1. Introduction
Ground surface temperature (GST) is a key land surface parameter. Through interaction with the atmosphere, it plays a key role in energy balance of the land surface and has significant impacts on climate variability (Liu and Avissar 1999; Hu and Feng 2004; Zhang et al. 2005; Fan 2009; Wang et al. 2013). Since 1960, there have been approximately 800 weather stations in China for routine measurements of GST. This makes it possible to understand temporal and spatial variation of GST over China and its relationship with climate variability during this period (Zhou and Huang 2006; Yang and Zhang 2016; Wang et al. 2016; Wang et al. 2017). Based on observations of GST, Wang et al. (2016) found that the GST in China featured an abrupt change around 2000, with a descending (an ascending) trend before (after) 2000, and the ascending trend in winter was stronger than that in summer. Wang et al. (2017) also found a dramatic increase in the difference between GST and surface air temperature (SAT) in northern China in winter since 2005. The observation dataset they used had been subjected to initial quality control by the China Meteorological Administration (CMA). However, no homogeneity check and correction was performed.
Moreover, lack of reliable observational GST data makes assessing model simulations difficult. Considering the important role of GST in land–atmosphere interaction, the quality of reanalysis products of the GST in China was assessed in several studies (e.g., Yang and Zhang 2018; Zhou et al. 2017). Compared with observations obtained in China during 1979–2003, Zhou et al. (2017) examined the GST obtained from eight widely used reanalysis datasets and showed that the interannual variability of GST in these reanalysis datasets well resembles that of the observations in most parts of China, but not for the Tibetan Plateau. Yang and Zhang (2018) compared soil temperatures from the ERA-Interim/Land, MERRA-2, CFSR, and Global Land Data Assimilation System, version 2.0 (GLDAS-2.0) with the observations obtained in China during 1981–2005. In the two studies, the authors noticed that the observation system had been changed from manual to automatic measurements near 2005, and higher temperatures might be obtained from automatic observations. To avoid the impact of observing system changes, the end points of their study period are 2003 and 2005, respectively.
The long records of observed climate data are generally subject to inhomogeneities due to various factors, such as relocation of the observation site, change of observation method, and transition to automated instruments (Mitchell 1953; Heino 1994). Analysis based on parallel observations of collocated manual and automatic weather stations in China showed that there existed serious biases of GST between the two systems (Liu et al. 2008; Ren et al. 2013). Specifically, there was a large difference in daily mean GST between automatic and manual observations in northern China, which was primarily attributed to biases in the automatic system under snow-cover condition during the cold season. Therefore, the transition of observation systems may have led to serious inhomogeneity in the historical records of GST in China. Also, other nonclimatic factors (e.g., relocation of station and change in observation environment) may affect the homogeneity of GST data (Liu et al. 2008). However, the historical records of GST datasets used in the past studies in China generally lack homogenization prior to these analyses; only Zhou et al. (2017) adjusted the inhomogeneities in GST data before 2003 using the RHtests software package. Note that the metadata information, which is one key criterion for inhomogeneity test and adjustment, was not applied in their work.
In the present study, we aim to remove the influence of nonclimatic factors on the inhomogeneities of historical records of GST in China. We construct a homogenized dataset of monthly GST for the period from 1961 to 2016. Based on this homogenized dataset, we evaluate the performances of five different reanalysis products from GLDAS in terms of temporal and spatial variation of average GST in China.
The remainder of this paper is organized as follows. In section 2, we describe the datasets and methods used in this study. In section 3, we present detailed process in the homogenization of GST in China. In section 4, we show the temporal and spatial trends of GST derived from the homogenized dataset during 1961–2016. In section 5, we evaluate the performances of GLDAS reanalysis products against the homogenized dataset, followed by a summary in section 6.
2. Data and methods
The data used in the present study are listed below.
The dataset of fundamental daily surface meteorological elements in China (V3.0), which is publicly released by the National Meteorological Information Center (NMIC) of the CMA. Quality control had been completed before this dataset was released. Threshold value check, allowable value check, station extreme-value check, instantaneous-value check, internal consistency check between daily mean and daily extreme values, temporal consistency check, spatial consistency check, and manual review were conducted. However, no homogeneity check and correction was performed. GST elements include daily mean, daily maximum, and daily minimum ground surface temperature.
The dataset of manual and automatic weather station observations for the period of 2000–07, which is archived at the NMIC. This dataset includes 791 weather stations, and each station provides 1–2 years of parallel observations of manual and automatic weather stations (Fig. 1). These parallel observations are used to correct systematic errors caused by the transition from manual to automatic observations.
The ground weather station metadata dataset of China (V1.0), published by the NMIC. This dataset is implemented to identify discontinuities in the historic ground surface temperature data caused by various factors, such as change in station location. This dataset consists of all metadata collected at national weather stations since these stations were set up, including more than 20 items, such as station name, station ID number, station level, station location, obstacles near the station, observational elements, and observational instruments.
Geographic locations of weather stations in China with collocated automatic and manual observations of GST. Colors indicate the first year of the collocated observation periods.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Geographic locations of weather stations in China with collocated automatic and manual observations of GST. Colors indicate the first year of the collocated observation periods.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Geographic locations of weather stations in China with collocated automatic and manual observations of GST. Colors indicate the first year of the collocated observation periods.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Five reanalysis products of average surface skin temperature (AvgSurfT) for 1979–2016 produced by GLDAS (Rodell et al. 2004; Rui 2011) are also used in the present study. These reanalysis products include 1) Noah 1°× 1° monthly mean reanalysis product in GLDAS-1 (hereafter Noah-V1); 2) MOS 1°× 1° monthly mean reanalysis product in GLDAS-1; 3) CLM 1°×1° monthly mean reanalysis product in GLDAS-1; 4) Noah 0.25°× 0.25° monthly mean reanalysis product in GLDAS-2.0 (hereafter Noah-V2; Rui and Beaudoing 2015); and 5) Catchment Land Surface Model (CLSM) 0.25°×0.25° daily reanalysis product in GLDAS-2.0. The GLDAS datasets are archived and distributed by the Goddard Earth Sciences (GES) Data and Information Services Center (DISC).
