Sciences in Cold and Arid Regions  2017, 9 (2): 127-141   PDF    

Article Information

HongYan Bao, Kai Yang, ChengHai Wang . 2017.
Characteristics of GLDAS soil-moisture data on the Tibet Plateau
Sciences in Cold and Arid Regions, 9(2): 127-141
http://dx.doi.org/10.3724/SP.J.1226.2017.00127

Article History

Received: September 7, 2016
Accepted: February 13, 2017
Characteristics of GLDAS soil-moisture data on the Tibet Plateau
HongYan Bao, Kai Yang, ChengHai Wang     
Key Laboratory of Arid Climate Change and Disaster Reduction of Gansu Province, College of Atmospheric Sciences, Lanzhou University, Lanzhou, Gansu 730000, China
Abstract: In this paper, the applicability of soil-moisture (SM) datasets of GLDAS (Global Land Data Assimilation System) in an alpine region (Tibet Plateau, TP) is investigated. The relations and discrepancies between the GLDAS-NOAH SM (0~10 cm) and the observations are compared; the possible reasons for errors over the TP are explored. The results show that GLDAS SM biases mainly show up in errors of values in the nonfrozen period (April to October) and changes of SM along with the temperature, especially during the freezing-thawing process in the frozen period (November to March). The biases of GLDAS SM in the nonfrozen period are mainly caused by the GLDAS precipitation-forcing data. The errors of GLDAS SM in the frozen period are speculated to be induced by the freeze-thaw parameterization scheme in the land-surface model.
Key words: soil moisture     GLDAS     Tibet Plateau     error analysis    
1 Introduction

Soil-moisture (SM) plays a crucial role in climate changes. As one of the important physical qualities of the land-surface process, SM is a significant variable used to measure dry and wet climatic changes; it also affects the interactions between the land surface and the atmosphere and the exchanges of heat, moisture, momentum, etc., directly or indirectly, by changing the surface evaporation and physical properties of soil, such as thermal capacity and the growth condition of vegetation (Seneviratne et al., 2010; Wang CH et al., 2008, 2010). Both observations and simulations indicate that SM anomalies can persist from weeks to seasons and affect the subsequent climate. In the early 1950s, Namias (1958, 1963) suggested the seasonal anomalies of SM play important roles in the seasonal changes of atmospheric circulation. Sensitivity experiments showed that dry soil not only is conducive to the rise of atmospheric temperature and the increase of the sensible heat flux but also reduces surface evaporation and is adverse to precipitation in the subsequent period (Walker and Rowntree, 1977). Nevertheless, wet soil is good for continuous precipitation but also works against the rise of temperature in the subsequent period. Using numerical simulations, Rowntree and Bolton (1983) proved that anomalies of SM have significant effects on subsequent precipitation, humidity, and temperature. Yeh et al. (1984) performed numerical experiments to simulate the influences of large-scale irrigation on late evapotranspiration and precipitation; and results suggested that the anomalies of SM caused by irrigation affect not only late evapotranspiration, precipitation, and temperature but also the large-scale circulation system. A study by Chahine (1992) further proposed that SM is the second important parameter in climate changes relative to the sea surface temperature (SST), and its effect is even larger than the role of SST on land. Statistics show that only 35% of the rainfall comes from marine evaporation driven by winds and 65% comes from the land evaporation over land. Wang and Guo (2012) and Guo and Wang (2014) suggested that much atmospheric precipitation comes from water's recycling caused by surface evaporation, which on the TP may be associated with the soil thaw and snowmelt. The changes of SM associated with the thawing and freezing processes have connections with the transition between the dry and wet seasons on the TP (Wang and Shang, 2007). Similarly, Chen SJ et al. (2013) found a negative correlation between SM and air temperature and a positive correlation between SM and precipitation over semi-arid region of the Loess Plateau. Several recent studies (Guo et al., 2007; Yang et al., 2008; Dirmeyer et al., 2009) showed that SM has significant effects on short-term climate change and plays an important role in short-term climate prediction, especially seasonal climate prediction.

Studies have recognized the important role of SM in climate change. However, largely due to spatial–temporal variations of SM, the distribution of observational sites is nonuniform on complex terrain, especially for high-latitude and high-altitude regions, resulting in a lack of temporal–spatial continuity in the SM dataset. It is very difficult to obtain observational SM on a large scale. Remote-sensing observation has the spatial advantage of global coverage, but its accuracy depends on the inversion algorithms; and it is difficult to get the data in deeper soil depths (Ma et al., 2001). Consequently, the accuracy of SM data impairs the land–atmosphere interaction studies based on SM.

As the "roof of the world", the TP has large areas of frozen ground. The diabatic heating caused by phase changes of SM affects interaction between the land and the atmosphere, further affecting weather, climate, and circulation characteristics in Asia (Liu et al., 1989; Zheng and Wang, 2001). Nonetheless, due to the complex conditions of the terrain, SM observations on the TP are scarce and uneven, especially in the western region; for a wide range of spatial and temporal scales, SM data of the TP is lacking.

