remote sensing of environment -...

Remote Sensing of Environment 163 (2015) 91–110

Contents lists available at ScienceDirect

Remote Sensing of Environment

j ourna l homepage: www.e lsev ie r .com/ locate / rse

Evaluation of remotely sensed and reanalysis soil moisture products overthe Tibetan Plateau using in-situ observations

Jiangyuan Zeng a,b, Zhen Li a,⁎, Quan Chen a, Haiyun Bi a,b, Jianxiu Qiu c, Pengfei Zou a,b

a Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences, Beijing 100094, Chinab University of Chinese Academy of Sciences, Beijing 100049, Chinac Guangdong Provincial Key Laboratory of Urbanization and Geo-simulation, School of Geography and Planning, Sun Yat-sen University, Guangzhou 501275, China

⁎ Corresponding author at: Institute of Remote SensAcademy of Sciences, Beijing 100094, China. Tel.: +86 10

E-mail address: [email protected] (Z. Li).

http://dx.doi.org/10.1016/j.rse.2015.03.0080034-4257/© 2015 Elsevier Inc. All rights reserved.

a b s t r a c t
a r t i c l e i n f o
Article history:Received 15 September 2014Received in revised form 13 March 2015Accepted 16 March 2015Available online 4 April 2015

Keywords:Remotely sensed soil moistureReanalysis soil moistureIn-situValidationTibetan Plateau

Long-term and large-scale remotely sensed and reanalysis soil moisture products of the Tibetan Plateau are veryimportant for understanding the land–atmosphere interactions in this area and their impacts on theweather andclimate in the Asian continent. However, it is of great importance to assess the reliability of these products beforeusing them. In the study, in-situ soil moisture measurements from three networks which represent different cli-matic and vegetation conditions over the Tibetan Plateau are used to evaluate the skill of seven remotely sensedsoilmoisture products and one reanalysis soilmoisture product in the period of 2002–2012. The remotely sensedsoil moisture products include three widely used products derived from the Advanced Microwave ScanningRadiometer — Earth Observing System (AMSR-E): the National Aeronautics and Space Administration (NASA)soil moisture product, the Land Parameter Retrieval Model (LPRM) soil moisture product, and the JapanAerospace Exploration Agency (JAXA) soil moisture product; the remaining products are the JAXA AMSR2 soilmoisture product, the Soil Moisture and Ocean Salinity (SMOS) soil moisture product, the AdvancedScatterometer (ASCAT) soil moisture product, and the Essential Climate Variable (ECV) soil moisture productwhich is a new merged product using both active and passive soil moisture products. The reanalysis product isthe latest ERA-Interim product produced by the European Centre for Medium Range Weather Forecasts(ECMWF). The results show that all these products can generally capture the soil moisture dynamics well exceptfor theNASA productwhich significantly underestimates the soilmoisture and also lacks temporal dynamics. TheJAXA AMSR-E and AMSR2 products underestimate the ground measurements at most of the time, whereas theLPRM product gives much larger seasonal amplitude than the in-situ observations with a large positive bias. Itseems that the SMOS observations are severely affected by the well-known radio-frequency interference (RFI)which leads to a big noise and bias in the SMOS product. The ASCAT product overestimates the groundmeasure-ments, but it correlateswith in-situ soilmoisture verywell and is also less influenced by vegetation cover. In gen-eral, the ECV and ERA products outperform other products. Though the ECV product underestimates the soilmoisture, it shows the best correlationwith groundmeasurements and captures the variation of in-situ soilmois-ture very well while the ERA product is closest to the absolute values of soil moisture observations. Overall, mostof the products give reasonable results in terms of correlation in sparsely vegetated areas.We expect that the val-idation results can be used as a feedback to the algorithmdevelopers to further enhance the accuracy of soilmois-ture retrievals over the Tibetan Plateau.

© 2015 Elsevier Inc. All rights reserved.

1. Introduction

Soil moisture is not only an important part of the earth ecosystem,but also a critical link between the land surface and atmosphere. It hasbeenwidely used as a key variable in numerous environmental applica-tions, including the hydrological modeling (Brocca et al., 2012), the nu-merical weather forecasting (Dharssi, Bovis, Macpherson, & Jones,2011), the modeling of land surface evaporation (Miralles et al., 2011),

ing and Digital Earth, Chinese82178751.

and the prediction of surface runoff (Brocca et al., 2010). Global or con-tinental scale soilmoisture information is often needed for such applica-tions. In-situ measurements are able to provide accurate information ofsoil moisture. However, they are only representative over a very smallspatial scale due to the strong heterogeneity of the land surface. It isalso impractical to deploy dense stations all over the world. Therefore,in-situ observations cannot fully characterize the spatial and temporalvariability of soilmoisture at large scales. An alternativeway is to obtainsoil moisture from satellite microwave remote sensing. It has been con-sidered the best way tomonitor soil moisture both temporally and spa-tially since it is the only technology that provides a directmeasure of theabsolute soil moisture contents which are already spatially averaged,

http://crossmark.crossref.org/dialog/?doi=10.1016/j.rse.2015.03.008&domain=pdf

http://dx.doi.org/10.1016/j.rse.2015.03.008

mailto:[email protected]

http://dx.doi.org/10.1016/j.rse.2015.03.008

http://www.sciencedirect.com/science/journal/00344257

92 J. Zeng et al. / Remote Sensing of Environment 163 (2015) 91–110

and it also has the advantage of large-scale observation and high tempo-ral resolution (Owe, de Jeu, & Holmes, 2008).

Microwave remote sensing has been effectively applied to measuresoil moisture since the late 1970s (Schmugge, Jackson, & McKim,1980). Over the past decades, researchers have exerted great effortson development of soil moisture retrieval algorithms for various micro-wave remote sensing satellites/sensors, including the AdvancedScatterometer (ASCAT) (Naeimi, Scipal, Bartalis, Hasenauer, & Wagner,2009;Wagner, Lemoine, & Rott, 1999;Wen&Su, 2003a, 2003b), theAd-vanced Microwave Scanning Radiometer — Earth Observing System(AMSR-E) (Jackson, 1993; Jones, Jones, Kimball, & McDonald, 2011;Koike et al., 2004; Njoku & Chan, 2006; Njoku, Jackson, Lakshmi, Chan,& Nghiem, 2003; Owe et al., 2008; Paloscia, Macelloni, & Santi, 2006;Pan, Sahoo, & Wood, 2014; Santi et al., 2012; Zeng, Li, Chen, & Bi,2015), the Advanced Microwave Scanning Radiometer-2 (AMSR2)(Koike et al., 2004), the Windsat (Li et al., 2010; Parinussa, Holmes, &de Jeu, 2012), and the Soil Moisture and Ocean Salinity (SMOS) satellite(de Jeu, Holmes, Panciera, &Walker, 2009; Kerr et al., 2012; Saleh et al.,2009;Wigneron et al., 2007). Currently, several microwave satellite soilmoisture products are released to the public free of charge, such as theNational Aeronautics and Space Administration (NASA) AMSR-E soilmoisture product (Njoku & Chan, 2006), the Land Parameter RetrievalModel (LPRM) soil moisture product (Owe et al., 2008), the Japan Aero-space Exploration Agency (JAXA) AMSR-E and AMSR2 soil moistureproducts (Koike et al., 2004), the SMOS soil moisture product (Kerret al., 2012), and the ASCAT soil moisture product (Naeimi et al., 2009;Wagner et al., 1999). However, it is very important to assess the accura-cy and reliability of these products before using them. The purpose ofverification includes two aspects: one is that the validation results canbe used as a feedback to algorithm developers to help them furtherimprove the algorithms, and the other is to facilitate the potentialusers to understand the status of the products and thus can better usethem for practical applications. Some studies have assessed the AMSR-E, SMOS and ASCAT soil moisture products using model simulations(Al-Yaari, Wigneron, Ducharne, Kerr, de Rosnay, et al., 2014; Al-Yaari,Wigneron, Ducharne, Kerr, Wagner, et al., 2014; Dorigo et al., 2010;Draper et al., 2013; Qiu et al., 2013; Rötzer et al., 2014; Rüdiger et al.,2009; Wanders et al., 2012) or in-situ measurements (Albergel et al.,2012; Jackson et al., 2010, 2012; Leroux et al., 2014; Pan et al., 2012;Sahoo et al., 2008; Su et al., 2011; Wagner, Naeimi, Scipal, de Jeu, &Martínez-Fernández, 2007), while less evaluation work has been donefor the latest AMSR2, and the Essential Climate Variable (ECV) soilmois-ture products (Albergel et al., 2013; Dorigo et al., 2015). The latter is anew merged product using both active and passive soil moisture prod-ucts (Liu et al., 2011, 2012; Wagner et al., 2012), which urgently needsto be validated using more “ground truth”. In addition, most validationwork was performed in Europe (Albergel et al., 2011; Brocca et al.,2011; Lacava et al., 2012; Parinussa, Yilmaz, Anderson, Hain, & de Jeu,2014; Wagner et al., 2007; Wigneron et al., 2012), the United States(Al Bitar et al., 2012; Jackson et al., 2010, 2012; Leroux et al., 2014;Pan et al., 2012; Sahoo et al., 2008), and Australia (Draper, Walker,Steinle, de Jeu, & Holmes, 2009; Rüdiger et al., 2011; Su, Ryu, Young,Western, & Wagner, 2013), while less verification activities were con-ducted in Asia, especially over the Tibetan Plateau which is one of themost special and least explored areas on earth, but yet of great impor-tance for understanding the global change (Su et al., 2011).

The Tibetan Plateau, which is often called the “Third Pole” or the“Roof of the World”, is the highest and most extensive plateau in theworld (Yang et al., 2013). It is also themost prominent and complicatedterrain on the earth as well as one of the most sensitive areas to globalchange (Su et al., 2011; Xie, Ye, Liu, & Chongyi, 2010). Many studieshave demonstrated that the Tibetan Plateau directly impacts its sur-rounding climate and environment through atmospheric and hydrolog-ical processes, thus greatly influencing circulations over China, Asia andeven the globe (Ma et al., 2003; Yang et al., 2011). Soil moisture is one ofthe most important parameters, which plays a key role in the water

cycle and the climate of the plateau, as a result, affecting the monsoonsystem and precipitation patterns. Consequently, reliable long-termand large-scale remotely sensed soil moisture products of the TibetanPlateau are very crucial for understanding the land–atmosphere inter-actions of this region and their effects on the climate of Eastern andSouth-East Asia. However, due to the special geographical environmentand harsh climate conditions, the Tibetan Plateau was rarely explored,and thus there has been a lack of measurements of the key surface pa-rameter soil moisture in the region for a long time. As a result, veryfew evaluation activities have been performed in this region so far.Until recent years, Su et al. (2011) and Chen et al. (2013) evaluated sev-eral satellite soil moisture products in this area.

In the study, a detailed evaluation of seven remotely sensed soilmoisture products, including the NASA AMSR-E soil moisture product,the LPRM AMSR-E soil moisture product, the JAXA AMSR-E andAMSR2 soil moisture products, the SMOS soil moisture product, theASCAT soil moisture product, and the ECV soil moisture product, aswell as one soil moisture reanalysis product (ERA-Interim), is conduct-ed over three networks established in the Tibetan Plateau in the periodof 2002–2012. Dense sites have been deployed in these networkswhichrepresent different climatic and vegetation conditions that can ensure arobust validation of the soil moisture products. In the following section,the networks and the soil moisture products used for this study arebriefly introduced. Section 3 describes the method for the evaluationof the products. The results are then presented in Section 4. Section 5discusses the possible explanations for the results and also gives somesuggestions. Finally, some conclusions of the study are summarized inSection 6.

2. Materials

Due to the special geographical environment, extreme climate con-ditions and the high cost of conducting experimental work in the Tibet-an Plateau, this region has been a lack of soil moisture and temperaturemeasurements for a long time. Fortunately, great efforts have beenmade in the last decade to establish long-term observation networksin various biomes and climate conditions over the Tibetan Plateau.These measurements provide a lot of valuable experimental data forvalidation of various soil moisture products in the Tibetan Plateau. Thefollowing is a brief introduction of the existing soil moisture and tem-perature networks of the Tibetan Plateau and the soil moisture productsused in the study.