Homogenization in this study is conducted by using the penalized maximal F-test method with the aid of RHtestsV5 software (hereafter referred to as PMFred; Wang 2008; Wang and Feng 2013). PMFred is applicable for the test with no reference climatic series. The algorithm accounts for the effect of autocorrelated noise term empirically and deals with multiple changepoints using a recursive testing algorithm. The effects of unequal sample sizes on the detection power were also diminished using an empirical function in the algorithm. The mean adjustments in the study are derived by comparing the mean in the period before the shift to that after the shift. A period of up to 10 years before and after each shift is normally used; however, because of the existence of several shifts, the period is sometimes shortened but should be no less than 5 months. In the PMFred test, multiple changepoints in the series are detected sequentially by segmentation through a regression testing algorithm, and the mean adjustment is also calculated by segmentation.
In this study, the annual mean and seasonal mean are calculated from monthly values. The anomalies are computed relative to the period of 1981–2010.
3. Homogenization of GST in China
The homogenization of monthly GST during 1961–2016 is aimed to solve two major problems. First, by using the daily parallel observations from collocated manual and automatic weather stations, we can remove the inhomogeneities caused by transition to automated instruments. Second, based on the first step, we further utilize the PMFred method and metadata to diminish inhomogeneities caused by other nonnatural factors (e.g., relocation of station, environmental change around the station, and change in observational time) in the monthly mean GST. Based on these two steps, the homogenized dataset of monthly GST during 1961–2016 is established, as shown in the following sections.
a. Source of biases in daily mean GST
As has taken place in the majority of the world’s national meteorological services, CMA stations have gradually advanced from using manual to automated systems since the year 2000. However, replacement of manual observational instruments should be done with great care. The World Meteorological Organization (WMO) suggests that a sufficient period of parallel observations by collocated manual and automated systems should be considered to ensure that biases between the two systems can be identified (WMO 2017).
According to Specifications for Surface Meteorological Observation (CMA 2003) in China, when measuring the ground surface temperature, half of the sensing part is buried in the soil, and half is exposed to the ground. The part buried in the soil must be closely attached to the soil with no space, and the exposed part should be kept clean. However, it is necessary to elaborate on the difference between manual and automatic observation systems of GST in China. First, the instruments used by manual and automatic observation are different. The automatic observation employs platinum resistance temperature sensors, while the manual observation uses a glass liquid thermometer. Second, there is a distance of about 30 cm between the manual and automatic observation sites; this factor always has some small effects. Third, the manual and automatic observation methods are different when there is snow cover on the ground. The temperature sensor is always above the snow surface for manual observation, and thus the measured temperature is that above the snow surface. For the automatic observation, the measured temperature is that under the snow, but this circumstance will be noted in the observational record. That is, when there is snow on the ground, the automatic temperature sensor is covered by snow, and the measured temperature is that under the snow surface instead of that above the snow surface. Liu et al. (2008) analyzed the effects on manual and automatic observations from the abovementioned three factors and pointed out that when there was snow covering the ground in winter, the GST from automatic observations was higher than that from manual observations. The obvious difference is mainly attributed to different observational methods between the two kinds of observation systems. Without the influence of snow, the difference is mainly caused by the type of instrument between the two kinds of observations.
It is worth noting that the daily mean is calculated by average data from the four observations at 0200, 0800, 1400, and 2000 Beijing standard time (BST; UTC + 8 h) since 1959, but at 0100, 0700, 1300, and 1900 BST before 1959. Previous studies showed that the standard deviation of difference between manual and automatic observations of daily mean GST in China is about 2.0°C (e.g., Liu et al. 2008). Therefore, a value of 10°C (i.e., 5 times the standard deviation) is adopted as the threshold for judging anomalous behavior between manual and automatic observations. Analysis of differences between manual and automatic observations at the 791 stations in China during their parallel observation periods (Fig. 1) reveals that absolute differences equal to or greater than 10°C for more than 10 days are found at 111 stations (Fig. 2). Further analysis reveals two types of problems.