GLDAS is a global, high-resolution, land-data assimilation system established in recent years, aimed at using satellite- and ground-based observation data products, advanced land-surface models, and data assimilation technology to generate optimal surface conditions and flux data. GLDAS provides relatively reliable data for researching SM. GLDAS drives multiple offline land-surface models (not coupled to the atmosphere), integrates a huge quantity of observation-based data, executes globally at high resolutions (2.5 to 1 km), and is capable of producing results in near real time. GLDAS dataset is widely used in past years, and has better representation. Compared with other reanalysis data, the GLDAS dataset has more accurate driving data and reasonable simulation results (Rodell et al., 2004). Some studies (Zaitchik et al., 2010; Fu and Wang, 2014; Ji et al., 2014) suggested that compared with observation data, the GLDAS data has higher credibility. Some studies (Ghazanfari et al., 2013; Yang and Chen, 2015) have accomplished much by means of GLDAS data. Ghazanfari et al. (2013) proposed a new drought index that can determine drought areas effectively using the GLDAS dataset. Yang and Chen (2015) compared and analyzed terrestrial water-storage variations from GRACE (Gravity Recovery and Climate Experiment) and GLDAS on the Tianshan Mountains and adjacent areas of central Asia, concluding that terrestrial water storage inverted by GRACE and GLDAS shows good consistency, with significant liner relations. Spennemann et al. (2015) compared simulated SM anomalies derived from different versions of the GLDAS data, showing that GLDAS data can capture SM anomalies and variability, taking into account regional and seasonal dependencies and showing correspondence with other proxies used to characterize soil states in South America; thus they supported the use of GLDAS as an indicator of SM states and for developing new SM–monitoring indices. However, other studies (Chen Y et al., 2013; Ji et al., 2014; Park and Choi, 2014) pointed out that GLDAS data still has some problems. Park and Choi (2014) compared the evapotranspiration capacity of GLDAS and observation data in South Korea, suggesting that the variable from GLDAS data is lower although it has similar seasonal changes with observations. Ji et al. (2014) concluded that the GLDAS air-temperature data shows relatively low accuracy in some areas of Africa and South America, although its estimates are generally accurate. They suggested that GLDAS temperature data should be used cautiously in mountainous areas or places where weather stations are sparse.Chen Y et al. (2013) showed that the four GLDAS models tend to systematically underestimate surface SM (0~5 cm) but well simulate SM for the 20~40 cm layer on the central TP.

However, how does the GLDAS SM perform on TP where terrain is complex? What are the main causes for errors of the GLDAS SM? These issues are related to understanding parameters causing the errors of SM and are beneficial to improving the land-surface model. The other sections of this paper are organized as follows. Section 2 presents the data and method used. The errors characteristically noted between observational and GLDAS data are analyzed in section 3. Section 4 investigates the possible causes of the errors. Section 5 evaluates the annual variation characteristic of GLDAS data. The conclusion and summary are in section 6.

2 Data and methods

GLDAS has been developed jointly by the National Aeronautics and Space Administration (NASA) Goddard Space Flight Center (GSFC) and the National Oceanic and Atmospheric Administration (NOAA) National Centers for Environmental Prediction (NCEP). It generates a series of land-surface state (e.g., SM and surface temperature) and flux (e.g., evapotranspiration and sensible heat flux) products simulated by four land-surface models (CLM, Mosaic, Noah, and VIC) (Rodell et al., 2004). These high-quality global land-surface data are widely used in the study of weather and climate. Some studies (Wang W et al., 2014; Wang XJ et al., 2014; Liu et al., 2015) showed that the SM data of GLDAS–NOAH have good quality and are close to the observed values. And the observation data we selected have some restrictions. Thus, in this study, in consideration of temporal consistency, we use the SM (0~10 cm), surface temperature, and precipitation data of GLDAS–NOAH version 1; spatial resolution is 1°×1°, and temporal resolution is 3-hourly. In this study, the term GLDAS is used to refer to GLDAS–NOAH. For GLDAS, soil-moisture is the sum of liquid water and ice (Chen Y et al., 2013; Liu et al., 2015), whereas observed soil-moisture includes only liquid water. Thus we use statistically experiential formula about the relation between soil liquid water and soil temperature (Xu et al., 2001; Wang and Yu, 2009) to calculate indirectly the corresponding portion of soil liquid water of the GLDAS data and then compare those figures with the observations. The computation formula is as follows:

$\textit{θ} = \textit{α} {T ^b}$ ((1))

where θ means SM (unit, mm/mm), T means natural index of the surface temperature below 0 °C (unit, °C), α and b are the constants related to the soil natures, respectively, which are obtained by fitting relations of observed soil-moisture and surface temperature.

The observation data we collected from the TP comprehensive observation stations include SM (0~10 cm) and surface temperature, with a half-hour time resolution. Daily precipitation data came from the China Meteorological Data Service Center (http://data.cma.cn/en). Four observation locations are shown in Figure 1. Maqu and Naqu are located in the Plateau's cold and humid region and their soil is composed mainly of sand and clay. Maduo and Ali are located in the droughty (even desert) area, where the soil is composed of mainly sand. The observational period is 2010/01/01 (Year/Month/Day) to 2010/12/31 at Maqu, 2011/01/01 to 2011/12/31 at Naqu and Ali, and 2013/07/07 to 2014/05/20 at Maduo.

Figure 1 The location and altitude of observation stations

According to the duration of observation data collection at the four stations, we chose the corresponding period for GLDAS–NOAH to compare with the observations. By comparing the results from the bilinear interpolation method and the adjacent-grid match method for GLDAS data, we found that the values calculated by the adjacent-grid match method are closer to the values obtained by observation. The amount of precipitation is the sum of rainfall and snowfall. The SM is the soil-water content from 0 to 10 cm; the correction factor of 100 mm is used to eliminate the difference of units between GLDAS data and observation data.

The mean bias $\bar d$ is used to analyze the errors of the SM between GLDAS data and observations as follows:

$\bar d = \frac{1}{N}\sum\limits_{i = 1}^N {\left| {{G_i}} \right. - \left. {{O_i}} \right|} $ ((2))

where G i is the GLDAS variable (SM, precipitation, or surface temperature), O i is the corresponding observation variable, and N is the number of days.

To estimate quantitatively the relation between the observation and the GLDAS data, the linear correlation coefficient is used. The calculation method for the correlation coefficient is as follows:

$R = \frac{{ \displaystyle \frac{1}{N}\sum\limits_{i = 1}^N {({X_i} - \overline X )({Y_i} - \overline Y )} }}{{\sqrt { \displaystyle \frac {1}{N}\sum\limits_{i = 1}^N {{{({X_i} - \overline X )}^2}} } \sqrt {\displaystyle \frac{1}{N}\sum\limits_{i = 1}^N {{{({Y_i} - \overline Y )}^2}} } }}$ ((3))

where X i is the GLDAS variable (SM, precipitation, or surface temperature), Y i is the corresponding observation variable, $\overline X = \displaystyle \frac{1}{N}\sum\limits_{i = 1}^N {{X_i}} $ , $\overline Y = \displaystyle \frac{1}{N}\sum\limits_{i = 1}^N {{Y_i}} $ are the mean values of them, and N is the number of days.