2.1. In-situ measurements

2.1.1. CAMP/TibetThe soil moisture and temperature measurement system (SMTMS)

network of the CEOP (Coordinated Enhanced Observing Period) Asia–Australia Monsoon Project (CAMP) on the Tibetan Plateau (CAMP/Tibet) is a meso-scale observational network set up over the central Ti-betan Plateau (Ma et al., 2003). The objective of this network is to mon-itor soil moisture and temperature to develop the land surface processmodels and satellite-based soil moisture retrieval methods. The net-work consists of eight stations, including Amdo, BJ, D66, D105, D110,MS3608, MS3637 and Tuotuohe. The ground surface of all stations issparsely coveredwith short grasses (Wen, Su, &Ma, 2003). The locationand distribution of the stations are shown in Fig. 1. Each site recordedsoil moisture and temperature from 1 October 2002 to 31 March2004 at different depths (from 4 to 250 cm below surface) with a tem-poral resolution of 1 h, and soil moisture and temperature aremeasuredby using a Time-Domain Reflectometry (TDR) system and a platinumresistance thermometer (Pt100 sensor), respectively. The CAMP/Tibetnetwork is in a cold semiarid climate, and soil thawing and freezingbegin in May and November respectively owing to the cold environ-ment. In the study, the daily averaged soil moisture measured by thesensors at 4 cm of all sites of CAMP/Tibet network was used. For more

Fig. 1. Location of the three networks and distribution of the corresponding sites in eachnetwork. Yellow, blue and black rectangles represent the CAMP/Tibet, Naqu, and Maqunetwork regions respectively, and the stars, triangles, and circles in each rectangle repre-sent the sites in each network. The legend of the color map indicates the altitude abovemean sea level in meters.

93J. Zeng et al. / Remote Sensing of Environment 163 (2015) 91–110

details about this network, readers are kindly referred to Ma et al.(2003).

2.1.2. MaquThe Maqu soil moisture and temperature monitoring network is lo-

cated in the north-eastern part of the Tibetan Plateau, and was set up inJuly 2008 (Su et al., 2011). The network consists of 20 sites and coversan area of approximately 40 km × 80 km, shown in Fig. 1. In thesesites, soil moisture and temperature were measured from 1 July 2008to 31 July 2010 at various depths (from 5 to 80 cm below surface)with a temporal resolution of 15 min (Su et al., 2011). All sites aredistributed in the large valley of the Yellow River and the surroundinghills, covered by short grassland uniformly. The network has a humidand cold climate with rainy summers and dry winters (Dente,Vekerdy, Wen, & Su, 2012). The monsoon season of the Maqu networkregion is fromApril to October, and beyond the period, the soil is frozen.In the study, the daily averaged soil moisture measured by the sensorsat 5 cm of all the 20 sites of Maqu network was used. For more detailsabout Maqu network, readers are kindly referred to Su et al. (2011).

2.1.3. NaquThe Naqu soil moisture and temperature monitoring network is a

multi-scale network set up in the central Tibetan Plateau within anarea of 100 km× 100 km (Yang et al., 2013). It is made up of 56 stationswhich measures soil moisture and temperature at three spatial scales(1.0, 0.3, 0.1°) corresponding to large-scale, medium-scale, and small-scale networks respectively, and at four soil depths (0–5, 10, 20, and40 cm), shown in Fig. 1. These data were collected between 1 August2010 and 31 December 2012 at 30 min intervals. The land surface ofthe network region is typically characterized by alpine meadows. Dueto the low biomass, low air mass and air moisture which leads toweak attenuation of microwave signals, this region is an ideal placefor the validation of satellite-based and model-simulated soil moistureproducts (Yang et al., 2013). The network is in a cold semiarid climate,and soil thawing and freezing begin inMay and November respectively.

Table 1Main characteristics of the three networks.

Networks Sites Climate Measuring depth of surface soil mo

CAMP/Tibet 8 Cold semiarid 4Maqu 20 Cold humid 5Naqu 56 Cold semiarid 0–5

In the study, the daily averaged soil moisture measured by the sensorsat 0–5 cm from 38 stations of the Naqu large-scale network was used.Table 1 summarizes the main characteristics of each network. Formore details about Naqu network, readers are kindly referred to Yanget al. (2013).

Furthermore, in order to investigate the vegetation cover conditionsof the three network regions, the Normalized Difference VegetationIndex (NDVI) was used in the study. The NDVI data were obtainedfrom the Free Vegetation Products (www.vito-eodata.be), resultingfrom 10 days of SPOT-4/Vegetation acquisitions with a resolution of1 km. The averaged NDVI of the three network regions from 1 January2002 to 31 December 2012 is shown in Fig. 2. It can be seen that thereis a clear vegetation cover distinction between the three network re-gions. TheMaqu network region has the highest NDVI during the entireperiod while the CAMP/Tibet network region has the lowest NDVI,which indicates the densest and sparsest vegetation cover in the two re-gions respectively. Thus, these networks supply a representative cover-age of different climate and vegetation cover conditions on the TibetanPlateau, which will contribute to a more robust validation of the soilmoisture products in this region.

2.2. Soil moisture products

2.2.1. AMSR-EAMSR-E was the first satellite sensor to incorporate soil moisture as

a standard productwith an expectant accuracy of less than 0.06m3m−3

(Njoku & Chan, 2006; Njoku et al., 2003). It measures brightness tem-peratures at six dual-polarized frequencies in the range of 6.9–89 GHzand scans the Earth's surface in an ascending (13:30 local solar time)and descending (01:30 local solar time) mode. Several algorithms areroutinely applied to estimate soil moisture from AMSR-E observations.In the study, three most widely used soil moisture products derivedfrom AMSR-E: the NASA soil moisture product (version 6, hereaftercalled NASA product) (Njoku & Chan, 2006), the JAXA soil moistureproduct (version 700, hereafter called JAXA product) (Koike et al.,2004) and the LPRM soil moisture product (version 2, hereafter calledLPRM product) (Owe et al., 2008) were used. The original algorithm ofNASA is a multifrequency-polarization iterative algorithm, which per-forms an iterative minimization of the root mean square error (RMSE)between model simulations and satellite observations for three param-eter (soil moisture, vegetation water content and soil temperature) re-trievals (Njoku et al., 2003). NASAmodified its earlier version by using aglobal regressive method (Njoku & Chan, 2006). The algorithm usesmi-crowave polarization difference index (MPDI) (Becker & Choudhury,1988) of AMSR-E brightness temperatures at 10.65 and 18.7 GHz to ac-count for the effects of vegetation and roughness, and soil moisture isestimated using the deviation of MPDI at 10.65 GHz from a baselinevalue. In the JAXA algorithm, themethod of look-up tables is used to es-timate soil moisture and vegetation water content simultaneously(Koike et al., 2004). At first, a forward radiative transfer scheme isused to generate brightness temperatures for awide range of parametervalues (soils and vegetation) for multiple frequencies and polarizations.Then, the data sets of brightness temperatures are used to create look-up tables. Finally, soil moisture and vegetation water contents are esti-mated by utilizing MPDI at 10.65 GHz and the index of soil wetness(ISW) (Koike, Tsukamoto, Kumakura, & Lu, 1996) at 36.5 GHz and10.65 GHz horizontal channels. The LPRM is a three-parameter (soilmoisture, vegetation optical depth, and surface temperature) retrieval

isture (cm) Time step (min) Land use Date period

60 Grassland 2002.10.01–2004.03.3115 Grassland 2008.07.01–2010.07.3130 Grassland 2010.08.01–2012.12.31

http://www.vito-eodata.be

Fig. 2. Temporal variability of averaged NDVI of three network regions from 1 January 2002 to 31 December 2012.


model. In the algorithm, the vegetation optical depth is expressed as afunction of the soil dielectric constant and the MPDI to avoid a relianceon additional vegetation data sets (Meesters, de Jeu, & Owe, 2005). Theland surface temperature is derived independently from the verticallypolarized Ka-band observations of AMSR-E by using an empirical re-gressionmodel (Holmes, de Jeu, Owe, & Dolman, 2009). Thus, soil mois-ture is the only unknown parameter in the radiative transfer equationand can be finally obtained through a nonlinear iterative procedure. Inaddition, it should be noted that the LPRM releases two productswhich use C-band and X-band brightness temperature observationsfor soil moisture retrieval, respectively. Since there is no apparentradio-frequency interference (RFI) at C-band over the Tibetan Plateauas well as less signal attenuation from vegetation and atmosphere atC-band than at X-band (Njoku, Ashcroft, Chan, & Li, 2005; Zeng, Li,Chen, & Bi, 2014), we use the LPRM soil moisture product derivedfrom C-band observations in the study.

2.2.2. AMSR2AMSR2 boarded on Global Change Observation Mission (GCOM)-

W1 satellite was launched by JAXA on 18 May 2012, and has startedits scientific observation since 3 July 2012 (Imaoka et al., 2010). It is asuccessor of AMSR-E on theNASA's Aqua satellite, and continues the ob-servations of AMSR-E and constructs the long-term data accumulation(Imaoka et al., 2012). The ascending and descending overpasses ofAMSR2 are the same as those of AMSR-E and the basic instrument con-cept of AMSR2 is almost identical to that of AMSR-E as well. However, ithas several improvements over AMSR-E, including largermain reflector(2.0 m), additional channels at the C-band frequency band (7.3 GHz),improved calibration system, and increased reliability (Imaoka et al.,2012). The soil moisture retrieval algorithm of AMSR2 is the same asthat of JAXA AMSR-E. In the study, the AMSR2 Level 3 soil moistureproduct (version 1.11, hereafter called AMSR2 product) was used,which can be obtained through GCOM-W1 Data Providing Service(https://gcom-w1.jaxa.jp/auth.html).

2.2.3. ASCATThe ASCAT is a real aperture backscatter radar and was launched in

October 2006 on the Meteorological Operation-A (MetOp-A) satellite(Wagner et al., 2013). It operates in C-band (5.255 GHz) at VV polariza-tion and scans the Earth's surface in a descending (9:30 local solar time)and ascending (21:30 local solar time) mode. Soil moisture is retrievedfrom the ASCAT backscatter measurements using a time series-basedchange detection method developed at the Vienna University of

Technology (TU-Wien) by Wagner et al. (1999) and Naeimi et al.(2009). In the study, the ASCAT soil moisture productwe used is the up-dated version based on 2 years (2007–2008) of ASCAT observations(version TU-Wien-WARP 5.5, hereafter called ASCAT product) fromthe European Organisation for the Exploitation of Meteorological Satel-lites (EUMETSAT) (Wagner et al., 2010). This product is the relativemeasure of surface soil moisture ranging from 0 (dry) to 100% (wet) ob-tained by scaling normalized backscatter between the historically low-est and highest values. In order to be compatible with other soilmoisture products as well as the in-situ measurements, this product isrescaled to volumetric soil moisture by using the soil database fromthe Food and Agriculture Organization (FAO) (Reynolds, Jackson, &Rawls, 2000). For more details about the ASCAT product, readers arekindly referred to a comprehensive review written by Wagner et al.(2013).