As in Fig. 1, but for the 111 stations with problems in parallel observations of GST. The filled circle indicates the station with first type of problem, and the triangle indicates the second type of problem.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
As in Fig. 1, but for the 111 stations with problems in parallel observations of GST. The filled circle indicates the station with first type of problem, and the triangle indicates the second type of problem.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
As in Fig. 1, but for the 111 stations with problems in parallel observations of GST. The filled circle indicates the station with first type of problem, and the triangle indicates the second type of problem.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
The first type of problem was seen at 105 stations, which is attributed to the abovementioned third factor. These stations are mainly concentrated in Northeast China and Xinjiang Province. Figure 3a displays the situation at Jiaohe station of Jilin Province as an example of this type of problem. Such a bias is primarily related to different observation methods between manual and automatic systems when there is snow on the ground. Most of the incident solar radiation is absorbed when it goes through the snow layer, and only a small part can reach the soil surface. Thereby, the diurnal variation of temperature is small below the snow layer and large above the snow surface. In addition, snow is a good insulator, which explains why the difference between temperature over the bare soil and that below the snow layer increases with snow depth. Surface temperature below the snow layer changes little, with small diurnal variation. In contrast, the temperature variation above the snow is relatively large due to the influence of solar radiation. Therefore, the difference in GST between snow-covered soil and bare soil increases as the snow cover becomes thicker. In Northeast China and Xinjiang Province, where snow is easily accumulated in winter, the automatic observations are mainly maintained near 0°C. To address the problem that automatic observations cannot realistically represent surface temperature, snow temperature observation was added to surface meteorological observations in China. In accordance with CMA (2003), when the temperature sensor is covered by snow, it should be relocated to the snow surface to measure temperature over the snow surface. The requirement for snow surface temperature measurement is the same as that for manual ground surface temperature measurement. Liu et al. (2008) investigated the relationship between automatic snow surface temperature observations and manual ground surface temperature observations. They found little difference between observations obtained by the two different methods and suggested that snow temperature observations in northern China can be used to represent ground surface temperature. Unfortunately, snow temperature observations were not collected before 2007 and cannot be used to replace ground surface temperature observations at the 105 stations. For this reason, the data at these 105 stations are not processed in the present study and are not used for computation.
Time series of manual (blue) and automatic (red) observations of daily mean GST (°C): (a) Jiaohe station during 2004–05 and (b) Dege station during 2003–04.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Time series of manual (blue) and automatic (red) observations of daily mean GST (°C): (a) Jiaohe station during 2004–05 and (b) Dege station during 2003–04.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Time series of manual (blue) and automatic (red) observations of daily mean GST (°C): (a) Jiaohe station during 2004–05 and (b) Dege station during 2003–04.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
There are six stations in question where abnormal automatic observations are used. These stations are mainly located in Shanxi, Sichuan, and Fujian Provinces. Figure 3b presents the situation at Dege station of Sichuan, which is one such station. These observation anomalies mostly occurred in the first year of parallel observations, when automatic observations were still in debugging and testing stages. After formal operation of the automatic weather stations started, most of these anomalies no longer appeared. Since the data from parallel observations are important reference data for systematic error correction, we conduct a threshold check (5 times the standard deviation) of parallel observations of daily GST collected at the remaining 686 stations. The results show that 3 records from manual observations and 254 records from automatic observations are considered as outliers, accounting for 0.0006% and 0.0531%, respectively, of all the nonmissing daily parallel observation data. We exclude these outliers from our study.
b. Bias correction of daily mean GST
To remove the bias caused by the two kinds of observation, a linear regression equation is established based on parallel observations, in which the manual observation is taken as the independent variable, and the automatic observation as the dependent one. A significance test is also conducted for the regression by using the F test. Figure 4 depicts the distribution of corrected biases. We can see that the 492 stations in northern, northwestern, and northeastern China present positive values, while the other 194 stations in southern and southeastern China show negative values. Figure 5 displays the difference in linear trend of annual-mean GST after and before bias correction (former minus latter) during 1961–2016. We can see that the stations with the reduced trend at the 95% confidence level are mainly located in northern China, especially in the northwest. In contrast, there are several stations exhibiting enhanced trends, but only one of them is significant at the 95% confidence level.
Corrected bias of daily GST data (°C) in China.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Corrected bias of daily GST data (°C) in China.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Corrected bias of daily GST data (°C) in China.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Differences in linear trend (°C yr−1) of the annual-mean GST after and before bias correction (former minus latter) during 1961–2016. Filled triangles indicate the stations with significance at the 95% confidence level by using the Kendall test. The size of the triangle is proportional to the magnitude of the linear trend in steps.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Differences in linear trend (°C yr−1) of the annual-mean GST after and before bias correction (former minus latter) during 1961–2016. Filled triangles indicate the stations with significance at the 95% confidence level by using the Kendall test. The size of the triangle is proportional to the magnitude of the linear trend in steps.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Differences in linear trend (°C yr−1) of the annual-mean GST after and before bias correction (former minus latter) during 1961–2016. Filled triangles indicate the stations with significance at the 95% confidence level by using the Kendall test. The size of the triangle is proportional to the magnitude of the linear trend in steps.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
The above results indicate that the change in observation method from manual to automatic systems resulted in some inhomogeneities in the historical records of GST. These inhomogeneities led to an obvious overestimation of the warming trend over part of northern China.
c. Homogenization of monthly mean GST
Based on the bias-corrected daily mean GST, monthly mean time series are constructed for the 686 stations in China. Then, we further reduce the inhomogeneities caused by other factors (e.g., changes of station location, surrounding environment of station, and observation time) in the monthly mean data.
GST is usually under the influence of soil color, thermal conductivity, thermal capacity, and other factors (Wang et al. 2002); therefore, its variation exhibits distinct local features. This implies that it is difficult to find a homogenous representative reference data series from nearby stations. Here, we adopt the PMFred method to conduct a statistical test on the monthly mean and annual-mean GST in China. To reduce uncertainty in the statistical test, we also use metadata information to determine discontinuities. The main criteria for identifying discontinuities are as follows: 1) if a discontinuity can be detected from both monthly mean and annual-mean historical records, this discontinuity will be treated as the ultimate discontinuity, and 2) if a discontinuity is detected from monthly mean historical record and confirmed by the metadata, this discontinuity is identified as the ultimate discontinuity. Table 1 and Fig. 6 illustrate the statistical results of identified discontinuities at all the stations, and the distribution of the inhomogeneous stations. We find ultimately that discontinuities exist at 109 stations, accounting for ~16% of the total 686 stations. Out of the 109 inhomogeneous stations, 10 of them contain two discontinuities in the historical records, while the other 99 stations contain just one discontinuity. For all 119 discontinuities across the 109 stations, there are 12 discontinuities that could be detected simultaneously by satisfying both criteria 1 and 2, 24 discontinuities by just satisfying 1, and 83 discontinuities by just satisfying 2.
