.460.03
 
b. Spring (March, April, May)
Northeast China (NE)0.000.030.690.660.170.380.02
North China Plain (NCP)0.030.090.960.570.080.390.02
East China (E)0.050.030.380.090.000.100.02
Southeast China (SE)0.000.010.030.040.000.080.01
North Central China (NC)0.190.070.830.320.010.260.08
Southwest China (SW)0.140.050.500.070.010.000.00
Northwest China (NW)0.040.020.710.400.030.070.01
Tibetan Plateau (TP)0.280.120.820.430.020.110.00
Nationwide0.020.330.910.600.080.510.04
 
c. Autumn (September, October, November)
Northeast China (NE)0.050.160.360.430.010.110.22
North China Plain (NCP)0.090.030.760.200.100.300.13
East China (E)0.160.080.350.140.020.060.00
Southeast China (SE)0.000.020.060.050.020.060.01
North Central China (NC)0.200.110.680.410.000.160.02
Southwest China (SW)0.340.280.700.480.040.050.02
Northwest China (NW)0.120.010.630.500.020.200.02
Tibetan Plateau (TP)0.210.320.880.790.020.060.27
Nationwide0.280.200.800.510.030.450.00
 
d. Winter (December, January, February)
Northeast China (NE)0.040.000.010.010.060.000.00
North China Plain (NCP)0.060.120.560.600.020.080.03
East China (E)0.100.060.950.680.020.290.10
Southeast China (SE)0.000.240.630.460.120.130.07
North Central China (NC)0.020.000.200.100.000.000.00
Southwest China (SW)0.040.020.910.780.000.130.05
Northwest China (NW)0.060.020.310.270.010.040.03
Tibetan Plateau (TP)0.020.120.410.260.000.050.05
Nationwide0.110.160.740.800.010.200.10

[30] Annually, there is a more consistent relationship with surface specific humidity (q), significant in all eight climatic regions. Water vapor is the most abundant greenhouse gas in the atmosphere and it is assumed that in a warming world the atmosphere can hold more water vapor. In the latter half of the twentieth century, atmospheric moisture increases are observed over most of China, with increases in surface specific humidity of several percent per decade and larger increases in nighttime than daytime [Wang and Gaffen, 2001]. The nighttime minimum temperature is largely controlled by the greenhouse effect of water vapor in the lower atmosphere. The daytime maximum temperature depends on surface solar heating, which is strongly affected by cloud cover, and the amount of it that is released into the air as sensible and latent heat, which in turn depends on soil moisture content [Dai et al., 1999]. Although atmospheric water vapor increases both nighttime and daytime temperatures (its warming effect has little diurnal variation), the larger increases in nighttime surface specific humidity suggest the changes in water vapor plays an important role in the widespread significant decrease in the daily temperature range in China for the same period. With the increase of water vapor in the air leading to an additional increase in surface temperature, this “positive water vapor feedback” [Philipona et al., 2005] can further increase minimum temperatures.

[31] This can partly explain the higher rate of increase in Tmin, compared with Tmax, in China during the latter half of the last century. Besides water vapor, other factors such as global dimming and urbanization may also have contributed to the decrease of daily temperature range. The global dimming phenomenon of the 1950s through the 1980s is widely considered have contributed to the different rates of change between Tmin and Tmax, as dimming served to limit increases in daytime temperatures [Stanhill and Cohen, 2001].

[32] Kalnay and Cai [2003] report that the urban heat island effect can contribute to the higher rate of increase in Tmin compared with Tmax. However, on a large scale the influence of urbanization is thought to be limited [Easterling et al., 1997]; an analysis of more than 27% of global land area for the later half of the last century shows that urbanization has not systematically exaggerated the observed global warming trends in Tmin [Parker, 2004]. Controversy has persisted over the influence of urban warming on reported surface-air temperature trends in China. Zhou et al. [2004] presents evidence for a significant urbanization effect on climate based on analysis of impacts of land-use changes on surface temperature in southeast China for over several recent decades. Another analysis of the urban heat island effect on annual mean temperature finds little effect of urbanization on the observed warming in China over the past 50 years, which is similar with the result of previous estimates for other periods and locations [Li et al., 2004a]. Further research is needed on role of urbanization on climate trends in China.