Linear regression analysis is used to analyze quantitatively the possible sources of errors of GLDAS SM figures. The relative contributions (Huang and Yi, 1991) are as follows:

$\mathop R\nolimits_j = \frac{1}{m}\sum\limits_{i = 1}^m {\left[ {\mathop T\nolimits_{ij}^2 /(\sum\limits_{j = 1}^a {\mathop T\nolimits_{ij}^2 } )} \right]} $ ((4))

where T ij =b jx ij is the items in the regression equation, b j is the multiple regression coefficient, m is the length of the data sequence, and a is the number of variables in the regression equation.

3 The biases and its characteristics of GLDAS

Large amounts of permafrost and seasonal frozen soil are distributed over the TP, and the freezing–thawing process of frozen soil is one of the most prominent physical characteristics of the land surface on the TP (Wang et al., 2001; Chen, 2014). Owing to the different elements affecting SM, the changes of SM in the frozen period are distinguishable from its state in the nonfrozen periods (Guo et al., 2002). Therefore, to avoid the influences of random weather processes on the shift of the soil freezing and thawing phase, for analysis we divided the SM into the frozen period (about January to March and November to December); and the nonfrozen period (about April to October), which is defined as the days during which the surface temperature is higher than 0 °C for 10 consecutive days.

Figure 2 shows the correlation between GLDAS SM and observations at the four stations (Maqu, Naqu, Ali and Maduo) in frozen and nonfrozen periods; on the whole, the GLDAS SM positively correlates with the observations significantly; the correlation coefficients are all past the 95% significance t-test in the nonfrozen and frozen periods. The correlation coefficients in the nonfrozen period are bigger than those in the frozen period at all stations except Ali, where the correlation coefficients in the nonfrozen period are lower than those for the frozen period. In the nonfrozen period, GLDAS SM is obviously lower than observations at Maqu and Maduo, stations located north of the Tanggula Mountains on the TP, where their figures are easily influenced by winter monsoons (An et al., 2001; Liu, 2006; Wang et al., 2015), and significantly larger than observations at Naqu and Ali in the nonfrozen period. Naqu and Ali are located on the southern TP, where their figures are affected by summer monsoons (He et al., 1987; Gong et al., 2004; Zhang et al., 2008), the warm and wet air within close proximity in the nonfrozen period. However, in the frozen period, GLDAS SM is lower than observations at all the stations except for Maqu. Thus GLDAS SM has certain errors in both the frozen period and the nonfrozen, though GLDAS SM has better correlations with observations on the whole in the nonfrozen period.

Figure 2 Scatter diagram of observations versus GLDAS soil-moisture in the nonfrozen period (left column) and the frozen period (right column) for (a, e) Maqu; (b, f) Maduo; (c, g) Naqu; and (d, h) Ali (R is the correlation coefficient between observation and GLDAS soil-moisture; * means the correlation coefficient is significant at the 95% level by t-test)

To analyze quantitatively the error properties of GLDAS SM, the mean bias of GLDAS SM with observations in frozen and nonfrozen periods at the four stations is calculated in Table 1. The results show that the mean bias of GLDAS SM in the nonfrozen period is larger than in the frozen period at the four stations, though GLDAS SM has better correlations with observations in the nonfrozen period. Combined with the analysis of Figure 2, we find that GLDAS SM has errors in both the frozen and the nonfrozen periods, though it has better correlations and larger errors with observations in the nonfrozen period than in the frozen period.

Table 1 The mean bias of GLDAS soil-moisture with observations (unit: mm/mm)
Type Maqu Maduo Naqu Ali Average
Nonfrozen 0.058 0.102 0.050 0.075 0.071
Frozen 0.019 0.033 0.022 0.043 0.029
Note: The rightmost column gives the average of the mean bias for the four sites.
4 The possible causes for biases in GLDAS soil-moisture

For a given region, the SM is affected not only by soil characteristics such as topography (slope direction, gradient, etc.) and the land-use types but also by precipitation, temperature, evapotranspiration, solar radiation, wind speed, runoff, and many other factors (Zhang et al., 2011; Li Y et al., 2012; Chen SJ et al., 2013). Compared to other factors, precipitation and temperature are the most important affecting changes of SM. Precipitation is the direct and important source of SM, meaning SM increases with the increase of precipitation (Yeh et al., 1984; Chahine, 1992; Bengtsson, 2010). Surface temperature affects SM by impacts on the freezing–thawing processes and the evapotranspiration from the soil surface. On the one hand, soil thaw increases the soil-water content and soil freeze decreases it; on the other hand, the high temperature of the soil surface can enhance surface evapotranspiration, which decreases the SM (Guo et al., 2007; Wang et al., 2009; Chen SJ et al., 2013). Thus, the forcing dataset (precipitation and surface temperature) for driving the land-surface model is crucial and the key factor for GLDAS.

To evaluate the quality of primary forcing for GLDAS, the relations between GLDAS precipitation and surface temperature and the corresponding observational data are analyzed.

4.1 Relations of precipitation and surface temperature between GLDAS and observation

Figure 3 shows the correlation of GLDAS precipitation and observations in frozen and nonfrozen periods at the four sites. GLDAS precipitation has a significantly positive relation with the observations, passing the 95% significance test. Except for the Maqu station data, the correlation coefficients in the frozen period are larger than in the nonfrozen period, which implies that GLDAS precipitation data has better relations with observations in the frozen period. Nevertheless, GLDAS precipitation has some differences with the observations, with a positive bias at Maqu and Ali and a negative bias at Maduo and Naqu. As the observation precipitation is very little or none, GLDAS still shows more or little precipitation,i.e., GLDAS overestimates precipitation, especially in the nonfrozen period. Consequently, the error for precipitation may be one of the reasons causing the errors of SM, particularly in the nonfrozen period.