The number of stations with zero, one, two, and three or more discontinuities that have been identified and adjusted in the monthly GST time series. Total discontinuities = 119.
Spatial distribution of the numbers of discontinuity in monthly GST data of 109 stations during 1961–2016.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Spatial distribution of the numbers of discontinuity in monthly GST data of 109 stations during 1961–2016.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Spatial distribution of the numbers of discontinuity in monthly GST data of 109 stations during 1961–2016.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
By using the metadata, we find that 56 out of the 119 discontinuities were caused by relocation of the station, accounting for ~47% of the total discontinuities. This indicates that relocation of the station was the main cause for the inhomogeneity of historical monthly mean ground surface temperature data. For instance, results obtained from the homogeneity test for Linhe station (53513) in northern China show that a discontinuity occurred around December 2009. When compared with the metadata of the station, we find that this station changed its location on 1 January 2010, with the new site being 4500 m away and about 100-m change in the vertical direction (not shown). Some discontinuities, accounting for 13% of the total, are found to be related to the change in surrounding environment of the station, such as that at Ruili station (56838) in southwestern China. For this station, a discontinuity in August 1993 was detected by the homogeneity test and examining the metadata. We find that this discontinuity was actually caused by significant change in surrounding environment of the station as several new buildings were put up around the station during 1993/94 (not shown). Other discontinuities are found to be influenced by change of observation time. For example, Chuxiong station (56768) used 0100, 0700, 1300, and 1900 BST to collect observations for its daily mean before June 1959, but adopted 0200, 0800, 1400, and 2000 BST after that. Therefore, the discontinuity around June 1959 was caused by change of observation time (not shown).
Based on the inhomogeneity analysis of the 109 stations, we utilize mean adjustment method to adjust the abovementioned 119 discontinuities in the historical records. Figure 7 shows the difference in linear trend of annual-mean GST after and before the adjustment (former minus latter) for the 109 stations during 1961–2016. The filled triangles indicate the difference in linear trend after and before adjustment with significance at the 95% confidence level, and the linear trend after adjustment in these stations is considered to have undergone a significant change. We can see that 51 stations exhibit intensified warming trends after the adjustment, and 36 out of the 51 stations show significant changes. In contrast, 58 stations present decreased warming trends after the adjustment, and 52 of them show significant difference after the adjustment.
As in Fig. 5, but for the difference after and before inhomogeneity correction (former minus latter) for the 109 stations shown in Fig. 6.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
As in Fig. 5, but for the difference after and before inhomogeneity correction (former minus latter) for the 109 stations shown in Fig. 6.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
As in Fig. 5, but for the difference after and before inhomogeneity correction (former minus latter) for the 109 stations shown in Fig. 6.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Figure 8 illustrates two cases of cooling and warming trends after the adjustment. In Fig. 8a, the historical GST data at Zengcheng station (59294) in southern China show a significant warming trend before the adjustment, with a value of 0.3196°C decade−1. In January 2001, its location was changed, which caused discontinuity in the historical records. After the adjustment, the data present a cooling trend with a value of −0.0703°C decade−1. For the GST time series at Linhe station (53513) in northern China, a warming trend with a value of 0.4871°C decade−1 is seen before the adjustment. In the unadjusted time series, we can easily find the discontinuity around January 2010 when the relocation of station occurred. After the adjustment, the data show an enhanced warming trend of a value of 0.7361°C decade−1 (Fig. 8b).
Time series of annual-mean GST (°C) before (blue) and after (red) inhomogeneity correction for (a) Zengcheng station (59294) and (b) Linhe station (53513) during 1961–2016.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Time series of annual-mean GST (°C) before (blue) and after (red) inhomogeneity correction for (a) Zengcheng station (59294) and (b) Linhe station (53513) during 1961–2016.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Time series of annual-mean GST (°C) before (blue) and after (red) inhomogeneity correction for (a) Zengcheng station (59294) and (b) Linhe station (53513) during 1961–2016.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
The above results indicate that these factors, such as change of station location, change of surrounding environment of station, and change of observation time, are the main causes for the inhomogeneities in monthly mean GST in China. By utilizing the PMFred method and metadata, the inhomogeneities in the historical records of the monthly mean GST caused by these factors are significantly diminished.
4. Spatial and temporal characteristics of GST in China
Using this homogenized dataset, we analyze temporal and spatial variation in the monthly mean GST of China.
a. Temporal variation
Table 2 shows the annual-mean homogenized GST averaged over China with a significant warming trend of about 0.273°C decade−1 during 1961–2016, weaker than the warming trend of 0.323°C decade−1 from the original series. The overall change can be roughly divided into three phases, which are characterized by a weakly decreasing trend in the 1960s, followed by an increasing trend from the 1970s to the end of the 1990s, and then by a significantly slowed-down increase after the 2000s (Fig. 9). Compared to the original series, the homogenized series were somewhat warmer before 2005 and somewhat colder after 2005. Compared to the trends of the original seasonal series (Table 2), the bias adjustment and homogeneity decrease the trends of ground surface temperature in all four seasons. The linear trend of homogenized GST averaged over China exhibits great seasonal variation during 1961–2016 (see Fig. 10, Table 2). The linear trend is 0.312°C decade−1 in spring and 0.242°C decade−1 in fall. The smallest trend is in summer, with a value of 0.173°C decade−1, while the largest trend occurs in winter, with a value of 0.321°C decade−1.