2.2.4. ECVThe ECV soil moisture product is the first purely multi-decadal

satellite-based soil moisture product that spans over 35 years (fromNo-vember 1978 to December 2013) on a daily basis and at a spatial resolu-tion of 0.25°. It was developed as part of the European Space Agency's(ESA) Water Cycle Multimission Observation Strategy (WACMOS) andSoil Moisture Climate Change Initiative (CCI) projects (Dorigo et al.,2014; Liu et al., 2012; Wagner et al., 2012). The ECV soil moisture prod-uct was generated using active and passive soil moisture products,which were derived from the Scanning Multichannel MicrowaveRadiometer (SMMR) onboard Nimbus-7, the Special Sensor MicrowaveImager (SSM/I) of the Defense Meteorological Satellite Program, theTropical Rainfall Measuring Mission Microwave Imager (TMI), theAMSR-E onboard the Aqua satellite, the WindSat satellite, and theAMSR2 boarded on the GCOM-W1 satellite for the passive data sets,and the scatterometers (SCAT) onboard the European Remote Sensingsatellites (ERS-1/2) and the ASCAT onboard the MetOp-A satellite forthe active data sets (Liu et al., 2012). The soil moisture was estimatedby using the LPRM (Owe et al., 2008) for passive sensors and usingthe change detection algorithm (Naeimi et al., 2009; Wagner et al.,1999) for active sensors. The first version of ECV product is ECV SM01.0 which is released in June 2012 and covers the 32 year periodfrom 1978 to 2010. Recently, ESA released the new version of ECV prod-uct that is the ECV SM 02.0. The ECV SM 02.0 has been extended to theyear 2013 by including Windsat and AMSR-2 data. Besides three moreyears of data, the revised product (ECV SM 02.0) includes improvedgap filling, new data attributes, and a revision of processing algorithms

https://gcom-w1.jaxa.jp/auth.html


and merging procedures (http://www.esa-soilmoisture-cci.org/node/185). The new ECV SM 02.0 product consists of three surface soil mois-ture data sets: the “active product” created by fusing scatterometer soilmoisture products, the “passive product” created by fusing passive soilmoisture products, and the “combined product” blended based on the“active product” and the “passive product”. In the study, the “combinedproduct” (version 02.0, hereafter called ECV product) was used. Formore details about the data harmonization procedure, readers are kind-ly referred to Liu et al. (2011, 2012) and Wagner et al. (2012).

2.2.5. SMOSAs the first mission specifically designed for soil moisture monitor-

ing, the SMOS satellite was successfully launched in 2 November 2009and it has become a useful tool for monitoring key elements of the glob-al water cycle (Jacquette et al., 2010; Kerr et al., 2010). SMOS providesglobal coverage at the equator every three days with amorning ascend-ing orbit at 06:00 local solar time and an afternoon descending orbit at18:00 local solar time (Kerr et al., 2012). The forward model usedfor SMOS soil moisture retrieval is the L-band microwave emissionof the biosphere (L-MEB) model (Wigneron et al., 2007), which isbased on a zero-order radiative transfer equation (Mo, Choudhury,Schmugge,Wang, & Jackson, 1982). Soilmoisture and vegetation opticaldepths are retrieved simultaneously from multiangular observations ofSMOS by minimizing a cost function using a generalized least-squaresiterative algorithm (Wigneron et al., 2007). In the study, the SMOSLevel 3 daily soil moisture product (version 2.45, hereafter calledSMOS product) from CATDS (Centre Aval de Traitement des DonnéesSMOS) (Jacquette et al., 2010) was used.

2.2.6. ERA-InterimERA-Interim is the latest global atmospheric reanalysis product,

whichwas produced by the EuropeanCentre forMediumRangeWeath-er Forecasts (ECMWF) (Dee et al., 2011). It covers the period from 1 Jan-uary 1979 to present and continues in near real time. It is based on avariational data assimilation system which assimilates various types ofobservations including satellite and ground based measurements in aconsistent framework (Dee et al., 2011). Soil moisture is a prognosticvariable in ERA-Interim and is provided at 00:00, 06:00, 12:00and 18:00 UTC. ERA-Interim considers four layers of soil (0–7, 7–28,28–100, and 100–289 cm). Additionally, it should be noted that anothernew reanalysis product, the ERA-Interim/Land, offers several parame-terization improvements in the land surface scheme with respect tothe ERA-Interim dataset (Balsamo et al., 2013). However, the ERA-Interim/Land product only covers the period from 1979 to 2010 whichcannot cover the observation period of Naqu network, while the ERA-Interim product can be provided in near real time. In order to makefull use of the valuable measurements in the Tibetan Plateau to drawmore reliable conclusions, we used 0.25° daily averaged ERA-Interimsoilmoisture simulations of the upper layer (0–7 cm) (version 2.0, here-after called ERA product) for inter-comparison and evaluation in thestudy.

3. Methods

First, the NASA, SMOS and ASCAT soil moisture products wereresampled to the spatial resolution of 0.25° to match with other prod-ucts. Additionally, in order to reduce the inherent uncertainties in thescale difference between in-situ points and satellite pixels, the averagesurface soil moisture measurements of all sites of each network werecompared with the soil moisture products which were also averagedover all grids within each network. The same technique has been usedby many previous studies for validation purpose (Jackson et al., 2010,2012; Leroux et al., 2014; Su, de Rosnay, et al., 2013; Su et al., 2011). Be-sides, for the purpose of evaluating the performance of soil moisture re-trieval algorithms at different satellite overpasses, we classified the soilmoisture products according to their acquisition time that is divided

into nighttime and daytime. The nighttime corresponds to the descend-ing overpass of AMSRE and AMSR2, and the ascending overpass ofSMOS and ASCAT. The daytime corresponds to the ascending overpassof AMSRE and AMSR2, and the descending overpass of SMOS andASCAT. It should be noted that the ECV and ERA products used in thestudy are on a daily basis, and they were classified into the nighttimegroup in order to facilitate the comparison. The satellite-based productsare expected to have better accuracy during nighttime because thenear-surface soil moisture and temperature profiles are more uniformduring nighttime than during daytime, which is more consistent withthe basic assumption made in the algorithms. Four commonly usederrormetrics, the RMSE, themean bias (Bias), the correlation coefficient(R), and the unbiased RMSE (ubRMSE) as defined by Entekhabi, Reichl,Koster, and Crow (2010) were computed to quantify the level of agree-ment between the remotely sensed and reanalysis soil moisture prod-ucts and the in-situ measurements. The p-value was also used todetermine the significance level. If the p-value is small (e.g. less than0.05), it means that the correlation is not a coincidence (Albergelet al., 2011). Moreover, the Taylor diagram offers a good way of graph-ically describing how closely a set of patterns matches observations(Taylor, 2001). Since it gives a good view of the performance of differentproducts in a single diagram by considering three statistics includingcorrelation coefficient, centered (unbiased) RMSE and standard devia-tion, we also added the Taylor diagrams for analysis. In theory, the sat-ellite soil moisture retrieval algorithms cannot estimate soil moistureaccurately under frozen soil conditions (Wigneron et al., 2007), becausesoil freezing can lead to significant changes in the soil dielectric constantwhich would further result in the inapplicability of the soil dielectricmodel in this case (Dobson, Ulaby, Hallikainen, & El-Rayes, 1985).Therefore, the conclusions are mainly derived from the evaluation dur-ing unfrozen season in this study, although we still evaluate the prod-ucts during frozen season. Besides, it should be noted that though theremotely sensed datasets are associated with flags that can be used tofilter out the soilmoisture products,we evaluated the complete datasetsin order to make a fair comparison of them and the results can also becompared with previous similar work (Chen et al., 2013; Su, deRosnay, et al., 2013; Su et al., 2011). Moreover, for the frozen period,the ERA-Interim soil moisture product provides both liquid and solidwater while the sensor only measures liquid water. Therefore, theERA-Interim soil moisture product is not assessed during the frozenseason.

4. Results

In this section, we will analyze the results of each network respec-tively. For each network, firstly, we examined the temporal behaviorof observed and estimated soil moisture during the entire period,which can reveal how the soil moisture products perform seasonallyand annually. The complementary information (precipitation data)was also added to each time series, shown in Fig. 3. Four error metrics(RMSE, Bias, ubRMSE and R) used to quantify the accuracy of the eightsoil moisture products were computed separately for nighttime anddaytime, summarized in Tables 2, 3 and 4. To further investigate thetrue skill of each satellite algorithm, we examined the scatter plots ofobserved and estimated soil moisture only during the unfrozen seasonto make a clearer view of the consistency between them, shown inFig. 4. The Taylor diagrams during the unfrozen season were alsoshown in Fig. 5. Moreover, the abovementioned four error metrics ofeach network during the unfrozen period were also displayed inFigs. 6 to 9, respectively.

4.1. CAMP/Tibet network results

The precipitation varies significantly in the CAMP/Tibet network re-gion, and most of the precipitation occurs during the monsoon season(from May to October). As can be seen in Fig. 3(a) and (b), the in-situ

http://www.esa-soilmoisture-cci.org/node/185

http://www.esa-soilmoisture-cci.org/node/185

Fig. 3. Time series of the station-averaged and estimated soil moisture for nighttime (a) CAMP/Tibet, (c) Maqu, and (e) Naqu, and for daytime (b) CAMP/Tibet, (d) Maqu, and (f) Naqu during the entire period.

96J.Zeng

etal./Remote

SensingofEnvironm

ent163

(2015)91

–110

Fig.

3(con

tinue

d).


Fig. 3 (continued).


soil moisture responds to variation in the precipitationwell and there isa good agreement between them. Overall, the NASA product cannotcapture the soil moisture dynamics, and it almost maintains a constantvalue around 0.1 m3 m−3 during the entire period. The other productscan well reflect the soil moisture dynamic range during the unfrozenperiod but show different seasonal amplitudes. As mentioned above,the satellite soil moisture algorithms cannot accurately estimate soil

Table 2Errormetrics of soilmoisture for CAMP/Tibet network. RMSE (rootmean square error), Bias (mep-value is used to determine the significance level and NSmeans no significance for p-value greconcerning each error metric.

Period Products

RMSE Bias ubRMSE R p-V

Unfrozen season ECV 0.060 −0.054 0.026 0.763 0LPRM 0.152 0.127 0.084 0.634 0NASA 0.121 −0.116 0.034 0.498 0JAXA 0.123 −0.114 0.046 0.657 0ERA 0.090 0.085 0.030 0.746 0

Frozen season ECV 0.040 −0.005 0.040 −0.265 0LPRM – – – –

NASA 0.026 0.004 0.026 −0.033JAXA 0.058 −0.053 0.024 0.367 0ERA – – – –

moisture under frozen soil conditions due to the great changes in thesoil dielectric constant. Thus, these algorithms use different criteria(land use or temperature) to exclude unreliable values in this case.The LPRM uses an empirical regression model to estimate the land sur-face temperature, which is used to distinguish the unfrozen and frozenconditions (Holmes et al., 2009). However, Zeng et al. (2015) recentlyfound that the surface temperature derived from LPRM is not very

an bias) and ubRMSE (unbiased RMSE) are both inm3m−3, R is the correlation coefficient,ater than 0.05, N is the number of samples. Bold data in the table represent the best results

Nighttime Daytime

alue N RMSE Bias ubRMSE R p-Value N

.000 161 – – – – – –

.000 163 0.093 0.084 0.040 0.559 0.000 182

.000 186 0.107 −0.100 0.038 0.228 0.002 186

.000 172 0.094 −0.069 0.064 0.607 0.000 172

.000 215 – – – – – –

.003 65 – – – – – –

– – 0.130 0.125 0.036 −0.081 NS 265NS 285 0.025 0.004 0.025 0.281 0.000 285

.000 265 0.046 −0.039 0.024 0.378 0.000 265– – – – – – – –

Table 3Error metrics of soil moisture for Maqu network. RMSE (root mean square error), Bias (mean bias) and ubRMSE (unbiased RMSE) are both in m3 m−3, R is the correlation coefficient, p-value is used to determine the significance level and NS means no significance for p-value greater than 0.05, N is the number of samples. Bold data in the table represent the best resultsconcerning each error metric.

