1 Introduction

Drought is a creeping recurrent natural hazard that adversely influences agricultural productivity, human livelihoods, and ecosystems (Hao and Singh 2015; Chen et al. 2019). Generally, drought may last for months or years, extending from meteorological to agricultural and hydrological drought, causing severe natural disasters (Wilhite and Glantz 1985; McVicar and Jupp 1998; Yang et al. 2020). For example, the 2010 spring drought in southwestern China reduced winter wheat production by 48 and 31% in Yunnan and Guizhou provinces, respectively (Zhang et al. 2012). The 2011 summer drought in the Yangtze River basin cost approximately 2.4 billion US dollars (Jin et al. 2013; Yuan et al. 2015). Moreover, the 2019 severe drought in southern China affected several provinces, including Anhui, Hubei, Jiangsu, Jiangxi, and Zhejiang provinces, among which, Anhui province was the worst-affected region because it received about 60% less rainfall than normal between August and October (http://www.futuredirections.org.au/publication/southern-china-experiences-another-severe-drought/). To lessen drought damages, remote sensing (RS) techniques (Yuan et al. 2015; Rajasekaran et al. 2018; Blyverket et al. 2019) and land surface models (Mo and Lettenmaier 2014; Ayantobo and Wei 2019; Zhang et al. 2019a) have been used for timely drought monitoring and analysis, especially in large regions.

Soil moisture (SM) is the key component of the earth system that dominates the water transport between soil, vegetation, and atmosphere and controls the water and energy balance, making it a fundamental indicator for drought monitoring, notably for agricultural drought (Brocca et al. 2011; Liu et al. 2019). SM data can be acquired from the measured sites, land surface models (LSM), and remote sensing. The measured site affords the most accurate point-based SM data with poor spatial representatives, however, due to the high costs of maintenance, and the measurement sites are globally sparse. Temporarily, most observation data are no longer adequate for drought monitoring and analysis (Robock et al. 2000).

In contrast, the LSMs can provide large spatial scale and long-term grid-based SM data with a high spatiotemporal resolution (Wang et al. 2016). However, the LSMs data quality depends heavily on the accuracy of the forcing variables, model parameters, and the model physical structures (Fang et al. 2016). The performance of LSMs in drought monitoring has large uncertainties with land surface complexity and climate variability (Levine et al. 2016). To address this issue, RS products have been utilized to adjust and merge LSMs data for SM monitoring and drought analysis (Liu et al. 2019; Zhang et al. 2019b; Vergopolan et al. 2020).

In the past two decades, a series of satellites have been launched for carrying a radiometer (passive) or radar (active) or both, providing band information for SM retrieval (Liu et al. 2012; Al-Yaari et al. 2019). These RS SM data are provided separately (e.g., Soil Moisture Active Passive (SMAP) (Entekhabi et al. 2010), Soil Moisture and Ocean Salinity (SMOS) (Kerr et al. 2001)) or blended (such as the one produced by the ESA project: Climate Change Initiative (CCI). Usually, microwave measurements are not affected by cloud contamination and varying solar illumination (Liu et al. 2019). Therefore, the SM data retrieved directly from these satellite observations are less affected by error accumulation, despite their thin penetration into the bare soil (<5 cm) (Spennemann et al. 2015). The microwave sensors-derived SM is closely related to that generated from the first layer in most LSMs. SM in shallow surface responds fast to meteorological anomalies (e.g., precipitation, evapotranspiration), serving as a sensitive indicator for drought monitoring. However, differences in sensor design, retrieval algorithms, terrain, and vegetation state could result in varying quality and inconsistency of the satellite data (Dorigo et al. 2015).

Many researchers have cross-validated different SM datasets under normal conditions (Brocca et al. 2011; Dorigo et al. 2012; Fang et al. 2016), but data quality shows more substantial variations in extreme conditions (e.g., drought). This could be due to the immature model structures and sparse forcing data for the LSMs and also high noise for the RS datasets (Taylor et al. 2012; Dorigo et al. 2015; Wang et al. 2016). These variations could plausibly influence SM drought monitoring. Therefore, the comparison of SM datasets could identify their consistencies and discrepancies and reveal their applicabilities in drought monitoring (Liu et al. 2019).