Linear trends (°C decade−1) of the annual-mean and seasonal-mean time series of original and homogenized GST averaged over China during 1961–2016. An asterisk indicates significance at the 95% confidence level.
Original (black) and homogenized (red) time series of annual-mean GST averaged over China and its linear trend (black) during 1961–2016.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Original (black) and homogenized (red) time series of annual-mean GST averaged over China and its linear trend (black) during 1961–2016.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Original (black) and homogenized (red) time series of annual-mean GST averaged over China and its linear trend (black) during 1961–2016.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
As in Fig. 9, but for seasonal-mean GST: (a) spring, (b) summer, (c) fall, and (d) winter.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
As in Fig. 9, but for seasonal-mean GST: (a) spring, (b) summer, (c) fall, and (d) winter.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
As in Fig. 9, but for seasonal-mean GST: (a) spring, (b) summer, (c) fall, and (d) winter.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
In general, annual and seasonal variations in the linear trend, especially the amplitude of the trend, are quite different from those shown by Wang et al. (2016) and Wang et al. (2017). The differences are largely attributed to two reasons. First, the observations of stations in northeastern China are excluded in this study, where the automatic observations in winter should be used with caution (such as the above 105 stations mentioned in section 3a). Second, the bias between manual and automatic systems and the inhomogeneities caused by other nonclimatic factors are adjusted in this study.
b. Spatial variation
Figure 11 displays the spatial distribution of linear trend of annual-mean homogenized GST in China during 1961–2016. The annual-mean GST exhibits a significant warming trend over most parts of China, except at a few stations in central–southern China, where a significant cooling trend exists. Note that the magnitude of warming trend is generally larger in northern China and over the Tibetan Plateau than in southern China.
Spatial distribution of linear trend (°C decade−1) of annual-mean homogenized GST during 1961–2016. Filled triangles indicate the stations with significance at the 95% confidence level by using the Kendall test. The size of triangle is proportional to the magnitude of the linear trend in steps.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Spatial distribution of linear trend (°C decade−1) of annual-mean homogenized GST during 1961–2016. Filled triangles indicate the stations with significance at the 95% confidence level by using the Kendall test. The size of triangle is proportional to the magnitude of the linear trend in steps.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Spatial distribution of linear trend (°C decade−1) of annual-mean homogenized GST during 1961–2016. Filled triangles indicate the stations with significance at the 95% confidence level by using the Kendall test. The size of triangle is proportional to the magnitude of the linear trend in steps.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
A common feature of linear trend in each of the four seasons is that the magnitude of warming trend in northern China is larger than that in the south (Fig. 12), although distinct seasonal variation exists. In spring, GST increases in most parts of China, except at a few stations in southern China where a statistically nonsignificant decreasing trend exists (Fig. 12a). In contrast, GST shows a cooling trend in central–southern China in summer, although a warming trend is observed at the majority of stations in northern China and the Tibetan Plateau (Fig. 12b). In fall (Fig. 12c), GST shows a warming trend over most parts of China, but the magnitude of the trend is generally smaller than that in spring and winter. In winter (Fig. 12d), GST experiences the most significant warming over the whole country, with a notable trend higher than 0.6°C decade−1 in northwestern China.
As in Fig. 11, but for seasonal-mean GST: (a) spring, (b) summer, (c) fall, and (d) winter.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
As in Fig. 11, but for seasonal-mean GST: (a) spring, (b) summer, (c) fall, and (d) winter.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
As in Fig. 11, but for seasonal-mean GST: (a) spring, (b) summer, (c) fall, and (d) winter.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
5. Evaluation of GLDAS reanalysis products
Using the homogenized dataset, we can evaluate the performance of GLDAS reanalysis products in terms of linear trend in China. The grid data of GLDAS reanalysis products are interpolated to the 686 stations by using the nearest-neighbor value interpolation method.
a. Temporal variation
Figure 13a shows the time series of annual-mean GST anomaly, regionally averaged over China from the homogenized observation and the five reanalysis products of GLDAS during 1980–2016. Among the five reanalysis products, the variation in annual-mean GST from GLDAS-2.0 is generally consistent with that of the observation, with a correlation coefficient of 0.88 for Noah-V2 during the period of 1980–2010 and 0.95 for CLSM during the period of 1980–2014. The time series of Noah-V2 and CLSM before 2005 are more consistent with the observation than those after 2005. The annual-mean GST from Noah-V2 and CLSM were obviously warmer than that from the observation before 2005 and cooler than observations after 2005. The correlation coefficients of annual-mean GST between the observation and the remaining three products (CLM, MOS, and Noah-V1) are 0.05, −0.19, and 0.20, respectively, during 1980–2016, none of which is significant at the 95% confidence level. This indicates these three products exhibit poor performances in terms of interannual variation of annual-mean GST in China.
Time series of (a) annual mean and seasonal mean in (b) spring, (c) summer, (d) fall, and (e) winter of GST anomalies averaged over China for the homogenized observation (solid line) and five different reanalysis products of GLDAS (dashed lines) during 1980–2016.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Time series of (a) annual mean and seasonal mean in (b) spring, (c) summer, (d) fall, and (e) winter of GST anomalies averaged over China for the homogenized observation (solid line) and five different reanalysis products of GLDAS (dashed lines) during 1980–2016.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Time series of (a) annual mean and seasonal mean in (b) spring, (c) summer, (d) fall, and (e) winter of GST anomalies averaged over China for the homogenized observation (solid line) and five different reanalysis products of GLDAS (dashed lines) during 1980–2016.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Further analysis shows that the observed linear trend of annual-mean GST regionally averaged over China is 0.356°C decade−1 during 1980–2016, exceeding the 95% confidence level (Table 3). The Noah-V2 and CLSM show significant warming trend of 0.355°C decade−1 during 1980–2010 and of 0.203°C decade−1 during 1980–2014, respectively, which are consistent with the observations. In contrast, the CLM, MOS, and Noah-V1 all show cooling trends during 1980–2016, none exceeding the 95% confidence level. This indicates that the CLM, MOS, and Noah-V1 all underestimate the warming trend of annual-mean GST in China during 1980–2016.