Period Products Nighttime Daytime

RMSE Bias ubRMSE R p-Value N RMSE Bias ubRMSE R p-Value N

Unfrozen season ECV 0.182 −0.177 0.042 0.695 0.000 458 – – – – – –

ASCAT 0.114 0.100 0.055 0.704 0.000 250 0.096 0.088 0.038 0.797 0.000 219LPRM 0.102 0.072 0.072 0.348 0.000 343 0.083 0.036 0.075 0.227 0.000 342NASA 0.239 −0.233 0.053 0.421 0.000 343 0.209 −0.201 0.057 0.225 0.000 342JAXA 0.296 −0.287 0.072 0.194 0.000 343 0.219 −0.209 0.065 0.482 0.000 341SMOS 0.140 −0.032 0.136 0.241 NS 60 0.365 0.254 0.262 0.198 NS 57ERA 0.077 −0.060 0.048 0.602 0.000 459 – – – – – –

Frozen season ECV 0.102 −0.073 0.071 0.705 0.000 194 – – – – – –

ASCAT – – – – – – – – – – – –

LPRM 0.085 0.062 0.058 0.681 0.000 180 0.247 0.227 0.097 0.160 0.016 226NASA 0.099 −0.067 0.073 0.419 0.000 224 0.073 −0.044 0.058 0.696 0.000 226JAXA 0.170 −0.153 0.074 0.451 0.000 224 0.137 −0.124 0.058 0.733 0.000 226SMOS – – – – – – – – – – – –

ERA – – – – – – – – – – – –


accurate in the Tibetan Plateau. Specifically, the LPRM underestimatesthe surface temperature at AMSR-E descending overpass while it over-estimates the surface temperature at AMSR-E ascending overpass inthe Tibetan Plateau. As a result, the LPRM cannot exclude the unreliableretrievals during frozen season at daytime that corresponds to AMSR-Eascending overpass, shown in Fig. 3(b). The NASA and JAXA algorithmsdo not seem to have an effective criterion, resulting in no data being ex-cluded under the frozen soil conditions. On the other hand, the LPRMoverestimates the soilmoisturewith the largest Bias value for the unfro-zen period,which can be seen in Table 2 and Fig. 7. The ERA product alsooverestimates the soil moisture in the unfrozen period but displays highcorrelation with groundmeasurements. In contrast, the NASA and JAXAproducts underestimate soil moisture with a large negative Bias atmostof the time. The ECV product performs very well in the CAMP/Tibet net-work region with the lowest RMSE, Bias and ubRMSE values as well asthe highest R value. The Taylor diagrams shown in Fig. 5(a) and(b) also illustrate that the variability of ECV product is in a very goodagreement with that of in-situ measurements, whereas the variabilityof the ERA and NASA products is lower than that of in-situ measure-ments and the LPRM and JAXA products show higher variability thanground observations. Meanwhile, it should be noted that the ECV prod-uct was scaled to a reference climatology provided by the Global Land

Table 4Error metrics of soil moisture for Naqu network. RMSE (root mean square error), Bias (meanp-value is used to determine the significance level and NS means no significance for p-valueresults concerning each error metric.

Period Products

RMSE Bias ubRMSE R p-V

Unfrozen season ECV 0.098 −0.092 0.034 0.852 0ASCAT 0.116 0.103 0.053 0.810 0LPRM 0.137 0.121 0.064 0.838 0NASA 0.181 −0.173 0.053 0.620 0JAXA 0.114 −0.035 0.108 0.747 0

AMSR2 0.127 −0.050 0.117 0.788 0SMOS 0.125 −0.073 0.101 0.544 0ERA 0.070 0.050 0.049 0.618 0

Frozen season ECV 0.055 −0.032 0.045 0.727 0ASCAT – – – –

LPRM – – – –

NASA 0.044 0.013 0.042 −0.034JAXA 0.051 −0.034 0.038 0.684 0

AMSR2 0.052 −0.048 0.020 0.699 0SMOS 0.103 −0.093 0.044 0.417 0ERA – – – –

Data Assimilation System (GLDAS)-Noah surface soil moisture product(Rodell et al., 2004) during the data harmonization procedure (Liuet al., 2011, 2012; Wagner et al., 2012). Therefore, the ECV producthas the climatology of GLDAS-Noah model, so the absolute error ofECV product is dependent on Noah model but not on satellite sensors(Loew, Stacke, Dorigo, de Jeu, & Hagemann, 2013). However, althoughthis procedure affects the absolute soil moisture values, temporal varia-tions and trends of the original datasets are well preserved (Dorigoet al., 2015; Liu et al., 2012). Additionally, according to the RMSE valuesshown in Table 2 and Fig. 6, the LPRM, NASA and JAXA productshave better accuracy during daytime than during nighttime. It conflictswith the general expectation that the satellite-based retrievals shouldbe more accurate during nighttime since near-surface soil moistureand temperature profiles are more uniform and the temperaturedifference between the ground surface and the canopy is smaller duringnighttime than during daytime, which are supposed to be basic as-sumptions made in the soil moisture retrieval algorithms. Some previ-ous work also yielded similar results (Brocca et al., 2011; Chen et al.,2013), and Brocca et al. (2011) argued the reason may be that thehigher temperature during daytime makes the vegetation more trans-parent, resulting in less effect of vegetation on emission attenuationfrom soils.

bias) and ubRMSE (unbiased RMSE) are both in m3 m−3, R is the correlation coefficient,greater than 0.05, N is the number of samples. Bold data in the table represent the best

Nighttime Daytime

alue N RMSE Bias ubRMSE R p-Value N

.000 432 – – – – – –

.000 210 0.112 0.105 0.039 0.890 0.000 216

.000 181 0.095 0.089 0.033 0.833 0.000 184

.000 182 0.160 −0.153 0.047 0.683 0.000 184

.000 167 0.103 0.026 0.100 0.854 0.000 168

.000 91 0.124 0.018 0.123 0.885 0.000 89

.000 231 0.138 0.003 0.138 0.434 0.000 214

.000 460 – – – – – –

.000 100 – – – – – –

– – – – – – – –

– – 0.196 0.192 0.039 0.445 0.000 136NS 138 0.041 0.017 0.037 0.441 0.000 136

.000 125 0.037 −0.018 0.032 0.726 0.000 125

.000 46 0.041 −0.037 0.177 0.785 0.000 45

.000 76 0.065 −0.018 0.062 0.159 NS 111– – – – – – – –


4.2. Maqu network results

Asmentioned above, theMaqu network region has the densest veg-etation cover, and the NDVI ranges from 0.2 to 0.8 for the unfrozen sea-son. Therefore, it can be expected that this network would involvemaximal vegetation effects. Though the ECV product still underesti-mates the ground measurements with a relatively large negative Biasvalue for the unfrozen period, it is highly correlated with the groundmeasurements which is consistent with previous studies that themerged product can preserve the correlation coefficient very well(Dorigo et al., 2015; Liu et al., 2012). Moreover, it is particularly note-worthy that the number of observations of the ECV product is

Fig. 4. Scatter plots of the station-averaged and estimated soil moisture for nighttime (a) CAMduring the unfrozen period.

significantly improved compared with other microwave remotelysensed soil moisture products, which benefited from merging both theactive and passive datasets. This will be more beneficial to practical ap-plications since many of them require frequent soil moisture measure-ments. Due to the predictable vegetation effects in Maqu networkregion, all passive microwave remotely sensed soil moisture productsexcept for the LPRM at daytime have a large RMSE value more than0.1 m3 m−3 and a low correlation coefficient during the unfrozen sea-son. In contrast, though the active microwave satellite-based ASCATproduct overestimates the in-situ data, it shows much higher correla-tion (greater than 0.7) with the groundmeasurements than the passivemicrowave remotely sensed soil moisture products. It is in consensus

P/Tibet, (c) Maqu, and (e) Naqu, and for daytime (b) CAMP/Tibet, (d) Maqu, and (f) Naqu

Fig. 4 (continued).


with previous studies that theASCAT product is less sensitive to the veg-etation cover (Matgen et al., 2012; Qiu et al., 2013). The main reasonmay be that the change detection algorithm used for ASCAT can sepa-rate the seasonal vegetation cycle process from shorter time scale soilmoisture variation contained in the backscattered signal time serieswhich leads to more readily detection of the dynamics of the surfacesoil moisture (Qiu et al., 2013). In terms of the absolute accuracy, theERA product outperforms other six soilmoisture productswith the low-est RMSE value of 0.077m3m−3 and a relatively higher R value of 0.602,even though it underestimates the ground measurements with a nega-tive Bias. This can be explained by the fact that the accuracy of model-based product is mainly influenced by atmospheric forcing data andmodel parameters rather than by vegetation (Bi, Ma, & Wang, 2014;

Moradkhani, Hsu, Gupta, & Sorooshian, 2005), but it still shows lowervariation than in-situ data which can be clearly seen from the Taylor di-agram shown in Fig. 5(c). The LPRM, NASA, JAXA and ASCAT productsperform still somewhat better during daytime than during nighttimefor the unfrozen period.While the SMOS product performsmuch betterduring nighttime with much lower RMSE, Bias and ubRMSE values,which can be clearly seen from Figs. 6 to 8 and Table 3. But the resultsstill fall out of the acceptable range. In theory, the SMOS satellite canprovide the most reliable soil moisture, since the effects of vegetationand atmosphere can be minimized by using L-band frequency. Unfortu-nately, the SMOS brightness temperature measurements have been se-verely affected by the well-known RFI in some parts of the world,including large parts of Europe, China, Southern Asia, and the Middle

Fig. 4 (continued).


East (Oliva et al., 2012). The RFI disturbs the natural microwave emis-sion receivedby SMOS, thus resulting in unreliable soilmoisture estima-tions. The phenomenon that the SMOS observations have been affectedby RFI at the Tibetan Plateau has also been found by Dente, Su, andWen(2012) recently. Moreover, it seems that the occurrence probabilities ofRFI are higher at SMOS descending overpass than at ascending overpassin Maqu network region because of the highly undesirable results ob-tained during daytime. It is consistent with Leroux et al.'s (2014) workwho also found that the SMOS observations at descending overpassaremore likely to be contaminated by RFI. The JAXA and NASA productsperform theworstwith a RMSE value greater than 0.2m3m−3 and theyobviously underestimate the soil moisture for the unfrozen period. Ad-ditionally, it is unexpected that the LPRM product performs somewhatbetter at Maqu network region than at other two network regions interms of RMSE value for the unfrozen period, while other remotelysensed soil moisture products perform the worst at Maqu network re-gion. It may indicate that the LPRM can provide more accurate absolutevolumetric soilmoisture in cold humid climate area than in cold semiar-id climate area. For the frozen period, the LPRM product unexpectedlyshows the best conformity with the ground data during nighttime. Buton average, the NASA product is still closest to the in-situ data whilethe JAXA product displays the best correlation with the observations.

4.3. Naqu network results

The Naqu network is the latest network established in the TibetanPlateau, so its observation period covers the latest AMSR2 mission. Ithas moderately low aboveground biomass with NDVI ranging from 0.1to 0.55 for the unfrozen period. As mentioned above, the skill of all re-motely sensed soil moisture products is improved at Naqu network re-gion than at Maqu network region with a lower RMSE value and ahigher R value except for the LPRM product. The ERA product is stillclosest to the absolute ground measurements with the lowest RMSEvalue for the unfrozen period. Moreover, compared with the results ofother two network regions, it seems that the ERA product appears tooverestimate the soil moisture in cold semiarid climate area corre-sponding to the CAMP/Tibet and Naqu network regions, whereas it

underestimates the soil moisture in cold humid climate area corre-sponding to Maqu network region. The ECV product can capture thesoil moisture dynamics very well and exhibits similar variation within-situ data. It has a high correlation coefficient of 0.852 and a smallubRMSE value of 0.034 m3 m−3 indicating that it is highly related toin-situ observations. The ASCAT product still overestimates the groundmeasurements but performs very well in terms of the correlation, andit even shows the highest R value of 0.890 during daytime. The perfor-mance of LPRM product is comparable with the ASCAT product. It alsosystematically overestimates soil moisture with a large positive Biasvalue, but it is highly correlated with ground measurements in the un-frozen period, which can be obviously seen from Table 4, Figs. 7 and 9.The NASA product performs the worst with the highest RMSE and Biasvalue. It evidently underestimates soil moisture and still shows verylow variation comparedwith the groundmeasurements in the unfrozenperiod. The results of JAXA and AMSR2 products are about the sameorder of magnitude though their evaluation period is not consistent.This is expected since the AMSR2 is a successor of AMSR-E used to con-tinue AMSR-E observations and the soil moisture algorithm is notchanged. Although their absolute accuracy is still not very satisfactoryand they exhibit higher variation than in-situ measurements, theyshow favorable correlation with ground data. The AMSR2 producteven gives an R value up to 0.885 during daytime, indicating that theAMSR2 sensor has the potential to continue the long-term and large-scale soil moisture accumulation. The SMOS product is still very likelyto be contaminated by RFI with the lowest R value, and it performsslightly better during nighttime than during daytime. In contrast, the re-sults of other remotely sensed soil moisture products (NASA, LPRM,JAXA, AMSR2 and ASCAT) all perform slightly better during daytimethan during nighttime, which is consistent with the results found atCAMP/Tibet and Maqu network regions. For the frozen season, theECV productmisjudged the unfrozen and frozen soil conditions, but un-expectedly shows the best correlation with the ground measurements.The NASA, JAXA and AMSR2 products perform almost at the samelevel but better than the SMOS product in terms of RMSE value. Howev-er, the NASA product gives unrealistic R valuewhereas that of JAXA andAMSR2 products is favorable.