[33] Given the role of water vapor in controlling surface temperatures, the higher nighttime increase in surface specific humidity can play an important role in decreasing the number of days of frost. We find that on an annual basis, for the eight climatic regions of China and nationally, the number of frost days shows a significant relationship with surface specific humidity. In spring and autumn, there is a weaker relationship in the southern regions of China. These areas have fewer days of frost outside of winter, so the influence of increasing water vapor or daily minimum temperature in spring and autumn is not significant. In winter, the relationship is weaker in northern regions, as the daily Tmin there in winter is far below zero.

[34] The number of frost days is a threshold index. These seasonal differences in the relationship with water vapor are a property of that type of index. Meehl et al. [2004] examined monthly mean changes of vertically averaged moisture and found a consistent relationship across all areas with positive moisture anomalies associated with fewer frost days. However, in simulating future climate scenarios they found a less consistent association between changes in moisture and frost days, while sea level pressure had a consistent and significant relationship with frost days for both past and future climate. They thus concluded that sea level pressure, as a manifestation of regional atmospheric circulation, is a good indicator of the dominant process affecting regional changes in frost days. In our analysis, we found that sea level pressure, as well as total cloud cover, surface wind speed, and precipitation, makes a significant contribution to the variation of the number of frost days in some regions on an annual or seasonal basis. We also examined the numbers of frost days in relation to these and other variables on a monthly basis (not shown) and arrived at results similar to those of Meehl et al. [2004]: both sea level pressure and surface specific humidity show significant relationships with frost days nationally and for each of the eight climatic regions. Nevertheless, on an annual or seasonal basis, water vapor has the most important and consistent effect on the number of days of frost.

3.4. Spatial Trends of Change in the Frost-Free Season

[35] For agriculture, the length of the frost-free season is more important than the annual number of frost days. Warm-weather crops usually must be harvested before the first fall frost occurs. If the first fall frost in a particular year should come earlier than usual, these crops would be killed before they ripen. This could result in sharp decreases of crop yield. Even surviving crops could be more difficult to store under these circumstances because of their high moisture content, which could also lead to a lower sprouting rate for seed crops. The earlier the first fall frost, the greater the damage to such crops. Similarly, if the last spring frost comes significantly later than usual, damage can be done to crops seeded in spring and to wintering crops that are caught in the stage of ear differentiation. Obviously, the later the last spring frost, the greater likelihood of crop damages [Chen et al., 1993]. All else being equal, the lengthening of the frost-free season over the past half century may have reduced the impact of frost damage on agriculture, but further research into the relationship between frost dates and season length will be important.

[36] We examined changes in the dates of the first autumn and last spring freezes and the resulting length of the frost-free season (from the spring date to the fall date) following the method described by Easterling [2002]. Taking the averages for China as a whole, we find a significant change in all three of these indicators, much greater than that reported by Easterling for the United States over a similar period. Figure 9 illustrates the trends in these indicators for each region and for China as a whole in terms of days per decade.

Figure 9.

Trends in the average (a) date of the first fall frost, (b) date of the last spring frost, and (c) length of the frost-free season. Trends are in days per decade. The national average trends are given below each map.

[37] The spatial patterns of change in the date of the last spring freeze runs counter to the change in the number of spring frost days discussed above, with the trend toward earlier thawing accelerating faster in the lower, not higher, latitude regions. A possible explanation is that, having fewer days of frost, the southern regions see greater effects on the last date of frost from smaller changes in temperatures.

[38] The spatial patterns of change in the date of the first autumn frost are consistent with those for the number of fall days of frost in six of the eight climatic regions. East China is the only climatic region to show a trend toward earlier first autumn frosts, but the trend is not statistically significant. Overall, the trend toward a longer frost-free season is consistent with the results presented earlier and is statistically significant for seven of the eight climatic regions of China (again excluding only east China). Our results show that the changes in the length of the frost-free season are not fully explained by the changes in the number of frost days. As we have explained above, the number of frost days is highly correlated with averaged Tmin when average Tmin between −10°C and 10°C. However, changes in the length of the frost-free season show different relationships in different regions. As shown in Table 3, the two phenomena are uncorrelated in east China. In that climatic region, which has seen a significant rapid decrease of the annual number of frost days, there has been no significant increase in the length of the frost-free season. In China, growing season length decreases in all regions with the increase of the number of frost days, but the correlations between annual numbers of frost days and the growing season length are low in some climate regions.