As in Table 2, but for the homogenized observation and the five different reanalysis products from GLDAS during 1980–2016 (Noah-V2 only covers 1980–2010, while CLSM covers 1980–2014).
Figures 13b–e show the time series of seasonal-mean GST anomaly regionally averaged over China from the homogenized observation and the five reanalysis products from GLDAS during 1980–2016. We can see the performances of the five GLDAS reanalysis products are rather different in terms of seasonal-mean anomaly in China. For Noah-V2, the variation in GST well resembles that of the observation in spring and summer during 1980–2010, with correlation coefficients of 0.95 and 0.94 between the observed and assimilated time series in spring and summer, respectively. The correlation coefficients between the observed and assimilated GST are 0.89 and 0.83 in fall and winter, respectively. For the CLSM, the variation in GST well resembles that of the observation in spring and summer during 1980–2014, with correlation coefficients of 0.97 and 0.96 between the observed and assimilated time series in spring and summer, respectively. The correlation coefficients between the observed and assimilated GST are 0.94 and 0.91 in fall and winter, respectively. In a word, the GLDAS-2.0 products exhibit better performances in terms of variation in warm seasons than in cold seasons. For the CLM, MOS, and Noah-V1, the variation in GST in spring is more consistent with that of the observation during 1980–2016, compared to the other three seasons. The correlation coefficients of spring-mean GST between the observations and the reanalysis (CLM, MOS, and Noah-V1) are 0.60, 0.42, and 0.67, respectively, all significant at the 95% confidence level. The CLM, MOS, and Noah-V1 also show better performances in terms of winter-mean GST in China during 1985–99, with correlation coefficients of 0.90, 0.78, and 0.95 with the observation, respectively. These three reanalysis products exhibit the worst performance in terms of summer-mean GST in China during 1980–2016, with correlation coefficients of 0.34, 0.44, and 0.49 with the observations, respectively.
As seen in Table 3, the observed seasonal-mean GST regionally averaged over China exhibits a significant warming trend in all four seasons, with the highest value of 0.476°C decade−1 in spring and the lowest of 0.240°C decade−1 in summer. Among the five reanalysis products, Noah-V2 and CLSM generally show warming trends similar to that of the observation in all four seasons. The CLM, MOS, and Noah-V1, however, show cooling trends in all but spring, opposite to the observation.
There are several ways to explain the above results. It is important to note that the GLDAS datasets are based on models to generate various land surface variables from other meteorological data. The GLDAS-1 simulations are forced with a combination of GDAS, disaggregated CMAP, and AFWA radiation datasets. Therefore, the climatological consistency of GLDAS-1 is not very good. The main objective for GLDAS-2.0 is to create more climatologically consistent datasets, using the Global Meteorological Forcing Dataset from Princeton University (Sheffield et al. 2006), which is a reanalysis product that has been bias corrected using observation-based products, currently extended from 1948 to 2010. Therefore, GLDAS-2.0 shows higher correlation with the observation than GLDAS-1, but also has obvious differences with the observation after 2005, especially in fall and winter. In other words, the process of updating forcing data sources may introduce inhomogeneity into GLDAS-2.0, which results in the lower correlation after 2005, especially in fall and winter.
b. Spatial variation
Figure 14 shows the distribution pattern of linear trend of annual-mean GST from the homogenized observation and the five GLDAS reanalysis products during 1980–2016. The observed GST exhibits a warming trend over most parts of China, except at a few stations in southern China. The Noah-V2 and CLSM well capture most of the observed warming trend distribution in China, likely because bias-corrected forcing data were applied to GLDAS-2.0 (Sheffield et al. 2006; Rui and Beaudoing 2015). Note that the Noah-V2 tends to underestimate the warming trend in northwestern China and fails to reproduce the cooling trend in parts of southern China. The CLSM tends to underestimate the warming trend at some stations in southwestern and northwestern China. The other three products generally fail to reproduce the spatial pattern of the observed linear trend; in particular, they all show significant cooling trends in China, except in central China and part of western China.
As in Fig. 11, but for (a) homogenized observation, (b) Noah-V2 during 1980–2010, (c) CLSM during 1980–2014, (d) Noah-V1, (e) CLM, and (f) MOS during 1980–2016 with units of °C yr−1.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
As in Fig. 11, but for (a) homogenized observation, (b) Noah-V2 during 1980–2010, (c) CLSM during 1980–2014, (d) Noah-V1, (e) CLM, and (f) MOS during 1980–2016 with units of °C yr−1.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
As in Fig. 11, but for (a) homogenized observation, (b) Noah-V2 during 1980–2010, (c) CLSM during 1980–2014, (d) Noah-V1, (e) CLM, and (f) MOS during 1980–2016 with units of °C yr−1.