Fig. 5. Taylor diagram illustrating the statistics of the comparison between station-averaged and estimated soil moisture for nighttime (a) CAMP/Tibet, (c) Maqu, and (e) Naqu, and fordaytime (b) CAMP/Tibet, (d) Maqu, and (f) Naqu during the unfrozen period.


4.4. Summary performance

Based upon the results from the three networks in the period of2002–2012, the inferences for the soil moisture products used in thestudy are summarized. Moreover, since the satellite soil moisture algo-rithms cannot accurately retrieval soil moisture under frozen soil condi-tions due to the great changes in the soil dielectric constant, thefollowing conclusions are only made for the unfrozen period.

(1) With the exception of LPRM, all of the other remotely sensed soilmoisture products give better results at CAMP/Tibet and Naqunetwork regions than at Maqu network region which has thedensest vegetation cover, resulting in heavy emission attenua-tion from soils.

(2) The ECV product systematically underestimates the soil moistureat all network regions, but it shows very high correlation with

ground observations and can reflect the seasonal variation ofsoil moisture. The ASCAT product is highly correlated withground measurements with an average R value of 0.751 and0.850 at Maqu and Naqu network regions respectively and isalso less affected by the vegetation, though it overestimates thesoil moisture with a positive Bias. The LPRM can give relativelyreasonable estimations in terms of R value, but it always routine-ly estimates very high soil moisture that produces large positiveBias. In contrast, the NASA algorithm significantly underesti-mates soilmoisture and also cannot capture the soil moisture dy-namics. The JAXA and AMSR2 products perform better than theNASA product but still have an underestimation Bias at most ofthe time. The ERA product can achieve the lowest averageRMSE value of 0.079 m3m−3 at the three network regions, butit shows lower variation than in-situ data and it seems to overes-timate soil moisture at CAMP/Tibet and Naqu network regions

Fig. 5 (continued).


which both are in a cold semiarid climate, while it underesti-mates soil moisture at Maqu network which is in a cold humidclimate.

(3) The results of ASCAT, LPRM, NASA, JAXA and AMSR2 products areslightly better during daytime than during nighttime, indicatingthat the daytime overpass may have a positive effect overthe Tibetan Plateau. The SMOS observations are very likelyto be contaminated by the RFI at the Tibetan Plateau whichleads to unfavorable results, and the occurrence probabilities ofRFI are higher at SMOS descending overpass than at ascendingoverpass.

(4) Overall, both ECV and ERA products have better performanceagainst ground observations than other soil moisture products.Specifically, the ECV product is highly related to in-situ datawith the highest R value and the smallest ubRMSE value. Itshows similar variation with in-situ data and the number of ob-servations is also greatly improved. The ERA product is closestto the absolute ground measurements with the lowest RMSEvalue. TheNASA product performs theworst with the largest un-derestimation Bias and also lacks temporal dynamics.

5. Discussion

As the results shown above, although most of the remotely sensedsoil moisture products can well reflect the seasonal variation of groundsoilmoisture, someof themdonot provide expectant accuracy. It is verynecessary to investigate the error source of the soil moisture retrievals,and the errors may be the results of combined effects of many factors,especially in the Tibetan Plateau with very complex underlying surface.Possible explanations for the errors are as follows: 1) different spatialobservation scales between in-situ points and satellite pixels. Althoughgroundmeasurements fromdense stations are used tominimize the ef-fect of thiswell-known issue, thedifferences in spatial resolution still in-troduce some deviation, especially for the CAMP/Tibet network withless stations and larger area than the other two networks. Since nowa-days there are no ground soil moisture observations which can exactlyrepresent the same scale as that of the satellite observations, we haveto use the average point-based measurements as the “ground truth”.

However, some studies pointed out that spatial soil moisture patternsare difficult to characterize using in-situ measurements and few pointmeasurements are only able to reproduce the temporal dynamic ofsoil moisture, but not the absolute values (Koster et al., 2009; Wagneret al., 2013). Therefore, the readers should not only focus on the RMSEor Bias value, but also pay more attention to the correlation. Moreover,to further relieve the scale-related issue, the point soil moisturemeasurements can be upscaled to a coarse scale using some spatialupscaling methods proposed recently (Loew & Schlenz, 2011;Qin et al., 2013). 2) A mismatch between in-situ soil moisture measur-ing depth and the microwave penetration depth. In the study, the sur-face ground measurements used for validation are at 4 cm, 5 cmand 0–5 cm for CAMP/Tibet, Maqu and Naqu networks respectively.While the effective soil moisture sampling depth at L-band and C-bandis 0–3 cm and 0–1 cm respectively (Al-Yaari, Wigneron, Ducharne,Kerr, de Rosnay, et al., 2014), and it also varies according to the soilmois-ture content (Escorihuela, Chanzy, Wigneron, & Kerr, 2010; Owe et al.,2008). A recent study also found that the effective soilmoisture samplingdepth even at L-band is shallower than provided by widely used in-situmoisture sensors (Escorihuela et al., 2010). Thus, comparedwith groundobservations, satellite-based retrievals aremore sensitive to atmosphericforcings such as precipitation, radiation and wind, which will lead to thephenomenon that the remotely sensed products usually exhibit highertemporal variability than the in-situ measurements (Wagner et al.,2007). 3) Errors in the in-situ data caused by measurement accuracy ofsensors (Dorigo et al., 2013). 4) Inaccuracies in the retrieval algorithm(making some assumptions for simplicity) and related inputs (such asthe soil texture and land surface temperature data).

Among them, the fourth point is themain reason that leads to differ-ent performances of the soil moisture algorithms. The passive micro-wave satellite-based soil moisture (NASA, LPRM, JAXA, and SMOS)algorithms are based on the same principles and radiative transferequation. Specifically, they all use a zero-order radiative transfermodel that is the so called τ–ω model (Mo et al., 1982) as the forwardmodel to represent the electromagnetic radiation from the Earth's sur-face. However, except for soil moisture, the radiative transfer equationdemands estimation of many other parameters including vegetationproperties (vegetation optical depth and single scattering albedo), sur-face roughness and temperature. The difference between each

Fig. 6. RMSE regarding comparison between station-averaged measurements and soilmoisture products in each network for (a) nighttime and (b) daytime during the unfrozenperiod.

Fig. 7. Bias regarding comparison between station-averagedmeasurements and soil mois-ture products in each network for (a) nighttime and (b) daytime during the unfrozenperiod.


algorithm is how they determine these parameters to get the best solu-tion of soil moisture. The JAXA algorithm assumes that the vegetationoptical depth is linearly related to vegetation water content which canbe determined by theNDVI fromoptical data. However, optical observa-tions are often influenced by clouds and atmospheric conditions, whichsometimes hamper the use of the algorithm. In contrast, the LPRM doesnot need any ancillary data, thus is considered to have the greatest po-tential to estimate soil moisture on a global scale. However, the estima-tion of surface temperature parameter in LPRM is from a globalempirical regression model, which is developed based on the physicalbasis that the Ka-band (36.5 GHz) vertical brightness temperature ishighly sensitive to the surface temperature (de Jeu & Owe, 2003;Holmes et al., 2009; Owe & Van de Griend, 2001; Salama et al., 2012;Zeng et al., 2015). But the ground measurements that used to establishthe empirical model are mainly from the United States and someEuropean countries. Therefore, it may have regional dependenceand be inapplicable outside these regions, especially in the TibetanPlateau with its quite different geographical environment and climate

conditions. A recent study also validated the accuracy of the land surfacetemperature derived from LPRM at the CAMP/Tibet, Maqu and Naqunetwork regions (Zeng et al., 2015). The results showed that the LPRMtemperature retrievals are not very accurate in the Tibetan Plateauand sometimes even bring large errors. Moreover, only land skin tem-perature information can be obtained from the Ka-band vertical bright-ness temperature observations, while the effective soil temperature ismore important for the retrieval of soil moisture information (Chanzy,Raju, & Wigneron, 1997). The effective soil temperature, which is a su-perposition of the intensities emitted at various depths within the soil(Choudhury, Schmugge, &Mo, 1982), is heavily dependent on soilmois-ture and soil temperature information from deeper layers (Lv, Wen,Zeng, Tian, & Su, 2014). Some researchers have developed two-layer al-gorithms to calculate the effective soil temperaturewhich can eliminatethe influence of temperature more effectively (Holmes et al., 2006; Lvet al., 2014; Wigneron, Chanzy, de Rosnay, Rüdiger, & Calvet, 2008;Wigneron, Laguerre, & Kerr, 2001). Consequently, the inaccurate

Fig. 8. ubRMSE regarding comparison between station-averaged measurements and soilmoisture products in each network for (a) nighttime and (b) daytime during the unfrozenperiod.

Fig. 9. R regarding comparison between station-averaged measurements and soil mois-ture products in each network for (a) nighttime and (b) daytime during the unfrozenperiod.


estimation of land surface temperature must be one factor that leads tothe deviation of LPRM product in the Tibetan Plateau.

Besides, the surface roughness is also a very important parameterthat influences the accuracy of soilmoisture retrieval. Surface roughnessis considered to increase the soil emissivity due to the increase insurface area of the emitting surface (Schmugge, 1985). At present, theQ/H model (Wang & Choudhury, 1981) is the most widely used rough-nessmodel in the soil moisture algorithms. The “Q” and “H” are two pa-rameters in the Q/H model used to parameterize the effects of surfaceroughness. Theoretically, the two roughness parameters can be charac-terized by the surface root mean square (RMS) height and horizontalroughness correlation length (Wang & Choudhury, 1981; Wen et al.,2003; Wigneron et al., 2011). However, there are very few data avail-able to quantify these two parameters globally, especially in some spe-cial areas such as the Tibetan Plateau. Moreover, the effects of incidenceangle and frequency on the roughness parameters have not been stud-ied thoroughly (Holmes, 2003; Owe, de Jeu, & Walker, 2001), either. Asa result, the “Q” and “H” are usually determined empirically. Due to the

lack of auxiliary data to quantify “Q” and “H”, some algorithms such asthe JAXA algorithm, the LPRM, and the SMOS algorithm have to assumethe two parameters as global constant values. However, the global con-stant roughness assumptionmade in the algorithms is inconsistentwiththe actual surface, and thus may introduce retrieval errors sometimes.Su et al. (2011) pointed out that the variability of roughness in the Tibet-an Plateau is quite large. Therefore, the constant roughness parameterassumption is not valid in this area, which can be one reason for the de-viation of JAXA, LPRM and SMOS products. Though somemethods havebeen developed to correct the effects of surface roughness on surface re-flectivity recently (Chen, Shi, Wigneron, & Chen, 2010; Guo, Shi, Liu, &Du, 2013; Shi et al., 2006), none of thesemethods can be applied to veg-etation coverage area directly since they are developed for bare groundsurface.