Table 3. Correlation Coefficient (r) in Eight Climate Regionsa
 Case 1Case 2Case 3
  • a

    Case 1: Annual number of frost days versus length of the frost-free season; Case 2: Spring number of frost days versus date of last spring frost; Case 3: Fall number of frost days versus date of first fall frost. Asterisks (*) indicate significance at the 0.01 level.

Northeast China (NE)−0.62*0.74*−0.52*
North China Plain (NCP)−0.71*0.78*−0.53*
East China (E)0.020.72*−0.29
Southeast China (SE)−0.330.19−0.32
North Central China (NC)−0.63*0.61*−0.61*
Southwest China (SW)−0.160.57*−0.13
Northwest China (NW)−0.59*0.65*−0.63*
Tibetan Plateau (TP)−0.68*0.64*−0.52*

[39] According to our results, the changes in the number of frost-days and the changes in the length of frost-free-season have been behaving differently in China. Frost-free season is defined as the number of continuous days between the last frost day in the spring and the first frost day in the fall, while the frost days are calculated as the total number of days with daily minimum temperature (Tmin) below 0°C. With global warming, increasing Tmin, the number of frost days has decreased as we have seen for all the regions in China though the decreasing magnitude is different among the regions. Meanwhile, changes in the annual (seasonal) number of frost days may not necessarily mean the changes in frozen-free length (the last spring frost date or first fall frost date). For example, under global warming the last date of frost in spring may not change, while the number of days with Tmin < 0°C before the last frost date may have significantly decreased. As a result, the correlation between frost days and frost-free season is spatially variable as we have seen for China where the correlation is high for most regions except the East, Southeast and Southwest regions (Table 3). This is not unique to China, similar results have been reported for other parts of the world. As Meehl et al. [2004] pointed out that the relationship between frost days and frost-free season is complicated because many factors, in addition to the number of frost days, may also affect the growing season length.

[40] However, we should be cautious in interpreting these results, especially for southeast China, as frost days there are concentrated in the winter: with relatively few days of frost in fall and spring, there is a low correlation between the numbers of frost days in those seasons and the last or first dates of frost. Shifts in the onset of spring warming and fall cooling can also influence the length of the frost-free season. To investigate these relationships, we also plotted scattergrams giving numbers of frost days against the frost-free season length (Figure 10). The plots clearly show strong relationships between the number of frost days and the length of the frost-free season (r = −0.94), considering data from all 268 sites. The number of frost days and the length of the frost-free season are highly correlated in the spatial domain, but this does not always hold true in the temporal domain.

Figure 10.

Scatterplot of the 1955–2000 annual number of frost days and frost-free season length.

3.5. Temporal Variation in the Frost-Free Season

[41] In order to analyze change in the timing and length of the frost-free season over time, we examined the time series of the date of the last spring frost, the date of the first autumn frost, and the length of the intervening frost-free season. Figure 11 shows the time series for China as a whole. We observe two distinct regimes. Before 1980, the three variables fluctuate around the base period average. From 1980 to 2000, the first autumn frost arrived later and the last spring frost came sooner, producing a rapidly lengthening frost-free season. A sharp change is evident around 1993. This increase in the length of the frost-free season lags about 10 years behind the rapid increase in Tmin.

Figure 11.

Temporal changes in the national average date of the first fall frost and last spring frost, and the length of the frost-free season, 1955–2000. Values shown are differences from the 1961–1990 mean, smoothed with a 9-point binomial filter with reflected ends.

[42] Figure 12 portrays the time series of regional average values of the same three variables. We can see the year-to-year variability is smaller in the north than in the south, excepting the Tibetan Plateau. The north central, southwest, northwest, southeast, and Tibetan Plateau regions show distinct changes in the dates of the last spring frost and first autumn frost, and in the length of the intervening frost-free season. Seven of the climatic regions (all except the east China region) display significant increases in the length of the frost-free season, but the upward trends for different regions begin in different years.

Figure 12.

Temporal changes in the regional average dates of the first fall frost and last spring frost, and the length of the frost-free season, 1955–2000. Values shown are differences from the 1961–1990 mean, smoothed with a 9-point binomial filter with reflected ends.