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
Figure 15 shows the spatial distribution of differences in linear trends of annual-mean GST between each of the five reanalysis products and the homogenized observation. The results confirm that the warming trend is seriously underestimated in most parts of China but overestimated in central China by CLM, MOS, and Noah-V1. In contrast, Noah-V2 and CLSM generally overestimate the warming trend in central China, part of southwestern China, and the coastal region of southern China, while they tend to underestimate the warming trend in western China. Similar results are obtained for differences in linear trends of seasonal-mean GST between each of the five reanalysis products and the homogenized observation (not shown).
As in Fig. 14, but for the difference between individual GLDAS reanalysis and the homogenized observation (former minus latter).
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
As in Fig. 14, but for the difference between individual GLDAS reanalysis and the homogenized observation (former minus latter).
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
As in Fig. 14, but for the difference between individual GLDAS reanalysis and the homogenized observation (former minus latter).
Citation: Journal of Climate 32, 4; 10.1175/JCLI-D-18-0275.1
6. Summary
In this study, we use the GST data from the National Reference and Basic Meteorological Stations dataset V3.0 of China, the parallel observations of collocated manual and automatic weather stations in China, and the metadata of weather stations to correct the biases between daily automatic and manual observations of GST. We then construct a monthly mean GST dataset for 686 stations in China by removing the inhomogeneities caused by nonclimatic factors, such as relocation of station and change in observational practice. The bias analysis using observation data from collocated manual and automatic stations shows that the automatic observation records in winter from the majority of the stations over northeastern and northwestern China contain serious errors. These biases are caused by the observational practice difference between manual and automated instruments under snow cover. Using these parallel observation data and linear regression method, we successfully correct biases between manual and automatic observations due to the change in observational system. We further utilize the PMFred method to diminish inhomogeneities in the historical records of monthly mean GST caused by other factors. Our analysis shows that 109 of the 686 stations have discontinuous points, accounting for ~16% of the total stations used in this study. Among them, 56 discontinuities are caused by relocation of station, which is about 47% of the total discontinuities. These results indicate that relocation of station is the second major cause for the inhomogeneity of monthly mean GST in China.
Based on this homogenized dataset, we analyze the spatial and temporal variations of monthly mean GST during 1961–2016. Results indicate that the linear trend of annual-mean GST in China is 0.273°C decade−1. The warming trend is most obvious in winter and spring, at 0.321°C and 0.312°C decade−1, respectively, and weakest in summer, at 0.173°C decade−1. Considering the spatial distribution of linear trends of annual-mean GST, most parts of China experience warming trends, except southern China. The warming trend is more significant in northern China than in southern China, with the maximum warming in northwestern China and the Tibetan Plateau. The seasonal-mean GST exhibits a warming trend over most parts of China in all seasons, with the largest value in winter. In contrast, GST shows a cooling trend in southeastern China in summer.
The five GLDAS reanalysis products have been widely used in climate studies. However, the quality of these reanalysis data in China is not well assessed due to the lack of long-term homogenized observations. Using the homogenized dataset for the period of 1980–2016, we evaluate the temporal and spatial variations of assimilated average surface skin temperatures (AvgSurfT) from the five products. It is shown that the Noah-V2 and the CLSM reanalysis from GLDAS-2.0 have the best performance in reproducing observed warming trends for both annual-mean and seasonal-mean GST in China. However, the Noah-V1, CLM, and MOS reanalysis products from GLDAS-1 all fail to reproduce the warming trend. For spatial distribution, the annual-mean trends in Noah-V1, CLM, and MOS are generally underestimated for most regions of China, but overestimated over central China. The spatial distributions of Noah-V2 and CLSM are similar to those of the observation. Seasonally, the amplitude of trend differences between reanalysis and observations is larger in fall and winter than in spring and summer.
Previous studies used historical observations to focus on quality assessment of modeled GST in China in different reanalyses (e.g., Zhou et al. 2017; Yang and Zhang 2018) and on revision of land surface parameterizations (e.g., surface roughness and surface conductivity) to improve GST in reanalysis (e.g., Zeng et al. 2012; Wang 2014). The historical GST data that are used as a reference to calibrate land surface parameterizations, however, contain systematic biases. We anticipate that our high-quality, homogenized GST data help improve the land surface parameterizations in these models and thus improve these reanalysis products.
In this study, we mainly evaluate the trend performance of GLDAS reanalysis products using bias-corrected GST data in China. In the East Asian monsoon region, the knowledge of the intraseasonal and interannual variations of the monsoon is also important for us to fully understand monsoon activity, especially from the viewpoint of land surface processes. In this respect, we need to explore the interseasonal and interannual variations of GST in China and to evaluate the performance of GLDAS reanalysis products in the near future.
Acknowledgments
The authors thank the two anonymous reviewers whose constructive comments are helpful for improving the overall quality of the paper. The authors also thank Dr. Ziyan Zheng for his insightful suggestions, which helped us improve this work significantly. This research is jointly supported by the National Innovation Project for Meteorological Science and Technology: Quality Control, Fusion, and Reanalysis of Meteorological Observations, the National Key Research and Development Program of China (Grant 2016YFA0601501), the China Meteorological Administration Special Foundation for Climate Change (CCSF201803), the Jiangsu Collaborative Innovation Center for Climate Change of China, and the Zhuhai Joint Innovative Center for Climate, Environment and Ecosystem.
REFERENCES
CMA, 2003: Specifications for Surface Meteorological Observation (in Chinese). China Meteorological Press, 151 pp.
Fan, X., 2009: Impacts of soil heating condition on precipitation simulations in the Weather Research and Forecasting Model. Mon. Wea. Rev., 137, 2263–2285, https://doi.org/10.1175/2009MWR2684.1.
Heino, R., 1994: Climate in Finland during the Period of Meteorological Observations. Finnish Meteorological Institute, 209 pp.