Furthermore, in the study, the validation results show that the NASAand JAXA algorithms seem to have relatively poorer performance interms of RMSE values. The reason may be that the two algorithms cali-brated somekey parameters and coefficients in the algorithms in specif-ic regions. The JAXA algorithm calibrated the surface roughness and


some vegetation parameters inMongolia, while the NASA algorithm is aglobal empirical regression algorithm and some key coefficients in thealgorithmaremainly calibrated in the United States. Since these param-eters/coefficients which are calibrated in specific regions exert an im-portant role in the soil moisture algorithms, these algorithms mayhave regional dependence and be unsuitable for other regions. Thus,though some previous studies have shown that the JAXA and NASAalgorithms can achieve relatively favorable results in Mongolia(Kaihotsu et al., 2009) and the United States (Jackson et al., 2010) re-spectively, they still cannot provide accurate soil moisture estimationsat the Tibetan Plateau. It is suggested that these parameters/coefficientsshould be recalibrated at the Tibetan Plateau to improve the estimationaccuracy of soil moisture in this region.

Besides, the well-known RFI is also a very important factor that in-fluences the accuracy of soil moisture retrievals. Since AMSR-E is amultifrequency-polarization sensor, the algorithms developed forAMSR-E can switch between C-band and X-band observations to avoidthe RFI. However, it is still a big problem for the SMOS satellite whichcan only provide single bandobservations. A recent study demonstratedthat the RFI had the largest influence over Asia, especially in the centralAsia and East Asia (Leroux, Kerr, Richaume, & Fieuzal, 2013). Therefore,though some exciting results of SMOS have been found in the UnitedSates (Jackson et al., 2012; Leroux et al., 2014) and Australia (Albergelet al., 2012; Su, Ryu, et al., 2013) where observations are rarely contam-inated by RFI at L-band, it cannot get favorable results at the Tibetan Pla-teau. Additionally, the results in the study show that the ASCAT productis better or comparable with the AMSR-E products which is consistentwith Brocca et al.'s (2011) work. The ASCAT product is highly correlatedwith in-situ data but still overestimates the ground measurements.Wagner et al. (2013) pointed out that the deviation of the ASCAT re-trievals may be caused by some imperfections of the ASCAT data pro-cessing over the high-elevation areas. Moreover, in the study, thetexture data derived from FAO is used to convert the ASCAT productfrom degree of saturation (%) into volumetric soil moisture content(m3m−3). However, recent studies found that the soil texture informa-tion from FAO cannot be representative for the Tibetan Plateau(Albergel et al., 2010; Chen et al., 2013; Su, de Rosnay, et al., 2013). Su,de Rosnay, et al. (2013) also found that there is no difference in soil tex-ture of FAO for almost the entire Tibetan Plateau. Thus, the inaccuratesoil texture information should be one factor that leads to the deviationof ASCAT product in the Tibetan Plateau as well as other remotelysensed soil moisture products. The ECV product is the first purelymulti-decadal satellite-based soil moisture dataset. Its harmonizationprocedure is based on a rescaling of the microwave soil moisture datausing the soil moisture statistics from the Noah land surface model(Liu et al., 2012; Loew et al., 2013). Although this procedure affectsthe absolute soil moisture values, temporal variations and trends ofthe original datasets are well preserved (Dorigo et al., 2015; Liu et al.,2012). Some studies also demonstrated that most of applications re-quire the information on soil moisture dynamic rather than its absolutevalue (Koster et al., 2009; Miralles, Crow, & Cosh, 2010). In the study,the results show that ECV product can capture the soilmoisture dynam-ics in the Tibetan Plateau verywell. The results also show that the num-ber of observations of the merged product is significantly improvedthan the individual microwave satellite-based soil moisture productwhich will be of great benefit to practical applications since many ofthem require frequent soil moisture measurements. It is expected thatthe ECV product can be extended with additional observations fromSMOS and the ongoing Soil Moisture Active Passive (SMAP) mission(Entekhabi, Njoku, et al., 2010) to constitute a more comprehensiveproduct.

6. Conclusions

Validation of different remotely sensed and reanalysis soil mois-ture products is very important for the data application and the

improvement of the retrieval algorithms. In this study, a detailed evalu-ation of seven remotely sensed and one reanalysis soil moisture prod-ucts was performed in the period of 2002–2012 over the TibetanPlateauwhich is oneof themost sensitive areas to global climate changeand is also very crucial for understanding the climate system. Theground soil moisture measurements used for validation are from threedense networks named CAMP/Tibet,Maqu andNaqu established in var-ious locations over the Tibetan Plateau. Both the CAMP/Tibet and Naqunetworks are in a cold semiarid climate, and the Maqu network is in acold humid climate. Moreover, each network was located in regions ofdifferent vegetation covers. The Maqu network region has the highestbiomass, followedbyNaqu and CAMP/Tibet network regions. Therefore,the three networks are under different biomes and climate conditionsthat can provide a more robust validation of the soil moisture productsover the Tibetan Plateau.

Overall, except for the NASA product, the seasonal variation of otherproducts is generally in good agreement with the ground measure-ments over the three network regions. It seems that the NASA productcannot capture the soil moisture dynamics and evidently underesti-mates the soil moisture during the entire period. The JAXA AMSR-Eand AMSR2 products perform somewhat better, but still underestimatethe ground measurements at most of the time. The reason for the devi-ation of these three products may be that the key parameters and coef-ficients in the algorithms are calibrated in specific regionswhichmaybeunsuitable for the Tibetan Plateau. The LPRMproduct can get reasonableresults in terms of correlation, but givesmuch larger seasonal amplitudethan in-situ observations, and it routinely overestimates the soil mois-ture with a very large positive Bias. The errors of LPRM are supposedto be mainly caused by the inaccurate surface temperature estimationin the algorithm. The empirical surface temperature model of LPRM isnot very accurate at the Tibetan Plateau and thus more accurate surfacetemperature estimates should be provided to improve the LPRM. Thepresence of RFI is probably themost important factor that causes biasesin the SMOS product. Moreover, the occurrence probabilities of RFI arehigher at SMOS descending overpass than at the ascending overpass.Thus, it seems better to use other sensors such as AMSR2 to monitorsoil moisture at Asia which is the most contaminated region by RFI atL-band. The ASCAT can correlate with in-situ data very well with an av-erage correlation value of 0.751 and 0.850 at Maqu and Naqu networkregions respectively and is also less influenced by the vegetation, butit still overestimates the ground measurements which may be causedby the defects of the ASCAT data processing over the high-elevationareas as well as the uncertainties in available soil texture data. There-fore, with further improvements in processing techniques andobtaining more accurate soil texture data, it can be expected to achievemore high-quality ASCAT retrievals. In general, the ERA and ECV prod-ucts perform the best compared with other products used in thestudy. The ERA product shows the highest absolute accuracy with thelowest RMSE value, while the ECV product best correlates with groundmeasurements with the highest correlation value. Though it still under-estimates the soil moisture, it exhibits similar variation with groundmeasurements and the number of observations is also greatly im-proved. The results confirm the effectiveness of merging active and pas-sive soil moisture products over the Tibetan Plateau, which is verybeneficial to understanding the land–atmosphere interactions of theplateau since the ECV product can provide information on soil moisturedynamics over a period of more than 35 years. It also reveals great po-tential for extending the merged product with additional observationsfrom SMOS and the ongoing SMAPmission to form amore comprehen-sive product.

Acknowledgments

We thank four reviewers for their valuable and detailed commentsthat have greatly improved the paper. We are grateful to Prof. KunYang, Prof. Zhongbo (Bob) Su, and Prof. Yaoming Ma for kindly


providing the in-situ data. We also thank National Snow and Ice DataCenter (NSIDC), Japan Aerospace Exploration Agency (JAXA), VrijeUniversiteit Amsterdam, Centre Aval de Traitement des DonnéesSMOS (CATDS), Vienna University of Technology (TU-Wien),European Organisation for the Exploitation of Meteorological Satellites(EUMETSAT), European Space Agency (ESA) and the European Centrefor Medium-Range Weather Forecasts (ECMWF) for providing the re-motely sensed and reanalysis soil moisture products. This work wassupported by the National Natural Science Foundation of China (grantnumbers: 41471065, 41101391), and the Director Foundation for Post-graduates of Institute of Remote Sensing and Digital Earth ChineseAcademy of Sciences (grant number: Y3ZZ17101B).

References

Al Bitar, A., Leroux, D., Kerr, Y.H., Merlin, O., Richaume, P., Sahoo, A., et al. (2012). Evalu-ation of SMOS soil moisture products over continental US using the SCAN/SNOTELnetwork. IEEE Transactions on Geoscience and Remote Sensing, 50(5), 1572–1586.

Albergel, C., Calvet, J.C., de Rosnay, P., Balsamo, G., Wagner,W., Hasenauer, S., et al. (2010).Cross-evaluation of modelled and remotely sensed surface soil moisture with in situdata in Southwestern France. Hydrology and Earth System Sciences, 14, 2177–2191.

Albergel, C., de Rosnay, P., Gruhier, C., Muñoz-Sabater, J., Hasenauer, S., Isaksen, L., et al.(2012). Evaluation of remotely sensed andmodelled soilmoisture products using glob-al ground-based in situ observations. Remote Sensing of Environment, 118, 215–226.

Albergel, C., Dorigo, W., Reichle, R.H., Balsamo, G., De Rosnay, P., Muñoz-Sabater, J., et al.(2013). Skill and global trend analysis of soil moisture from reanalyses and micro-wave remote sensing. Journal of Hydrometeorology, 14(4), 1259–1277.

Albergel, C., Zakharova, E., Calvet, J.C., Zribi, M., Pardé, M., Wigneron, J.P., et al. (2011). Afirst assessment of the SMOS data in southwestern France using in situ and airbornesoil moisture estimates: The CAROLS airborne campaign. Remote Sensing ofEnvironment, 115(10), 2718–2728.

Al-Yaari, A., Wigneron, J.P., Ducharne, A., Kerr, Y., de Rosnay, P., de Jeu, R., et al. (2014).Global-scale evaluation of two satellite-based passive microwave soil moisturedatasets (SMOS and AMSR-E) with respect to Land Data Assimilation System esti-mates. Remote Sensing of Environment, 149, 181–195.

Al-Yaari, A., Wigneron, J.P., Ducharne, A., Kerr, Y.H., Wagner, W., De Lannoy, G., et al.(2014). Global-scale comparison of passive (SMOS) and active (ASCAT) satellitebased microwave soil moisture retrievals with soil moisture simulations (MERRA-Land). Remote Sensing of Environment, 152, 614–626.

Balsamo, G., Albergel, C., Beljaars, A., Boussetta, S., Cloke, H., Dee, D., et al. (2013). ERA In-terim/Land: A global land water resources dataset. Hydrology and Earth SystemSciences Discussions, 10, 14705–14745.

Becker, F., & Choudhury, B.J. (1988). Relative sensitivity of normalized differencevegetation index (NDVI) and microwave polarization difference index (MPDI) forvegetation and desertification monitoring. Remote Sensing of Environment, 24(2),297–311.

Bi, H.Y., Ma, J.W., & Wang, F.J. (2014). An improved particle filter algorithm based on en-semble Kalman filter and Markov chain Monte Carlo method. IEEE Journal of SelectedTopics in Applied Earth Observations and Remote Sensing. http://dx.doi.org/10.1109/JSTARS.2014.2322096.

Brocca, L., Hasenauer, S., Lacava, T., Melone, F., Moramarco, T., Wagner, W., et al. (2011).Soil moisture estimation through ASCAT and AMSR-E sensors: An intercomparisonand validation study across Europe. Remote Sensing of Environment, 115(12),3390–3408.

Brocca, L., Melone, F., Moramarco, T., Wagner, W., Naeimi, V., Bartalis, Z., et al. (2010).Improving runoff prediction through the assimilation of the ASCAT soil moistureproduct. Hydrology and Earth System Sciences, 14(10), 1881–1893.

Brocca, L., Moramarco, T., Melone, F., Wagner, W., Hasenauer, S., & Hahn, S. (2012).Assimilation of surface- and root-zone ASCAT soil moisture products into rainfall–runoff modeling. IEEE Transactions on Geoscience and Remote Sensing, 50(7),2542–2555.

Chanzy, A., Raju, S., & Wigneron, J.P. (1997). Estimation of soil microwave effective tem-perature at L and C bands. IEEE Transactions on Geoscience and Remote Sensing, 35(3),570–580.

Chen, L., Shi, J., Wigneron, J.P., & Chen, K.S. (2010). A parameterized surface emissionmodel at L-band for soil moisture retrieval. IEEE Geoscience and Remote SensingLetters, 7(1), 127–130.