Figure 3 Scatter diagram of observations versus GLDAS precipitation in the nonfrozen period (left column) and the frozen period (right column) for (a, e) Maqu; (b, f) Maduo; (c, g) Naqu; and (d, h) Ali (R is the correlation coefficient between observation and GLDAS precipitation; * means the correlation coefficient is significant at the 95% level by t-test)

The TP located in the low-middle-latitude region is the area of the most widespread and thickest frozen ground and lower temperature, an area for which the frozen ground has unique seasonal changes (Wang et al., 2001). The seasonal frozen layer is sensitive to changes of soil temperature. The surface temperature influences the soil-water content through the process of freezing–thawing in winter and surface ET in the nonfrozen period (Guo et al., 2007; Wang et al., 2009; Chen SJ et al., 2013). Figure 4 shows the relations between GLDAS and observational surface temperature. It shows that GLDAS surface temperature has a significantly positive correlation with observations, and the correlation coefficients pass the 95% significance test in both the nonfrozen period and the frozen at the four sites. The correlation coefficients for the nonfrozen period are larger than those for the frozen period, which implies the bias of surface temperature between GLDAS and observation is smaller in the nonfrozen period. Except for Ali, where GLDAS surface temperature is lower than observations, GLDAS surface temperature is close to the observations in the nonfrozen period. However, in the frozen period—except for Maqu, where GLDAS surface temperature has good consistency with observations—GLDAS surface temperature is much lower than the observations and has larger errors. The errors between GLDAS surface temperature and observations in the nonfrozen period are much less than in the frozen period. The errors of GLDAS surface temperature in the frozen period may be the main reason for the SM errors.

Figure 4 Scatter diagram of observations versus GLDAS surface temperature in the nonfrozen period (left column) and the frozen period (right column) for (a, e) Maqu; (b, f) Maduo; (c, g) Naqu; and (d, h) Ali (R is the correlation coefficient between observation and GLDAS precipitation; * means the correlation coefficient is significant at the 95% level by t-test)

The mean bias of GLDAS precipitation and surface temperature is calculated in Table 2. It shows that the mean bias of precipitation in the nonfrozen period is much larger than in the frozen periods, and the mean bias of surface temperature in the frozen period is much larger than in the nonfrozen period at the four stations, which is consistent with the results of the correlation analysis. The differences of the mean bias of surface temperature are not large in the frozen and the nonfrozen periods at Maqu and Ali. It should be noted that Ali is located in a desert area where precipitation is scarce, and SM changes are more dependent on the freezing and thawing process of soil. Therefore, the bias of surface temperature also is the dominant factor in the SM errors for Ali in the nonfrozen period.

Table 2 The mean bias of GLDAS precipitation and surface temperature with corresponding observations
Type Maqu Maduo Naqu Ali Average
Nonfrozen PRE (mm) 1.83 1.80 1.99 1.88 1.89
TEMP (°C) 1.25 1.83 3.54 9.19 3.95
Frozen PRE (mm) 0.10 0.23 0.10 1.77 0.55
TEMP (°C) 1.47 8.67 4.74 9.50 6.10
Note: PRE is short for precipitation, and TEMP is short for surface temperature. The rightmost column gives the average of mean bias for the four sites.
4.2 The relation of GLDAS soil-moisture with precipitation and surface temperature

To analyze further the causes of the bias of GLDAS SM, the relations between SM and precipitation of GLDAS and observations in the nonfrozen (Figure 5) and the frozen periods are analyzed, respectively. Figure 5 shows that the observation SM has a positive correlation with precipitation at the four sites in the nonfrozen period. Except for Ali's, the correlation coefficients pass the 95% significance t-test. For GLDAS data, the SM has a significantly positive correlation with precipitation at the four stations, and the correlation coefficients are all larger than for observations. The changes of GLDAS SM along with its precipitation show some differences from observations. With the emergence of the observed precipitation, the changes of observation SM are very slight, even changeless at first. With the increase of observed precipitation, observation SM shows slow increase, even invariability. But the forcing of GLDAS precipitation on SM is very obvious. GLDAS SM increases quickly, with its precipitation appearing and increasing, which is different than the observations. In the frozen period, GLDAS and observation precipitation are little; and the correlations between their SM and precipitation are not important (figure omitted). There are some errors of the precipitation-simulation data and its forcing effects in SM in the GLDAS–NOAH model, which are more obvious in the nonfrozen period. The errors of precipitation-simulation data could be the main cause of errors of GLDAS SM in the nonfrozen period.

Figure 5 Scatter diagram showing precipitation versus soil-moisture from observation (left column) and GLDAS (right column) in the nonfrozen period; (a, e) Maqu; (b, f) Maduo; (c, g) Naqu; and (d, h) Ali (R is the correlation coefficient between precipitation and soil-moisture; * indicates the correlation coefficient is significant at 95% level by t-test)

The observed SM generally is the liquid water in the soil. One of the significant characteristics is the phase change occurring at surface temperature at about 0 °C. When the surface temperature is just below 0 °C or even lower, the surface soil begins to freeze, which can decrease the observed SM of the surface soil. On the contrary, the surface soil begins to thaw when the surface temperature is just above 0 °C or even higher, and the surface SM increases.