Hu, Q., and S. Feng, 2004: Why has the land memory changed? J. Climate, 17, 3236–3243, https://doi.org/10.1175/1520-0442(2004)017<3236:WHTLMC>2.0.CO;2.
Liu, X. N., Z. H. Ren, and Y. Wang, 2008: Differences between automatic-observed and manual-observed surface temperature (in Chinese). J. Appl. Meteor. Sci., 19, 554–563.
Liu, Y., and R. Avissar, 1999: A study of persistence in the land–atmosphere system using a general circulation model and observations. J. Climate, 12, 2139–2153, https://doi.org/10.1175/1520-0442(1999)012<2139:ASOPIT>2.0.CO;2.
Mitchell, J. M., 1953: On the causes of instrumentally observed secular temperature trends. J. Meteor., 10, 244–261, https://doi.org/10.1175/1520-0469(1953)010<0244:OTCOIO>2.0.CO;2.
Ren, Z. H., G. A. Wang, and F. L. Zhou, 2013: Difference between soil temperatures obtained through automatic observation and manual observation and analysis of its causes (in Chinese). Turang Xuebao, 50, 657–663.
Rodell, M., and Coauthors, 2004: The Global Land Data Assimilation System. Bull. Amer. Meteor. Soc., 85, 381–394, https://doi.org/10.1175/BAMS-85-3-381.
Rui, H., 2011: Global Land Data Assimilation System version 1 (GLDAS-1) products. NASA GSFC Rep., 32 pp., https://hydro1.gesdisc.eosdis.nasa.gov/data/s4pa/GLDAS_V1/README.GLDAS.pdf.
Rui, H., and H. Beaudoing, 2015: Global Land Data Assimilation System version 2 (GLDAS-2) products. NASA GES DISC Rep., 22 pp., https://hydro1.gesdisc.eosdis.nasa.gov/data/s4pa/GLDAS/README_GLDAS2.pdf.
Sheffield, J., G. Goteti, and E. F. Wood, 2006: Development of a 50-yr high-resolution global dataset of meteorological forcings for land surface modeling. J. Climate, 19, 3088–3111, https://doi.org/10.1175/JCLI3790.1.
Wang, J. L., and Coauthors, 2016: Variation in ground temperature at a depth of 0 cm and the relationship with air temperature in China from 1961 to 2010 (in Chinese). J. Resour. Sci., 38, 1733–1741, https://doi.org/10.18402/resci.2016.09.11.
Wang, K., 2014: Measurement biases explain discrepancies between the observed and simulated decadal variability of surface incident solar radiation. Sci. Rep., 4, 6144, https://doi.org/10.1038/srep06144.
Wang, S. L., Y. J. Ding, L. Zhao, and K. Toshio, 2002: The influence of local factor on surface layer ground temperature in Qinghai-Xizang Plateau (in Chinese). Plateau Meteor., 21, 85–89.
Wang, X. L., 2008: Penalized maximal F test for detecting undocumented mean shift without trend change. J. Atmos. Oceanic Technol., 25, 368–384, https://doi.org/10.1175/2007JTECHA982.1.
Wang, X. L., and Y. Feng, 2013: RHtestsV4 user manual. Environment Canada Rep., Climate Research Division, Science and Technology Branch, 29 pp., http://etccdi.pacificclimate.org/RHtest/RHtestsV4_UserManual_10Dec2014.pdf.
Wang, Y., W. Chen, J. Zhang, and D. Nath, 2013: Relationship between soil temperature in May over Northwest China and the East Asian summer monsoon precipitation. Acta Meteor. Sin., 27, 716–724, https://doi.org/10.1007/s13351-013-0505-0.
Wang, Y., Z.-Z. Hu, and F. Yan, 2017: Spatiotemporal variations of differences between surface air and ground temperatures in China. J. Geophys. Res. Atmos., 122, 7990–7999, https://doi.org/10.1002/2016JD026110.
WMO, 2017: Challenges in the transition from conventional to automatic meteorological observing networks for long-term climate records. WMO Rep. 1202, 31 pp., https://library.wmo.int/doc_num.php?explnum_id=4217.
Yang, K., and J. Y. Zhang, 2016: Spatiotemporal characteristics of soil temperature memory in China from observation. Theor. Appl. Climatol., 126, 739–749, https://doi.org/10.1007/s00704-015-1613-9.
Yang, K., and J. Y. Zhang, 2018: Evaluation of reanalysis datasets against observational soil temperature data over China. Climate Dyn., 50, 317–337, https://doi.org/10.1007/s00382-017-3610-4.
Zeng, X., Z. Wang, and A. Wang, 2012: Surface skin temperature and the interplay between sensible and ground heat fluxes over arid regions. J. Hydrometeor., 13, 1359–1370, https://doi.org/10.1175/JHM-D-11-0117.1.
Zhang, Y., W. Chen, S. L. Smith, D. W. Riseborough, and J. Cihlar, 2005: Soil temperature in Canada during the twentieth century: Complex responses to atmospheric climate change. J. Geophys. Res., 110, D03112, https://doi.org/10.1029/2004JD004910.
Zhou, C., K. Wang, and Q. Ma, 2017: Evaluation of eight current reanalyses in simulation land surface temperature from 1979 to 2003 in China. J. Climate, 30, 7379–7398, https://doi.org/10.1175/JCLI-D-16-0903.1.
Zhou, L., and R. Huang, 2006: Characteristics of interdecadal variability of the difference between surface temperature and surface air temperature in spring in arid and semi-arid region of Northwest China and its impact on summer precipitation in North China (in Chinese). Climatic Environ. Res., 11, 1–13.