Among various SM data datasets, the GLDAS-Noah and the ESA CCI SM products have been widely applied for drought monitoring (Spennemann et al. 2015; Yuan et al. 2015; Zhang et al. 2019b). Therefore, we compare these two datasets for drought characterization across China and its respective characteristics. The objectives of this study include (i) to investigate the consistencies and discrepancies between these two datasets in monitoring droughts in different regions across China; (ii) to evaluate the capacity of the datasets in long-term drought monitoring; and (iii) to evaluate the capabilities of two datasets in drought detection during extreme droughts occurring.

2 Data

2.1 ESA CCI soil moisture

The ESA CCI Soil Moisture project (http://www.esasoilmoisture-cci.org) has been established to satisfy some needs based on passive and active microwave products in support of climate research (Dorigo et al. 2017). The latest release (v04.5) provides global soil moisture data from 1978 to 2018 with a spatial resolution of 0.25° × 0.25° and in units of m3/m3, merging three active (AMI-WS, MetOp-A ASCAT, and MetOp-BASCAT) and seven passive (SMMR, SSM/I, TMI, WindSat, AMSR-E, AMSR2, and SMOS) microwave products. The ESA CCI SM consists of three types (active, passive, and active-passive) of microwave products. The surface SM derived from the ESA CCI passive-active combined product was obtained from 1979 to 2014, and the monthly averaged ESA CCI SM values were subsequently computed for all grids across China.

2.2 GLDAS-Noah soil moisture

NASA Global Land Data Assimilation System Version 2 (GLDAS-2) has three components: GLDAS-2.0, GLDAS-2.1, and GLDAS-2.2. GLDAS-2.0 is forced entirely with the Princeton meteorological forcing input data and provides a temporally consistent series from 1948 through 2014 (Rodell et al. 2004; Beaudoing and Rodell 2019). The surface (0–10cm) SM data from GLDAS-2.0/Noah are used for drought analysis in this study.

2.3 Precipitation data

Monthly gridded precipitation data from 1979 to 2014, with a spatial resolution of 0.5° across China, were acquired from the National Meteorological Information Center (NMIC) which is a department of the China Meteorological Administration (CMA) (http://data.cma.cn/). This dataset was generated using the thin plate spline (TPS) of ANUSPLIN software based on the observed precipitation data from 2472 ground stations in China from 1961 to the latest (McVicar and Jupp 1998). This dataset has been tested using the measured in situ precipitation data across China and showed very high precision (Zhao et al. 2014). The dataset was resampled to 0.25° resolution, as same as ESA CCI and GLDAS-Noah, using a bilinear interpolation method.

3 Methods

3.1 ESA CCI SM gap filling

The ESA CCI SM spans from 1978 to date, making it suitable for drought research (Brocca et al. 2011; Blyverket et al. 2019; Liu et al. 2019). However, the data gaps of the dataset hinder its application (Yuan et al. 2015). Therefore, for each pixel, we filled the missing values with GLDAS-Noah SM data, using the cumulative distribution function (CDF) matching approach (Reichle 2004; Drusch 2005; Brocca et al. 2011) as follows:

  1. (1)

    Plotting the CDF curves of monthly ESA CCI SM and GLDAS-Noah SM, respectively.

  2. (2)

    The polynomial regression is applied to GLDAS-Noah SM and the differences between ESA CCI and GLDAS-Noah SM.

  3. (3)

    Match the CDF of GLDAS-Noah SM to ESA CCI SM, thereby yielding the GLDAS-CDF series (Fig. 1).