Figure 6 shows the correlations between SM and temperature in the frozen period for observation and GLDAS. The relation between surface temperature and SM from observation and GLDAS likely is an exponentiation function, as Formula (1) shows. SM positively correlates with surface temperature in both GLDAS and observations at the four sites in the frozen periods, and the correlation coefficients all pass the 95% significance t-test. The observation SM changes little when the surface temperature is below 0 °C and dramatically increases near 0 °C, which implies the thawing of soil. For GLDAS data, the positive correlations between SM and surface temperature are more significant than those of the observations, and the correlation coefficients are 0.980 and 0.997. SM increases quickly with the increase in surface temperature, although surface temperature is far below 0 °C. This pattern implies that the freezing–thawing parameterization scheme in GLDAS–NOAH is speculated to be problematic; this pattern also is a disadvantage for most of the land-surface models.

Figure 6 Scatter diagram showing surface temperature versus soil-moisture for observation (left column) and for GLDAS (right column) in the frozen period; (a, e) Maqu; (b, f) Maduo; (c, g) Naqu; and (d, h) Ali (R is the correlation coefficient between surface temperature and soil-moisture; * indicates the correlation coefficient is significant at the 95% level by t-test)

In the nonfrozen period, with surface temperature rising, variation of observational SM shows a slowly increasing and then a decreasing trend; the amplitude of the variation is small. The variation of GLDAS SM is similar to that of the observations. That is to say, the SM in the shallow layer is also affected by surface evapotranspiration. When the surface temperature rises, surface evapotranspiration is enhanced, thus the shallow-layer SM decreases.

Precipitation and surface temperature have different effects on the SM in different periods; similarly, the main causes of GLDAS SM biases also change. To analyze quantitatively the influences of GLDAS precipitation and surface temperature on SM errors, the relative contribution rates of GLDAS and observation precipitation and surface temperature on their SM are calculated, respectively (Table 3), combined with the multiple regression analysis in the frozen and nonfrozen periods. In Table 3, for observation data in the nonfrozen period, SM is mainly influenced by precipitation at Maqu and Naqu and by temperature at Maduo and Ali. In the frozen period, SM is mainly influenced by temperature and almost unaffected by precipitation at the four sites. Actually, Maqu and Naqu are located in the Plateau's cold and humid region, where the increases of soil-water mainly rely on precipitation in the nonfrozen period and the thaw of the soil in the frozen period. Maduo and Ali are located in the droughty (even desert) area, where precipitation is scant, and the increases of SM mainly rely on the thaw of the soil. For GLDAS data, the relative contribution rates can roughly reflect the facts. But the relative contribution rates of precipitation are larger than those of observations, and the relative contribution rates of surface temperature are lower than observations at Maduo, Naqu, and Ali in the nonfrozen period, with Maqu showing the opposite. GLDAS SM is mainly influenced by precipitation at Maduo in nonfrozen period. In frozen period, the relative contribution rate of surface temperature is larger than found with observations, except at Maduo. The differences mean that GLDAS overestimates the effect of precipitation on SM in the nonfrozen period and the effect of surface temperature on SM in the frozen period, which give further support to the inference that the errors for precipitation may be one of the reasons causing the errors of SM, particularly in the nonfrozen period and the errors of GLDAS surface temperature in the frozen period may be the main reason for the SM errors.

Table 3 The relative contribution ratio of precipitation and surface temperature on soil-moisture at different times (unit: %)
Type Maqu Maduo Naqu Ali
OBS GLDAS OBS GLDAS OBS GLDAS OBS GLDAS
Nonfrozen PRE 92.39 75.96 25.21 63.39 71.16 96.23 1.30 8.35
TEMP 7.61 24.04 74.79 36.61 28.84 3.77 98.70 91.65
Frozen PRE 3.73 0.76 0.01 1.60 1.26 0.23 1.03 0.29
TEMP 96.27 99.24 99.99 98.40 98.74 99.72 98.96 99.71
Note: PRE is short for precipitation, and TEMP is short for surface temperature.
5 Annual evaluation of soil-moisture, precipitation, and temperature

Koster et al. (2002) through simulation experiments of the land–atmosphere interaction, showed that the positive anomaly of precipitation can increase SM; and the wet soil can increase the surface evapotranspiration, which in turn can influence precipitation through the local water cycle or changes in large-scale circulation. By studying the effects of water storage in the atmosphere on precipitation cycles, Dominguez et al. (2006) suggested that SM plays an important role in maintaining subsequent precipitation. Through research on climatic responses in the west of Northwest China, Wang L et al. (2008) proposed there are responses between SM and climate, as well as widespread negative correlations between SM and temperature. Some studies (Shukla and Mintz, 1982; Ma et al., 2015) also suggested that the SM anomalies have significant influences on the subsequent short-term climate, mainly on precipitation and temperature.