Chen, Y.Y., Yang, K., Qin, J., Zhao, L., Tang, W.J., & Han, M.L. (2013). Evaluation of AMSR-Eretrievals and GLDAS simulations against observations of a soil moisture network onthe central Tibetan Plateau. Journal of Geophysical Research, [Atmospheres], 118(10),4466–4475. http://dx.doi.org/10.1002/jgrd.50301.

Choudhury, B.J., Schmugge, T.J., & Mo, T. (1982). A parameterization of effective soiltemperature for microwave emission. Journal of Geophysical Research, 87,1301–1304.

de Jeu, R.A.M., Holmes, T.R.H., Panciera, R., & Walker, J.P. (2009). Parameterization of theland parameter retrieval model for L-band observations using the NAFE'05 data set.IEEE Geoscience and Remote Sensing Letters, 6(4), 630–634.

de Jeu, R.A.M., & Owe, M. (2003). Further validation of a new methodology for surfacemoisture and vegetation optical depth retrieval. International Journal of RemoteSensing, 24(22), 4559–4578.

Dee, D.P., Uppala, S.M., Simmons, A.J., Berrisford, P., Poli, P., Kobayashi, S., et al.(2011). The ERA-Interim reanalysis: Configuration and performance of the dataassimilation system. Quarterly Journal of the Royal Meteorological Society,137(656), 553–597.

Dente, L., Su, Z., & Wen, J. (2012). Validation of SMOS soil moisture products over theMaqu and Twente regions. Sensors, 12(8), 9965–9986.

Dente, L., Vekerdy, Z., Wen, J., & Su, Z. (2012). Maqu network for validation of satellite-derived soil moisture products. International Journal of Applied Earth Observationand Geoinformation, 17, 55–65.

Dharssi, I., Bovis, K.J., Macpherson, B., & Jones, C.P. (2011). Operational assimilation ofASCAT surface soil wetness at the Met Office. Hydrology and Earth System Sciences,15(8), 2729–2746.

Dobson, M.C., Ulaby, F.T., Hallikainen, M.T., & El-Rayes, M.A. (1985). Microwave dielectricbehavior of wet soil—Part II: Dielectric mixingmodels. IEEE Transactions on Geoscienceand Remote Sensing, GE-23(1), 35–46.

Dorigo, W.A., Gruber, A., De Jeu, R.A.M., Wagner, W., Stacke, T., Loew, A., et al. (2015).Evaluation of the ESA CCI soil moisture product using ground-based observations.Remote Sensing of Environment, 162, 380–395.

Dorigo, W.A., Scipal, K., Parinussa, R.M., Liu, Y.Y., Wagner, W., de Jeu, R.A.M., et al. (2010).Error characterisation of global active and passive microwave soil moisture datasets.Hydrology and Earth System Sciences, 14, 2605–2616.

Dorigo, W.A., Xaver, A., Vreugdenhil, M., Gruber, A., Hegyiová, A., Sanchis-Dufau, A.D.,et al. (2013). Global automated quality control of in situ soil moisture data fromthe international soil moisture network. Vadose Zone Journal, 12(3). http://dx.doi.org/10.2136/vzj2012.0097.

Draper, C., Reichle, R., de Jeu, R., Naeimi, V., Parinussa, R., &Wagner,W. (2013). Estimatingroot mean square errors in remotely sensed soil moisture over continental scale do-mains. Remote Sensing of Environment, 137, 288–298.

Draper, C.S., Walker, J.P., Steinle, P.J., de Jeu, R.A.M., & Holmes, T.R.H. (2009). An evaluationof AMSR-E derived soil moisture over Australia. Remote Sensing of Environment,113(4), 703–710.

Entekhabi, D., Njoku, E.G., O'Neill, P.E., Kellogg, K.H., Crow, W.T., Edelstein, W.N., et al.(2010). The soil moisture active passive (SMAP) mission. Proceedings of the IEEE,98(5), 704–716. http://dx.doi.org/10.1109/JPROC.2010.2043918.

Entekhabi, D., Reichl, R.H., Koster, R.D., & Crow, W.T. (2010). Performance metrics for soilmoisture retrievals and application requirements. Journal of Hydrometeorology, 11,832–840.

Escorihuela, M.J., Chanzy, A., Wigneron, J.P., & Kerr, Y.H. (2010). Effective soil moisturesampling depth of L-band radiometry: A case study. Remote Sensing of Environment,114(5), 995–1001.

Guo, P., Shi, J.C., Liu, Q., & Du, J.Y. (2013). A new algorithm for soil moisture retrieval withL-band radiometer. IEEE Journal of Selected Topics in Applied Earth Observations andRemote Sensing, 6(3), 1147–1155.

Holmes, T.R.H. (2003). Measuring surface soil parameters using passive microwave remotesensing. The ELBARA campaign 2003, HD-no.: 639. Amsterdam: Vrije Universiteit.

Holmes, T.R.H., de Jeu, R.A.M., Owe, M., & Dolman, A.J. (2009). Land surfacetemperature from Ka band (37 GHz) passive microwave observations.Journal of Geophysical Research: Atmospheres, 114, D04113. http://dx.doi.org/10.1029/2008JD010257.

Holmes, T.R.H., de Rosnay, P., de Jeu, R., Wigneron, J.P., Kerr, Y., Calvet, J.C., et al. (2006). Anew parameterization of the effective temperature for L band radiometry.Geophysical Research Letters, 33(7). http://dx.doi.org/10.1029/2006GL025724.

Imaoka, K., Kachi, M., Fujii, H., Murakami, H., Hori, M., Ono, A., et al. (2010). Global ChangeObservation Mission (GCOM) for monitoring carbon, water cycles, and climatechange. Proceedings of the IEEE, 98(5), 717–734. http://dx.doi.org/10.1109/JPROC.2009.2036869.

Imaoka, K., Maeda, T., Kachi, M., Kasahara, M., Ito, N., & Nakagawa, K. (2012). Status ofAMSR2 instrument on GCOM-W1. Proceedings of SPIE, 8528, 852815.

Jackson, T.J. (1993). Measuring surface soil moisture using passive microwave remotesensing. Hydrological Processes, 7(2), 139–152.

Jackson, T.J., Bindlish, R., Cosh, M.H., Zhao, T.J., Starks, P.J., Bosch, D.D., et al. (2012).Validation of Soil Moisture and Ocean Salinity (SMOS) soil moisture over watershednetworks in the US. IEEE Transactions on Geoscience and Remote Sensing, 50(5),1530–1543.

Jackson, T.J., Cosh, M.H., Bindlish, R., Starks, P.J., Bosch, D.D., Seyfried, M., et al. (2010). Val-idation of advanced microwave scanning radiometer soil moisture products. IEEETransactions on Geoscience and Remote Sensing, 48(12), 4256–4272.

Jacquette, E., Al Bitar, A., Mialon, A., Kerr, Y., Quesney, A., Cabot, F., et al. (2010). SMOSCATDS level 3 global products over land. Proceedings of SPIE, 7824, 78240K.

Jones, M.O., Jones, L.A., Kimball, J.S., & McDonald, K.C. (2011). Satellite passive microwaveremote sensing for monitoring global land surface phenology. Remote Sensing ofEnvironment, 115(4), 1102–1114.

Kaihotsu, I., Koike, T., Yamanaka, T., Fujii, H., Ohta, T., Tamagawa, K., et al. (2009).Validation of soil moisture estimation by AMSR-E in the Mongolian plateau. Journalof the Remote Sensing Society of Japan, 29, 271–281.

Kerr, Y.H., Waldteufel, P., Richaume, P., Wigneron, J.P., Ferrazzoli, P., Mahmoodi, A., et al.(2012). The SMOS soil moisture retrieval algorithm. IEEE Transactions on Geoscienceand Remote Sensing, 50(5), 1384–1403.

Kerr, Y.H., Waldteufel, P., Wigneron, J.P., Delwart, S., Cabot, F., Boutin, J., et al. (2010). TheSMOS mission: New tool for monitoring key elements of the global water cycle.Proceedings of the IEEE, 98(5), 666–687. http://dx.doi.org/10.1109/JPROC.2010.2043032.

Koike, T., Nakamura, Y., Kaihotsu, I., Davva, G., Matsuura, N., Tamagawa, K., et al. (2004).Development of an Advanced Microwave Scanning Radiometer (AMSR-E) algorithmof soil moisture and vegetation water content. Annual Journal of HydraulicEngineering, JSCE, 48, 217–222.

http://refhub.elsevier.com/S0034-4257(15)00104-2/rf0025


























http://dx.doi.org/10.1109/JSTARS.2014.2322096

http://dx.doi.org/10.1109/JSTARS.2014.2322096















http://dx.doi.org/10.1002/jgrd.50301




























http://dx.doi.org/10.2136/vzj2012.0097







http://dx.doi.org/10.1109/JPROC.2010.2043918












http://dx.doi.org/10.1029/2008JD010257

http://dx.doi.org/10.1029/2006GL025724




























Koike, T., Tsukamoto, T., Kumakura, T., & Lu, M. (1996). Spatial and seasonal distributionof surface wetness derived from satellite data. Proceeding of the International Work-shop on Macro-scale Hydrological Modeling (pp. 87–96).

Koster, R.D., Guo, Z., Yang, R., Dirmeyer, P.A., Mitchell, K., & Puma, M.J. (2009). On the na-ture of soil moisture in land surface models. Journal of Climate, 22(16), 4322–4335.

Lacava, T., Matgen, P., Brocca, L., Bittelli, M., Pergola, N., Moramarco, T., et al. (2012). A firstassessment of the SMOS soil moisture product with in situ and modeled data in Italyand Luxembourg. IEEE Transactions on Geoscience and Remote Sensing, 50(5),1612–1622.

Leroux, D.J., Kerr, Y.H., Al Bitar, A., Bindlish, R., Jackson, T.J., Berthelot, B., et al. (2014).Comparison between SMOS, VUA, ASCAT, and ECMWF soil moisture products overfour watersheds in US. IEEE Transactions on Geoscience and Remote Sensing, 52(3),1562–1571.

Leroux, D.J., Kerr, Y.H., Richaume, P., & Fieuzal, R. (2013). Spatial distribution and possiblesources of SMOS errors at the global scale. Remote Sensing of Environment, 133,240–250.

Li, L., Gaiser, P.W., Gao, B.C., Bevilacqua, R.M., Jackson, T.J., Njoku, E.G., et al. (2010).WindSat global soil moisture retrieval and validation. IEEE Transactions onGeoscience and Remote Sensing, 48(5), 2224–2241.

Liu, Y.Y., Dorigo, W.A., Parinussa, R.M., De Jeu, R.A.M., Wagner, W., McCabe, M.F., et al.(2012). Trend-preserving blending of passive and active microwave soil moisture re-trievals. Remote Sensing of Environment, 123, 280–297.

Liu, Y.Y., Parinussa, R.M., Dorigo, W.A., de Jeu, R.A.M., Wagner, W., Van Dijk, A.I.J.M., et al.(2011). Developing an improved soil moisture dataset by blending passive and activemicrowave satellite-based retrievals. Hydrology and Earth System Sciences, 15(2),425–436.

Loew, A., & Schlenz, F. (2011). A dynamic approach for evaluating coarse scale satellitesoil moisture products. Hydrology and Earth System Sciences, 15, 75–90.

Loew, A., Stacke, T., Dorigo, W., de Jeu, R., & Hagemann, S. (2013). Potential and limita-tions of multidecadal satellite soil moisture observations for selected climate modelevaluation studies. Hydrology and Earth System Sciences, 17, 3523–3542.

Lv, S.N., Wen, J., Zeng, Y.J., Tian, H., & Su, Z.B. (2014). An improved two-layer algorithmfor estimating effective soil temperature in microwave radiometry using in situtemperature and soil moisture measurements. Remote Sensing of Environment, 152,356–363.