Fig. 1
figure 1

Example of cumulative distribution function matching method (a) implemented to rescale GLDAS SM products, against ESA CCI SM data. The GLDAS-CDF product (b) is derived by applying a 5th-order polynomial fitting to the difference between the ranked GLDAS SM and ESA CCI SM data

Full size image

3.2 Standardized Precipitation Index (SPI)

The metrological drought was assessed for comparison, and it was evaluated using the monthly gridded precipitation data. Precipitation significantly and directly influences the SM, especially surface SM. Considering the instant response of shallow SM to precipitation (Szalai and Szinell 2000; Sims et al. 2002), 1-month SPI values were calculated in this study. The formula of SPI is bellowed (Mckee et al. 1993; Lloyd-Hughes and Saunders 2002; Liu et al. 2014):

$$ SPI=\left\{\begin{array}{l}\begin{array}{l}-\Big(t-\frac{C_0+{C}_1t+{C}_2{t}^2}{1+{d}_1t+{d}_2{t}^2+{d}_3{t}^3}\kern0.24em \;\kern0.24em for\;\kern0.24em \;\;0<H(x)\le 0.5\\ {}t=\sqrt{\ln \left(1/H{(x)}^2\right)}\end{array}\\ {}t-\frac{C_0+{C}_1t+{C}_2{t}^2}{1+{d}_1t+{d}_2{t}^2+{d}_3{t}^3}\kern0.24em \;\kern0.24em \;\kern0.24em for\;\kern0.24em \;\;0.5<H(x)<1.0\\ {}t=\sqrt{\ln \left(1/\left(1-H{(x)}^2\right)\right)}\end{array}\right. $$
(1)

where the coefficients C0, C1, C2, d1, d2, and d3 are given as follows: C0=2.515517, C1=0.802853, C2=0.010328, d1=1.432788, d2=0.189269, and d3=0.001308. H(x) represents the cumulative probability, which is obtained from the Gamma CDF. One can refer to Ayantobo et al. (2017) for more details. Drought classification criteria are as shown in Table 1.

Table 1 Drought classification criteria for soil moisture droughts (SSI) and meteorological droughts (SPI)
Full size table

3.3 Drought identification

Droughts across China were characterized by the Standardized Soil Moisture Index (SSI) (Hao and AghaKouchak 2013; AghaKouchak 2014; Hao et al. 2014) from the ESA CCI SM and GLDAS-Noah datasets. The SSI can be calculated similarly to SPI (Mckee et al. 1993). The CDF is estimated using a nonparametric approach where the empirical probability (P) of the SM data is derived using the empirical Gringorten plotting position (Gringorten 1963). Then the P is transformed into a standard Gaussian distribution function, SSI = Φ-1(P), where Φ is the standard Gaussian distribution function, and the greatest negative value of SSI represents the most severe drought. According to the SSI classification criteria (Hao et al. 2014), this study focused on the moderate (SSI<−0.8) and above grade droughts, which are referred to as droughts hereinafter.

3.4 SSI correlations

The correlations (R) of ESA CCI SM-derived SSI and GLDAS-Noah SM-derived SM were calculated for each calendar month at the pixel scale, using the formula given below:

$$ r=\frac{n\sum \limits_i{SM}_{ESAi}{SM}_{GLDASi}-\sum \limits_i{SM}_{ESAi}\sum \limits_i{SM}_{GLDASi}}{\sqrt{n\sum \limits_i{SM_{ESAi}}^2-{\left(\sum \limits_i{SM}_{ESAi}\right)}^2\sqrt{n\sum \limits_i{SM_{GLDASi}}^2-{\left(\sum \limits_i{SM}_{GLDASi}\right)}^2}}} $$
(2)

where SMESAi and SMGLDASi represent the ith year’s ESA CCI SM and the GLDAS-Noah SM.

3.5 Drought characterization and trend detection

In this paper, drought duration, frequency, and severity were calculated to determine their effectiveness for drought monitoring and analysis. Drought duration refers to the consecutive months where the drought index is below the truncation level. Drought frequency represents the reciprocal of the average drought interval in a given period, while drought severity refers to the cumulative sum of values during a drought period (Ayantobo et al. 2017)