Figure 7 shows the evaluation of shallow-layer (0~10 cm) SM, precipitation, and surface temperature from GLDAS and from observation over time. For observation data of Maqu (Figure 7a), the changes in SM are not obvious in January to February or in December due to the frozen soil and scarce precipitation. In early or mid-March, as surface temperature rises above 0 °C, SM rapidly increases, meaning the ground begins to thaw. In April to May after the thawing of soil, with the increase of surface temperature, a series of precipitation events occur, increasing in intensity and frequency, which may reflect the influence of SM on subsequent precipitation. In June to July, with the arrival of the rainy season, SM increases with an increase in precipitation and decreases with a decrease of precipitation. In early August, the surface temperature rises to the maximum for the year, after which, precipitation begins to reduce dramatically, which may be due to enhancement of surface evaporation and decrease of SM in early August, and a similar situation appears in mid-September. From mid-August to early November, due to the rapid reduction of precipitation, the effects of precipitation on SM become weak; and the effects of surface temperature on it enhances. Therefore, SM increases (decreases) with a decrease (increases) of surface temperature. In mid- to late November, as the surface temperature drops below 0 °C, SM significantly reduces, meaning the soil begins to freeze. For GLDAS data of Maqu ( Figure 7a), in the frozen period (November to March), SM changes with the fluctuations in surface temperature and is higher than that found by observation; and these changes don't well reflect the thawing process of soil near the surface temperature of 0 °C. At the same time, the errors of surface temperature are large. In the nonfrozen period (April to October), the change trend of SM is consistent with observations; and the value is low. The surface temperature is relatively close to that of observations; and due to greater precipitation, SM mainly changes along with the changes in precipitation. As for the observation data of Maduo ( Figure 7b), the soil began to freeze in mid- to late October 2013, and the SM changed little in the frozen period (November 2013 to March 2014). In early April 2014, the soil began to thaw. In May 2014, precipitation began to increase. The changes of SM were mainly affected by precipitation in the wet season (June to August) and by surface temperature in the dry season (September to mid- to late October), which is consistent with the situation at Maqu. For GLDAS data of Maduo (Figure 7b), in the frozen period (November 2013 to March 2014), the surface temperature is obviously low, and with error rising to the highest level; SM still has small fluctuations with the changes of surface temperature and doesn't reflect well the thawing process of the soil. In the nonfrozen period (July to October 2013 and April to May 2014), precipitation and SM were obviously low. At Naqu (Figure 7c), the observations are similar to those for Maqu and Maduo. For GLDAS data for Naqu (Figure 7c), in the nonfrozen period (May to October), the change trend of SM is consistent with observations though SM is larger than found by observation and precipitation is less. In the frozen period (November to March), GLDAS data is consistent for Maduo observations. At Ali (Figure 7d), because it is located in a semi-arid (even desert) area and its precipitation is slight, observed SM shows an obvious increase in the period (early April) when soil begins to thaw and shows a decrease in the period (November) when the soil begins to freeze. In other periods, SM changes little; however, the GLDA SM has big errors: On the one hand, these may be caused by the inaccuracy of its precipitation-simulation data. On the other hand, they may be associated with errors of its surface temperature, which is consistent with the results above.

Figure 7 The annual variation of soil-moisture (mm/mm), precipitation (mm), and surface temperature (°C) for (a) Maqu, (b) Maduo, (c) Naqu, and (d) Ali

To sum up, the changes in SM on the TP are affected significantly by the freezing and thawing process of soil. SM changes little in the frozen period and significantly increases during the process of soil thawing, which is opposite in the process of soil freezing. Plateau precipitation and surface temperature also have significant effects on SM, and the effects change in different periods. In the wet season (about June to August), SM is mainly affected by precipitation and increases with the increase in precipitation. In the dry season (about September to October), because of the reduction in precipitation, SM is mainly affected by the surface temperature and decreases with the increase of the surface temperature. In turn, SM also has effects on subsequent precipitation, mainly on the increase in precipitation after the soil thaws, with an increase of SM; once the soil is frozen, this effect doesn't occur. In the frozen period, GLDAS SM changes with the fluctuations of surface temperature and doesn't well simulate the freezing and thawing processes, a result mainly caused by the faultiness of the freezing–thawing parameterization scheme of the GLDAS–NOAH model. In the nonfrozen period, the change trend of GLDAS SM is consistent with observations, though there are some errors in its values, mainly caused by the inaccuracy of the precipitation-simulation data. At the same time, at Ali in the desert area, the contribution of the errors of GLDAS precipitation and surface temperature to the errors of its SM are obvious in both frozen and nonfrozen periods.

6 Conclusion and discussion

GLDAS as a widely used SM dataset provides a tool for understanding spatial–temporal characteristics of SM on a global scale. However, in regions of complex terrain, the performance of GLDAS SM remains largely uncertain. This study investigated GLDAS characteristics based on four observational sites over the TP, sites located in regions with different climate regimes. The results show that, over the TP, GLDAS SM has errors both in the frozen period and the nonfrozen period, though it has better correlations and larger errors as compared with observations in the nonfrozen period than in the frozen period, which implies GLDAS SM data has certain a representativeness in complex terrain. The SM biases mainly show up in the errors of values in the nonfrozen period and the changes of SM along with the temperature, especially in the freezing–thawing process in the frozen period.

The biases of a forcing dataset such as precipitation and surface temperature can affect the property of GLDAS SM. Basically, GLDAS precipitation and surface temperature are consistent with the corresponding observational data, respectively. However, the correlation of precipitation between GLDAS and observations in the frozen period is better than in the nonfrozen period over the TP. The relation of surface temperature between GLDAS and observations in the nonfrozen period is better than in the frozen period. In the nonfrozen period, SM is mainly affected by precipitation in the wet season (e.g., summer) and by surface temperature in the dry season (e.g., autumn). The SM variation is dominated by precipitation and surface temperature from atmospheric forcing (Ma et al., 2001; Xu et al., 2001). The main reason for GLDAS SM errors in the nonfrozen period is the bias from GLDAS precipitation.

Soil-moisture increases rapidly during the thawing process and decreases rapidly during the freezing process. However, the variations of GLDAS SM can't well describe this basic feature, especially the thawing process. The bias of GLDAS SM in the frozen period reveals the faultiness of the freezing–thawing parameterization scheme in the GLDAS–NOAH model, particularly for the process of phase change (i.e., around the phase point), simulation of SM is almost inadequate.

Due to the limited observation data and the limitations of the GLDAS–NOAH data, as well as the nearly incomprehensible complexity of the land surface (soil properties, altitude,etc.), hence, the analysis of the errors and climate effects of GLDAS–NOAH SM has some limitations. Nonetheless, the GLDAS soil-moisture dataset is accurate and reasonable assimilation data. Over TP, the observation stations of SM are scarce and the temporal scales of observation data are short, which limit the study of SM on the TP. By using the BATS (biosphere atmosphere transfer scheme) land-surface model in RegCM3, Yu and Wang (2012) amended the SM figures over the TP; and the results of the amended SM are close to those of observations.