[43] Either the fall or spring dates, or both, can influence the length of the frost-free season. We performed a regression analysis to measure their relative influence. Table 4 shows the R2 values by region. In the southern regions, the autumn date has the higher R2 values, while for two regions the spring date does not have a statistically significant relationship at all; thus, the autumn date appears to have the greater influence on the length of the frost-free season. In the northern regions except for the northeast, the spring date has the higher R2 values, though the difference in influence between the spring and autumn dates is small. For its part, northeast China has a marginally higher R2 value for the autumn date, but here too it appears that the spring and autumn dates each have a comparable influence on the length of the frost-free season.

Table 4. Coefficients of Determination (R2) for Simple Linear Regression Models Predicting the Length of the Frost-Free Season, 1955–2000a
 Independent Variables
Date of Last spring FrostDate of First Autumn Frost
  • a

    Asterisks (*) indicate statistical significance at the 0.01 level.

Nationwide0.51*0.79*
Northeast China (NE)0.48*0.51*
North China Plain (NCP)0.76*0.64*
East China (E)0.030.74*
Southeast China (SE)0.25*0.94*
North Central China (NC)0.57*0.56*
Southwest China (SW)0.120.88*
Northwest China (NW)0.62*0.48*
Tibetan Plateau (TP)0.59*0.76*

4. Summary and Conclusions

[44] In this paper, we have concentrated on the spatial and temporal character of the change in frost in China over the second part of the last century, a period where climate change has been dominated by the effects of anthropogenic global warming. Nationally, the annual number of frost days has decreased at the rate of 2.9 d per decade over the period of 1955–2000. The spatial pattern of change in frost differs from the pattern observed for Tmin, with a faster rate of decrease in frost in east China, the Tibetan Plateau, and the North China plain, and a slower rate of decrease in the southeast. As a threshold index, frost is more sensitive to change in Tmin along the margin where the 0°C line is moving poleward, as suggested by Meehl et al. [2004]. Similar to findings in Europe, in China the high latitude areas generally saw little change in the number of winter frost days, but higher rates of decrease in spring and/or autumn; low latitude areas saw the reverse of that pattern. Our results show that the relationship between Tmin and frost days is strong in both the temporal and spatial domains when seasonal average Tmin falls between −10°C and 10°C.

[45] We classified the time series of national average numbers of annual and seasonal days of frost into three distinct regimes: oscillating higher than the base period average before 1973, stable near the average from 1973 to 1985, and decreasing rapidly after 1985. Comparing annual and seasonal numbers of frost days, we divided the eight climatic regions into three types: (A) three northern regions with change concentrated in the spring and autumn; (B) three southern regions with change concentrated in the winter, and (C) two remaining regions (the North China Plain and the Tibetan Plateau) where change occurred in all three seasons. In most of the regions of types A and B, the annual number of frost days fluctuated around the base period average up until 1985, then decreased rapidly; however, the southeastern region saw a different temporal pattern, fluctuating around 0.5 d higher than the base period average before 1978 and around 0.5 d lower from that year forward. The regions of type C followed the national pattern of three temporal regimes.

[46] We have documented significant changes in the length of the frost-free season, with first autumn frosts coming later and last spring freezes coming earlier, averaged across the whole country. Consistent with earlier reports, we report a lengthening frost-free season in all eight of China's climatic regions, statistically significant in all but the east China region. Examining spatiotemporal patterns, we find greater decreases in the number of spring days of frost at high latitudes, but greater changes in the date of the last spring frost at low latitudes. The patterns of change in the date of first autumn frost are consistent with the changes in the number of autumn frost days except in the east and southeast climatic regions. Temporally, the length of the frost-free season began to increase rapidly with a later first autumn frost and earlier last spring frost beginning in 1980, lagging about a decade behind the initial rapid increase in Tmin. The decline in the annual number of days of frost lags the Tmin increase by about 15 years.

[47] This analysis confirms the reduction in annual days of frost and the lengthening of the frost-free season across China, as has also been reported in every other country and region where the phenomenon has been studied. As the occurrence of frost has a considerable impact on many natural processes and human activities, especially agriculture, these trends have important consequences for environmental policy and management.

Acknowledgments

[48] This research was partially supported by the Natural Science Foundation of China (30671678). The Chinese Academy of Sciences and Rutgers University provided computational facilities for data analysis.