The trends for SSI and the corresponding drought characteristics are examined utilizing the Mann-Kendall (M-K) method, a very popular method for trend identification (Dorigo et al. 2012; Liu et al. 2012). The M-K trend test defined as in Eq. 3 is used to quantify the significance of trends in the time series (Raziei 2017),

$$ S\kern0.5em =\kern0.5em \sum \limits_{i=1}^{n-1}\sum \limits_{j=i+1}^n\mathit{\operatorname{sign}}\left({x}_j-{x}_i\right) $$
(3)

where xi and xj are the sequential data values in the time series (j > i) and n represents the length of the dataset. The sign(xj−xi) is equal to 1, 0, or −1 when the resulting value is greater than, equal to, or less than zero, respectively. If the sample size n is greater than 30, the standard normal test statistic z can be computed to quantify the trend using Eq. 4:

$$ z\kern0.5em =\kern0.5em \left\{\begin{array}{l}\frac{S-1}{\sqrt{Var(S)}}\kern0.5em if\kern0.5em S>0\\ {}0\kern6em if\kern0.5em S=0\\ {}\frac{S+1}{\sqrt{Var(S)}}\kern0.5em if\kern0.5em S<0\end{array}\right. $$
(4)

where Var(S) is the variance of S, given by:

$$ Var(S)=\frac{1}{18}\left[n\left(n-1\right)\left(2n+5\right)-\sum \limits_t{f}_t\left({f}_t-1\right)\left(2{f}_t+5\right)\right] $$
(5)

where t varies over the set of tied ranks and ft is the number of times that the rank t appears. The M-K z-score was used to test the significance level for Gaussian distribution, with p < 0.05 representing a significant trend and vice versa.

3.6 Comparison with typical droughts

To evaluate the performances of these two SM datasets in drought monitoring, several historical droughts were chosen for drought comparison across China. These events consist of the 2010 spring drought in southwestern China (Zhang et al. 2012), the 2010–2011 severe winter drought in the North China Plain (Jin et al. 2013), and the 2011 summer drought in the middle and lower reaches of the Yangtze River basin of China (Yuan et al. 2015).

4 Results and discussion

4.1 SSI correlations between ESA CCI SM and GLDAS-Noah

Figure 2 shows R values between SSI values derived from ESA CCI and GLDAS-Noah, with the median, upper, and lower quartiles for each calendar month identified. In most cases, the R values are more than 0.2, and the median correlations are more than 0.4.

Fig. 2
figure 2

Pearson’s correlation coefficient (r) between the monthly SSI calculated using ESA CCI SM and GLDAS-Noah SM datasets for each calendar month over the study region

Full size image

Figure 3 shows the R patterns of SSI for seasons and the whole year. The data gaps in ESA CCI, which could either be due to changing input sensor constellation or to natural physical phenomena (e.g., soil frost), lead to the missing R values in some regions (white blank). For the whole year, the significant R values dominate over 70% across China. Insignificant even negative correlations mainly appeared in the northwestern arid regions, Tibetan regions, and northeastern cold regions. This figure also shows that correlations vary with the season, with more significant correlations in more regions in warm seasons and lesser in cold seasons. For the entire month, the SSI calculated using ESA CCI SM showed good agreement with GLDAS-Noah SM data. The low correlations observed in the high latitude (northeastern China), high altitude (Tibet Plateau), and arid regions (Xinjiang and Inner Mongolia) were ascribed to the inability of sensors to obtain SM data in freeze-thaw zones (Dorigo et al. 2015) and large uncertainties of LSM in these regions (Ferguson and Wood 2011). The GLDAS outputs accuracy is controlled by inputs accuracy, parameters uncertainty, and feedback mechanisms for land-atmosphere (Ferguson and Wood 2011; Taylor et al. 2012; Liu et al. 2019). In northwestern arid regions and Tibetan, the gauge density for GLDAS model inputs (e.g., precipitation) is sparser than in other regions, which makes more errors and high uncertainty for model outputs. Besides, in freeze/thaw surface (Tibetan Plateau, and northeastern China), due to the physical limitations (frozen ground impacting the backscatter measurements), it is a challenging task to obtain accurate retrieval of SM from microwave measurements (Zwieback et al. 2015), in which the noise of satellite data may drown out the signal (Reichle 2004).