Acknowledgments:

This work is supported by the National Science Foundation of China (Nos. 91437217, 41275061, 41471034, and 41661144017) and the China National Basic Research Program (2013CBA01800). We thank the China Meteorological Data Service Center (CMDC) (http://data.cma.cn/en) and the International Soil Moisture Network (https://ismn.geo.tuwien.ac.at) for providing partial observation data. The GLDAS data used in this study were archived and distributed by the Goddard Earth Sciences (GES) Data and Information Services Center (DISC) (http://disc.sci.gsfc. nasa.gov/hydrology/data-holdings).

Reference
An ZS, Kutzbach JE, Prell WL, et al, 2001. Evolution of Asian monsoons and phased uplift of the Himalaya-Tibetan Plateau since Late Miocene times. Nature, 411(6833): 62–66. DOI: 10.1038/35075035
Bengtsson L, 2010. The global atmospheric water cycle. Environmental Research Letters, 5(2): 025202. DOI: 10.1088/1748-9326/5/2/025002
Chahine MT, 1992. The hydrological cycle and its influence on climate. Nature, 359(6394): 373–380. DOI: 10.1038/359373a0
Chen BL, 2014. A study of land surface energy and water in soil freezing and thawing process and impact on regional climate of the TP. Cold and Arid Regions Environmental and Engineering Research Institute, Chinese Academy of Sciences. (in Chinese)
Chen SJ, Guan XD, Huang JP, et al, 2013. Analysis of response of soil moisture to climate change in semi-arid Loess Plateau in China based on GLDAS data. Journal of Arid Meteorology, 31(4): 641–649.
Chen Y, Yang K, Qin J, et al, 2013. Evaluation of AMSR-E retrievals and GLDAS simulations against observations of a soil moisture network on the central Tibetan Plateau. Journal of Geophysical Research: Atmospheres, 118(10): 4466–4475. DOI: 10.1002/jgrd.50301
Dirmeyer PA, Schlosser CA, Brubaker KL, 2009. Precipitation, recycling, and land memory: An integrated analysis. Journal of Hydrometeorology, 10(1): 278–288. DOI: 10.1175/2008JHM1016.1
Dominguez F, Kumar P, Liang XZ, et al, 2006. Impact of atmospheric moisture storage on precipitation recycling. Journal of Climate, 19(8): 1513–1530. DOI: 10.1175/JCLI3691.1
Fu X, Wang B, 2014. Reliability evaluation of soil moisture and land surface temperature simulated by Global Land Data Assimilation System (GLDAS) using AMSR-E data. In: SPIE Asia Pacific Remote Sensing. International Society for Optics and Photonics, 92650O-92650O-5.
Ghazanfari S, Pande S, Hashemy M, et al, 2013. Diagnosis of GLDAS LSM based aridity index and dryland identification. Journal of Environmental Management, 119: 162–172. DOI: 10.1016/j.jenvman.2013.01.040
Gong YF, Ji LR, Duan TY, 2004. Precipitation character of rainy season of Qinghai-Xizang Plateau and onset over East Asia Monsoon. Plateau Meteorology, 23(3): 313–322.
Guo WD, Ma ZG, Wang HJ, 2007. Soil moisture—An important factor of seasonal precipitation prediction and its application. Climatic and Environmental Research, 12(1): 20–28.
Guo YP, Wang CH, 2014. Trends in precipitation recycling over the Qinghai-Xizang Plateau in last decades. Journal of Hydrology, 517(2): 826–835. DOI: 10.1016/j.jhydrol.2014.06.006
Guo ZR, Xing EC, Nie ZL, et al, 2002. Analysis on the characteristics of soil moisture transfer during freezing and thawing period. Advances in Water Science, 13(3): 298–302.
He H, Mcginnis JW, Song Z, et al, 1987. Onset of the Asian summer monsoon in 1979 and the effect of the Tibetan Plateau. Monthly Weather Review, 115(9): 1966–1995. DOI: 10.1175/1520-0493(1987)115<1966:OOTASM>2.0.CO;2
Huang JP, Yi YH, 1991. Inversion of a nonlinear dynamical model from the observation. Science in China (Series B), 34(10): 1246–1246.
Ji L, Senay GB, Verdin JP, 2014. Evaluation of the Global Land Data Assimilation System (GLDAS) air temperature data products. AGU Fall Meeting Abstracts, 16(6): 150731131106004. DOI: 10.1175/JHM-D-14-0230.1
Koster RD, Dirmeyer PA, Hahmann AN, et al, 2002. Comparing the degree of land-atmosphere interaction in four atmospheric general circulation models. Journal of Hydrometeorology, 3(3): 363–375. DOI: 10.1175/1525-7541(2002)003<0363:CTDOLA>2.0.CO;2
Li Y, Liu X, Li WP, 2012. Numerical simulation of land surface process at different underlying surfaces in Tibetan Plateau. Plateau Meteorology, 31(3): 581–591.
Liu C, Yu Y, Xie J, et al, 2015. Applicability of soil temperature and moisture in several datasets over Qinghai-Xizang Plateau. Plateau Meteorology, 34: 653–665.
Liu QC, 2006. The relationship between East Asia winter monsoon and the temperature and precipitation over Tibet Plateau. Qinghai Meteorology, 3: 2–5.
Liu XD, Luo SW, Qian YF, 1989. Numerical simulations of influences of different thermal characteristics on ground surface of Tibet Plateau on the over SE-Aisia. Plateau Meteorology, 8(3): 205–216.
Ma ZG, Fu ZB, Xie L, et al, 2001. Some problems the study on the relationship between soil moisture and climatic change. Advance in Earth Sciences, 16(4): 563–566.
Ma ZG, Wei HL, Fu ZB, 2015. Progress in the research in the relationship between soil moisture and climate change. Advance in Earth Sciences, 14(3): 88–94.
Namias J, 1958. Persistence of mid-tropospheric circulation between adjacent months and seasons. Rossby Memorial Volume, Rockefeller Institute Press and Oxford University Press, pp. 240–248.
Namias J, 1963. Surface-atmosphere interactions as fundamental causes of droughts and other climatic fluctuations. Arid Zone Research, 20: 345–359.
Park J, Choi M, 2014. Estimation of evapotranspiration from ground-based meteorological data and global land data assimilation system (GLDAS). Stochastic Environmental Research and Risk Assessment, 29(8): 1963–1992. DOI: 10.1007/s00477-014-1004-2
Rodell M, Houser PR, Jambor U, et al, 2004. The global land data assimilation system. Bulletin of the American Meteorological Society, 85(3): 381–394. DOI: 10.1175/BAMS-85-3-381
Rowntree PR, Bolton JA, 1983. Simulation of the atmospheric response to soil moisture anomalies over Europe. Quarterly Journal of the Royal Meteorological Society, 109(461): 501–526. DOI: 10.1002/(ISSN)1477-870X
Seneviratne SI, Corti T, Davin EL, et al, 2010. Investigatingsoil moisture-climate interactions in a changing climate: A review. Earth-Science Reviews, 99(3–4): 125–161. DOI: 10.1016/j.earscirev.2010.02.004
Shukla J, Mintz Y, 1982. Influence of land-surface evapotranspiration on the earth's climate. Science, 215(4539): 1498–1501. DOI: 10.1126/science.215.4539.1498
Spennemann PC, Rivera JA, Saulo AC, et al, 2015. A comparison of GLDAS soil moisture anomalies against standardized precipitation index and multisatellite estimations over South America. Journal of Hydrometeorology, 16(1): 158–171. DOI: 10.1175/JHM-D-13-0190.1
Walker J, Rowntree PR, 1977. The effect of soil moisture on circulation and rainfall in a tropical model. Quarterly Journal of the Royal Meteorological Society, 103(435): 29–46. DOI: 10.1002/(ISSN)1477-870X
Wang CH, Dong WJ, Wei ZG, 2001. The feature of seasonal frozen soil in Qinghai-Tibet Plateau. Acta Geographica Sinica, 56(5): 523–531.
Wang CH, Guo YP, 2012. Precipitable water conversion rates over the Qinghai‐Xizang (Tibet) Plateau: changing characteristics with global warming. Hydrological Processes, 26(10): 1509–1516. DOI: 10.1002/hyp.8246
Wang CH, Huang B, Pan BT, 2010. Simulating and analyzing energy balance in Qilian Mountains. Journal of Glaciology and Geocryology, 32(1): 78–82.
Wang CH, Li Y, Wang Y, 2015. Impacts of interdecadal variability of circulation and winter monsoon on winter snowfall over the Tibetan Plateau. Climatic and Environmental Research, 20(4): 421–432.
Wang CH, Shang DC, 2007. Effect of the variation of the soil temperature and moisture in the transition from dry season to wet season over northern Tibet Pleteau. Plateau Meteorology, 26(4): 677–685.
Wang CH, Shi R, Zuo HC, 2008. Analysis on simulation of characteristic of land surface in western Qinghai-Xizang Plateau during frozen and thawing. Plateau Meteorology, 27(2): 239–248.
Wang L, Wen J, Wei ZG, et al, 2008. Soil moisture over the west of northwest China and its response to climate. Plateau Meteorology, 27(6): 1257–1266.
Wang R, Li WP, Liu X, et al, 2009. Simulation of the impacts of spring soil moisture over the Tibetan Plateau on summer precipitation in China. Plateau Meteorology, 28(6): 1233–1241.
Wang W, Wang XJ, Wang P, 2014. Assessing the applicability of GLDAS monthly precipitation data in China. Advances in Water Science, 25(6): 769–778.
Wang XJ, Wang W, Wang P, 2014. Assessing the applicability of GLDAS monthly temperature data in China. Water Resources and Power, 11: 003.
Xu XZ, Wang JC, Zhang LX, 2001. Physics of Frozen Soils. Beijing Publishing House. (in Chinese)
Yang K, Wang CH, Bao HY, 2016. Contribution of soil moisture variability to summer precipitation in the Northern Hemisphere. Journal of Geophysical Research: Atmosphere, 121(12): 108–124. DOI:10.1002/2016JD025644.
Yang P, Chen YN, 2015. An analysis of terrestrial water storage variations from GRACE and GLDAS: The Tianshan Mountains and its adjacent areas, central Asia. Quaternary International, 358: 106–112. DOI: 10.1016/j.quaint.2014.09.077
Yang SF, Liu ZH, Yan Y, et al, 2008. Preliminary study on soil humidity and its relationships with soil temperature and air temperature in snow melting season. Arid Zone Research, 25(5): 642–646.
Yeh TC, Wetherald RT, Manabe S, 1984. The effect of soil moisture on the short-term climate and hydrology change-A numerical experiment. Monthly Weather Review, 112(3): 474–490. DOI: 10.1175/1520-0493(1984)112<0474:TEOSMO>2.0.CO;2
Yu L, Wang CH, 2012. The simulation and revision of sil moisture over Tibetan Plateau. Acta Physica Sinica, 61(2): 531–538. DOI: 10.1109/ICECC.2011.6066563
Zaitchik BF, Rodell M, Olivera F, 2010. Evaluation of the Global Land Data Assimilation System using global river discharge data and a source-to-sink routing scheme. Water Resources Research, 46(6): 2840–2849. DOI: 10.1029/2009WR007811
Zhang J, Li DL, Wang W, 2008. Influence of terrain on precipitation in Qinghai-Tibet Plateau during summer monsoon. Geographic Science, 28(2): 235–240.
Zhang YF, Wang XP, Pan YX, et al, 2011. The dependence of surface albedo on soil moisture in an arid desert area. Journal of Desert Research, 31(5): 1141–1148.
Zheng QL, Wang SS, 2001. Numerical study of the effects of dynamic and thermodynamic of Qinghai-Xizang Plateau on Tropical Atmopheric circulation in summer. Plateau Meteorology, 20(1): 14–21.