Fig. 3
figure 3

Spatial distribution of Pearson’s correlation coefficients between the SSI calculated using ESA CCI SM and GLDAS-Noah SM datasets for seasons and whole year over 1979–2014

Full size image

4.2 Assessment of drought characteristics

Figure 4 presents the spatial patterns of SM for ESA CCI and GLDAS-Noah and their averaged differences for the 1979–2014 period. Generally, the distribution patterns and values of the two datasets are in agreement, with SM values decreasing gradually from the southeast to the northwest. High SM (> 0.3 m3/m3) could be found in the humid and subhumid areas, while low SM (< 0.2 m3/m3) dominates the arid areas (Fig. 5). However, the ESA CCI SM was lower than GLDAS-Noah SM in humid and semi-humid regions (Fig. 5), while the ESA CCI SM was greater than GLDAS-Noah SM in other regions, especially in the Tibetan Plateau.

Fig. 4
figure 4

Spatial distribution of surface soil moisture (m3/m3) for a the ESA CCI SM, b the GLDAS-Noah SM, and c their differences subtracting from the ESA CCI SM by the GLDAS-Noah SM

Full size image
Fig. 5
figure 5

China aridity map (from http://ref.data.fao.org/map?entryId=221072ae2090-48a1-be6f-5a88f061431a&tab=about). AI is an aridity index calculated by dividing the average yearly precipitation using the average yearly potential evapotranspiration

Full size image

Figure 6 displays the patterns of averaged drought characteristics for the ESA CCI and the GLDAS-Noah SM over 1979–2014 for 1-month SSI. The spatial distribution of drought duration according to ESA CCI SM generally agreed with the GLDAS-Noah SM (Fig. 6a, b). According to our analysis, drought duration ranged from 1 to 4 months in the whole region. The durations over the arid and subarid regions are generally higher than that over the humid regions. Moreover, 82% of the regions have a duration of fewer than 2 months spreading over the humid regions (Fig. 5) for ESA CCI SM and 86% for GLDAS-Noah SM. Also, 52% of the region with a duration greater than 3 months are found in the arid regions (Fig. 5) for ESA CCI SM and 69% for GLDAS-Noah SM. Generally, the duration derived from GLDAS-Noah SSI was higher (for 81% total pixels) than that derived from ESA CCI SSI. The discrepancies in averaged drought duration are mainly found in the arid regions (southern Xinjiang, northern Tibetan Plateau) and some cold regions (eastern Inner Mongolia and Northeast).

Fig. 6
figure 6

Spatial distribution of averaged drought duration, severity, and frequency for the ESA CCI SM (a, c, e) and the GLDAS SM (b, d, f) over 1979–2014 based on SSI

Full size image

The spatial distribution of drought severity is consistent with that of drought duration for the two datasets (Fig. 6c, d). Droughts were more severe over the arid regions (i.e., with an averaged severity of about −2.85 and −3.40 for ESA CCI and GLDAS-Noah, respectively) than that over the humid regions (i.e., with averaged severity of about −2.58 and −2.90 for ESA CCI SM and GLDAS-Noah datasets, respectively). Major discrepancies were also found in the arid regions.

The two datasets displayed a consistent spatial pattern of drought frequency. Drought events had a relatively higher frequency in the humid and semi-humid regions (i.e., with an averaged frequency of about 0.34 month-1 and 0.37 month-1 for ESA CCI and GLDAS-Noah, respectively) and relatively lower frequency in the arid regions (i.e., with an averaged frequency of about 0.35 month-1 and 0.32 month-1 for ESA CCI and GLDAS-Noah SM, respectively). This suggested that drought occurred more frequently but with shorter duration and slight severity in humid and semi-humid regions, while it occurred more severely with longer duration and relatively lower frequency in the arid region.

The obvious disparity for drought characteristics monitored by ESA CCI and GLDAS-Noah SM datasets mostly concentrated in arid (northwestern China) and cold regions (northeastern and Tibetan regions). This is due to the low signal-to-noise ratio for satellite SM over arid regions and high uncertainty in LSM in high altitude, freeze-thaw zones, and dry climate zones.

4.3 Drought trend detection

Long-term trends of SSI and corresponding drought characteristics between ESA CCI SM and GLDAS-Noah SM were compared using the M-K trend test method. Figure 7 shows the Mann-Kendall z-score in SSI for ESA CCI SM and GLDAS-Noah SM. Both two SM datasets showed similar spatial patterns with drying trends over central, southern, and eastern China and wetting trends over northwestern China. Accordingly, 35.8 and 16.7% of the regions showed a significant drying trend, while 16.7 and 17.5% of the regions presented a significant wetting trend for ESA CCI SM and GLDAS-Noah SM, respectively. According to our results, most of the above regions were found in the humid and semi-humid zone. The grid to grid consistent ratios of the significant drying and wetting trends between ESA CCI SM and GLDAS-Noah SM were ~18.1 and 10.0%, respectively, and major discrepancies in SSI trends between the two datasets are mainly located in southwestern China (Fig. 7).

Fig. 7
figure 7

Spatial distribution of trend significance (M-K z-score) in SSI for a ESA CCI SM and b GLDAS SM over 1979–2014

Full size image

To further understand drought changes, the average SSI was calculated over southwestern China (i.e., Yunnan, Sichuan, Guizhou, and South Tibet) monthly from 1979 through 2014 (Fig. 8). The SSI for ESA CCI SM showed a slight and insignificant increasing trend from 1979 to 1999, while after 2000, SSI decreases significantly (R2=0.20, p < 0.01) with a change rate of −0.04/year (Fig. 8a). In contrast, the SSI for GLDAS-Noah SM had no obvious trend for the whole period (Fig. 8b). This showed that ESA CCI SM agreed well with GLDAS-Noah SM before 2000, but ESA CCI SM was generally lower than GLDAS-Noah SM between 2000 and 2014, especially in the winter season.

Fig. 8
figure 8

Averaged monthly SSI changes over southwest China (Yunnan, Sichuan, Guizhou, and south Tibet) from 1979 to 2014 for ESA CCI SM (a) and GLDAS SM (b)

Full size image

For drought duration, few pixels presented a significant drying trend (z-score > 1.96) for both two datasets (i.e., 7% for ESA CCI SM and 2% for GLDAS-Noah SM) (Fig. 9a, b). For the ESA CCI SM, major significant upward trends appeared in the humid and semi-humid regions, such as southwestern China, where drying trends had been previously detected (Liu et al. 2017). In the case of GLDAS-Noah SM, significant increasing (drying) trends appeared in some northeastern areas and parts of the central and southern China (Henan and Jiangxi provinces).

Fig. 9
figure 9

Spatial distribution of trend significance (M-K z-score) in drought duration, severity, and frequency for ESA CCI SM (a, c, e) and GLDAS SM (b, d, f) over 1979–2014

Full size image

The drought severity showed similar trends with drought duration, indicating that longer drought events corresponded to more severe drought events. Significant decreasing (drying) trends were detected over southwestern China for the ESA CCI SM, while the GLDAS-Noah SM showed decreasing trends over northeastern China, east Inner Mongolia, and parts of the central and southern China. Except for Inner Mongolia and Henan provinces, other regions with drying trends are either humid or semi-humid regions. Over the study period, most parts of northwestern China showed a wetting trend both for ESA CCI SM and GLDAS-Noah SM datasets, which was found to be consistent with previous studies (Peng and Zhou 2017).

For the trend significance in drought frequency (Fig. 9e, f), both recorded drying trends (z-score > 1.96) over some parts of northeastern and central China. Concerning the difference, the ESA CCI SM had more regions of increasing drought frequency than the GLDAS-Noah SM in southwestern China. Conversely, the GLDAS-Noah SM had more remarkable wetting trends than the ESA CCI SM, especially in northwestern China (Xinjiang and Qinghai provinces). The discrepancies between ESA CCI and GLDAS-Noah in drought trend detection stress the uncertainties in satellite-derived SM and LSM-modeled SM for drought identification (Seneviratne 2012; Sheffield et al. 2012). The differences are ascribed to the disturbances for satellite retrieval and LSM modeling, for instance, dense vegetation in southern humid China and freeze-thaw surface in northeastern China and Tibetan Plateau.

4.4 Comparison with historical drought events

To verify the quality of these two SM datasets for drought monitoring over China, the SPI-1 data computed using monthly grid precipitation data were used. Three drought events were selected for comparison, including the 2009/2010 southwestern drought (Fig. 10a, b, c), the 2013 southern drought (Fig. 10d, e, f), and the 2007 northern drought (Fig. 10g, h, i).

Fig. 10
figure 10

Extreme drought events detected from the grid precipitation (a, d, g), the GLDAS-Noah (b, e, h), and the ESA CCI (c, f, i) soil moisture. (a, b, c) The drought in February 2010 in southwestern China; (d, e, f) the drought in July 2013 in southern China; and (g, h, i) the drought in July 2007 in northern China

Full size image

The extreme drought events extending from autumn 2009 to spring 2010 over southwestern China were the driest event, having the lowest percentage of rainfall anomaly and the longest non-raining days during the winter season in the past 50 years (Yang et al. 2011). This event was the most severe, having the lowest percentage rainfall anomaly in the same period since 1880 (Yang et al. 2011), causing serious economic losses (nearly US$30 billion). The SPI showed that drought centers were located in Guangxi, Guizhou, and Hunan provinces. ESA CCI and GLDAS-Noah also captured spatial patterns of this event; however, they overestimated the severity in Yunnan province and Tibetan Plateau. Specifically, the SPI-1 categorized the drought as moderate and mild drought, while the ESA CCI and GLDAS-Noah classified it as severe, especially for GLDAS-Noah. These discrepancies could be ascribed to high altitude and sparse hydro-meteorological stations, which increased the uncertainty of LSM modeling.

Drought events over southern China in July 2013 were caused by a strong west Pacific subtropical high that extended to the north from its normal position. As shown in the SPI-1 plot (Fig. 10d), the huge drought centers were located in Guizhou, Hunan, and Zhejiang provinces. The GLDAS-Noah perfectly captured the event and thus classified it as a severe drought, which was the same as SPI-1. Comparatively, the ESA CCI displayed similar spatial patterns with a smaller extent and lower severity, especially in Zhejiang province. This may be associated with the dense vegetation over these regions, which compromised the accuracy of satellite SM (Albergel et al. 2013).

In 2007, there was a widespread drought in parts of northern and northeastern China. As displayed in Fig. 10g, the severe drought attacked most of the Heilongjiang province and eastern Inner Mongolia. The ESA CCI dataset displayed a similar pattern with grid precipitation, but with relatively low severity. However, the GLDAS-Noah dataset displayed a larger spatial drought extent, with higher severity. In central China (south Shaanxi, Henan, and Hubei provinces), northern Gansu province, and western Inner Mongolia, the grid precipitation-based SPI-1 showed a wetting trend. The ESA CCI dataset successfully monitored this status, while the GLDAS-Noah dataset failed to detect the wetting pattern. Overall, in central and northern China, the ESA CCI dataset outperforms the GLDAS-Noah dataset in extreme droughts monitoring.

5 Conclusions

ESA CCI SM and GLDAS-Noah soil moisture data were analyzed for drought identification and monitoring over 1979–2014 using SSI. Our results show that large regions of significant correlations between ESA CCI SSI and GLDAS-Noah SSI, indicating high consistency between the two datasets.

Both SM datasets display a similar pattern in drought duration, severity, and frequency. Generally, more severe droughts are found in the arid regions than in humid regions. Also, both SM datasets show a similar trend in drought variations of SSI, and significant drying trends were concentrated in arid regions. However, the two datasets also present a disparity in trends, especially in the Tibet Plateau regions.

To verify the capabilities of the two soil moisture datasets in drought monitoring, three extreme droughts were selected for comparison, and the grid precipitation-based SPI-1 was generated for comparison. It is demonstrated that the two datasets are capable of detecting and describing the spatial pattern of large-scale drought events. However, some discrepancies existed regarding drought severity and drought centers. The ESA CCI dataset is not efficient in south China with dense vegetation, while the GLDAS-Noah dataset is relatively incapable of detecting drought in high altitude and latitude regions. It is also demonstrated that the ESA CCI SM is more robust in delimiting the spatial pattern in the arid regions, while the GLDAS-Noah SM performed better in the humid and regions.

ESA CCI SM and GLDAS-Noah exhibit high consistencies in spatial patterns, also showing their potentials in drought monitoring across China. However, a substantial discrepancy exists between the two datasets, particularly over high altitude and latitude regions. In the future, multisource data assimilation could be an effective method for drought research.