Ma, Y., Su, Z., Koike, T., Yao, T., Ishikawa, H., Ueno, K.I., et al. (2003). Onmeasuring and re-mote sensing surface energy partitioning over the Tibetan Plateau—From GAME/Tibet to CAMP/Tibet. Physics and Chemistry of the Earth, Parts A/B/C, 28(1), 63–74.

Matgen, P., Heitz, S., Hasenauer, S., Hissler, C., Brocca, L., Hoffmann, L., et al. (2012). On thepotential of MetOp ASCAT-derived soil wetness indices as a new aperture for hydro-logical monitoring and prediction: A field evaluation over Luxembourg. HydrologicalProcesses, 26(15), 2346–2359.

Meesters, A.G., de Jeu, R.A.M., & Owe, M. (2005). Analytical derivation of the vegetationoptical depth from the microwave polarization difference index. IEEE Geoscienceand Remote Sensing Letters, 2(2), 121–123.

Miralles, D.G., Crow, W.T., & Cosh, M.H. (2010). Estimating spatial sampling errors incoarse-scale soil moisture estimates derived from point-scale observations. Journalof Hydrometeorology, 11, 1423–1429.

Miralles, D.G., Holmes, T.R.H., de Jeu, R.A.M., Gash, J.H., Meesters, A.G.C.A., & Dolman, A.J.(2011). Global land–surface evaporation estimated from satellite-based observations.Hydrology and Earth System Sciences, 15(2), 453–469.

Mo, T., Choudhury, B.J., Schmugge, T.J., Wang, J.R., & Jackson, T.J. (1982). A model for mi-crowave emission from vegetation-covered fields. Journal of Geophysical Research,Oceans, 87(C13), 11229–11237.

Moradkhani, H., Hsu, K.L., Gupta, H., & Sorooshian, S. (2005). Uncertainty assessment ofhydrologic model states and parameters: Sequential data assimilation using the par-ticle filter. Water Resources Research, 41(5), W05012. http://dx.doi.org/10.1029/2004WR003604.

Naeimi, V., Scipal, K., Bartalis, Z., Hasenauer, S., & Wagner, W. (2009). An improved soilmoisture retrieval algorithm for ERS and METOP scatterometer observations. IEEETransactions on Geoscience and Remote Sensing, 47(7), 1999–2013.

Njoku, E.G., Ashcroft, P., Chan, T.K., & Li, L. (2005). Global survey and statistics of radio-frequency interference in AMSR-E land observations. IEEE Transactions onGeoscience and Remote Sensing, 43(5), 938–947.

Njoku, E.G., & Chan, S.K. (2006). Vegetation and surface roughness effects on AMSR-E landobservations. Remote Sensing of Environment, 100(2), 190–199.

Njoku, E.G., Jackson, T.J., Lakshmi, V., Chan, T.K., & Nghiem, S.V. (2003). Soil moisture re-trieval from AMSR-E. IEEE Transactions on Geoscience and Remote Sensing, 41(2),215–229.

Oliva, R., Daganzo, E., Kerr, Y.H., Mecklenburg, S., Nieto, S., Richaume, P., et al. (2012).SMOS radio frequency interference scenario: Status and actions taken to improvethe RFI environment in the 1400–1427-MHz passive band. IEEE Transactions onGeoscience and Remote Sensing, 50(5), 1427–1439.

Owe, M., de Jeu, R.A.M., & Holmes, T.R.H. (2008). Multisensor historical climatology ofsatellite-derived global land surface moisture. Journal of Geophysical Research, EarthSurface, 113, F01002. http://dx.doi.org/10.1029/2007JF000769.

Owe,M., de Jeu, R.A.M., &Walker, J.P. (2001). Amethodology for surface soil moisture andvegetation optical depth retrieval using the microwave polarization difference index.IEEE Transactions on Geoscience and Remote Sensing, 39(8), 1643–1654.

Owe, M., & Van de Griend, A.A. (2001). On the relationship between thermodynamic sur-face temperature and high-frequency (37 GHz) vertically polarized brightness tem-perature under semi-arid conditions. International Journal of Remote Sensing, 22(17),3521–3532.

Paloscia, S., Macelloni, G., & Santi, E. (2006). Soil moisture estimates from AMSR-E bright-ness temperatures by using a dual-frequency algorithm. IEEE Transactions onGeoscience and Remote Sensing, 44(11), 3135–3144.

Pan, M., Sahoo, A.K., & Wood, E.F. (2014). Improving soil moisture retrievals from aphysically-based radiative transfer model. Remote Sensing of Environment, 140,130–140.

Pan, M., Sahoo, A.K., Wood, E.F., Al Bitar, A., Leroux, D., & Kerr, Y.H. (2012). An initial as-sessment of SMOS derived soil moisture over the continental United States. IEEEJournal of Selected Topics in Applied Earth Observations and Remote Sensing, 5(5),1448–1457.

Parinussa, R.M., Holmes, T.R.H., & de Jeu, R.A.M. (2012). Soil moisture retrievals from theWindSat spaceborne polarimetric microwave radiometer. IEEE Transactions onGeoscience and Remote Sensing, 50(7), 2683–2694.

Parinussa, R.M., Yilmaz, M.T., Anderson, M.C., Hain, C.R., & de Jeu, R.A.M. (2014). An inter-comparison of remotely sensed soil moisture products at various spatial scales overthe Iberian Peninsula. Hydrological Processes, 28(18), 4865–4876.

Qin, J., Yang, K., Lu, N., Chen, Y., Zhao, L., & Han, M. (2013). Spatial upscaling of in-situ soilmoisture measurements based on MODIS-derived apparent thermal inertia. RemoteSensing of Environment, 138, 1–9.

Qiu, J.X., Mo, X.G., Liu, S.X., Lin, Z.H., Yang, L.H., Song, X.F., et al. (2013). Intercomparison ofmicrowave remote-sensing soil moisture data sets based on distributed eco-hydrological model simulation and in situ measurements over the North ChinaPlain. International Journal of Remote Sensing, 34(19), 6587–6610.

Reynolds, C.A., Jackson, T.J., & Rawls, W.J. (2000). Estimating soil water-holding capacitiesby linking the Food and Agriculture Organization Soil map of the world with globalpedon databases and continuous pedotransfer functions. Water Resources Research,36(12), 3653–3662.

Rodell, M., Houser, P.R., Jambor, U., Gottschalck, J., Mitchell, K., Meng, C.J., et al. (2004). Theglobal land data assimilation system. Bulletin of the American Meteorological Society,85, 381–394. http://dx.doi.org/10.1175/BAMS-85-3-381.

Rötzer, K., Montzka, C., Bogena, H., Wagner, W., Kerr, Y.H., Kidd, R., et al. (2014).Catchment scale validation of SMOS and ASCAT soil moisture products using hydro-logical modeling and temporal stability analysis. Journal of Hydrology, 519, 934–946.

Rüdiger, C., Calvet, J.C., Gruhier, C., Holmes, T.R., de Jeu, R.A., & Wagner, W. (2009). An in-tercomparison of ERS-Scat and AMSR-E soil moisture observations with model simu-lations over France. Journal of Hydrometeorology, 10(2), 431–447.

Rüdiger, C., Walker, J.P., Kerr, Y.H., Mialon, A., Merlin, O., & Kim, E.J. (2011). Validation ofthe level 1c and level 2 SMOS products with airborne and ground-based observations.International Congress on Modelling and Simulation (MODSIM) (pp. 12–16).

Sahoo, A.K., Houser, P.R., Ferguson, C., Wood, E.F., Dirmeyer, P.A., & Kafatos, M. (2008).Evaluation of AMSR-E soil moisture results using the in-situ data over the LittleRiver Experimental Watershed, Georgia. Remote Sensing of Environment, 112(6),3142–3152.

Salama, M.S., Van der Velde, R., Zhong, L., Ma, Y.M., Ofwono, M., & Su, Z.B. (2012). Decadalvariations of land surface temperature anomalies observed over the Tibetan Plateauby the Special Sensor Microwave Imager (SSM/I) from 1987 to 2008. ClimaticChange, 114, 769–781.

Saleh, K., Kerr, Y.H., Richaume, P., Escorihuela, M.J., Panciera, R., Delwart, S., et al. (2009).Soil moisture retrievals at L-band using a two-step inversion approach (COSMOS/NAFE'05 Experiment). Remote Sensing of Environment, 113(6), 1304–1312.

Santi, E., Pettinato, S., Paloscia, S., Pampaloni, P., Macelloni, G., & Brogioni, M. (2012). Analgorithm for generating soil moisture and snow depth maps from microwavespaceborne radiometers: HydroAlgo. Hydrology and Earth System Sciences, 16,3659–3676.

Schmugge, T.J. (1985). Chapter 5: Remote sensing of soil moisture. In M.G. Anderson, & T.P.Burt (Eds.), Hydrological forecasting (pp. 101–124). New York: John Wiley and Sons.

Schmugge, T.J., Jackson, T.J., & McKim, H.L. (1980). Survey of methods for soil moisture de-termination. Water Resources Research, 16(6), 961–979.

Shi, J.C., Jiang, L.M., Zhang, L.X., Chen, K.S., Wigneron, J.P., Chanzy, A., et al. (2006). Physi-cally based estimation of bare-surface soil moisture with the passive radiometers.IEEE Transactions on Geoscience and Remote Sensing, 44(11), 3145–3153.

Su, C.H., Ryu, D., Young, R.I.,Western, A.W., &Wagner,W. (2013). Inter-comparison of mi-crowave satellite soil moisture retrievals over the Murrumbidgee Basin, southeastAustralia. Remote Sensing of Environment, 134, 1–11.

Su, Z., de Rosnay, P., Wen, J., Wang, L., & Zeng, Y. (2013). Evaluation of ECMWF's soil mois-ture analyses using observations on the Tibetan Plateau. Journal of GeophysicalResearch, 118(11), 5304–5318. http://dx.doi.org/10.1002/jgrd.50468.

Su, Z., Wen, J., Dente, L., van der Velde, R., Wang, L., Ma, Y., et al. (2011). The TibetanPlateau observatory of plateau scale soil moisture and soil temperature (Tibet-Obs)for quantifying uncertainties in coarse resolution satellite and model products.Hydrology and Earth System Sciences, 15(7), 2303–2316.

Taylor, K.E. (2001). Summarizing multiple aspects of model performance in a single dia-gram. Journal of Geophysical Research, 106, 7183–7192.

Wagner,W., Bartalis, Z., Naeimi, V., Park, S.E., Figa-Saldaña, J., & Bonekamp, H. (2010). Sta-tus of the METOP ASCAT soil moisture product. IEEE International Geoscience and Re-mote Sensing Symposium 2010 (IGARSS2010), 25–30 July 2010, Honolulu, Hawaii, U.S.A.(pp. 276–279).

Wagner, W., Dorigo, W., de Jeu, R., Fernandez, D., Benveniste, J., Haas, E., et al. (2012).Fusion of active and passive microwave observations to create an Essential ClimateVariable data record on soil moisture. XXII ISPRS Congress, Melbourne, Australia, 25August–1 September 2012 (pp. 315–321).

Wagner, W., Hahn, S., Kidd, R., Melzer, T., Bartalis, Z., Hasenauer, S., et al. (2013). TheASCAT soil moisture product: A review of its specifications, validation results, andemerging applications. Meteorologische Zeitschrift, 22, 5–33.

Wagner, W., Lemoine, G., & Rott, H. (1999). A method for estimating soil moisture fromERS scatterometer and soil data. Remote Sensing of Environment, 70, 191–207.

Wagner, W., Naeimi, V., Scipal, K., de Jeu, R., & Martínez-Fernández, J. (2007). Soilmoisture from operational meteorological satellites. Hydrogeology Journal, 15(1),121–131.


















































http://dx.doi.org/10.1029/2004WR003604

http://dx.doi.org/10.1029/2004WR003604















http://dx.doi.org/10.1029/2007JF000769



































http://dx.doi.org/10.1175/BAMS-85-3-381
































http://dx.doi.org/10.1002/jgrd.50468























Wanders, N., Karssenberg, D., Bierkens, M., Parinussa, R., de Jeu, R., van Dam, J., et al.(2012). Observation uncertainty of satellite soil moisture products determined withphysically-based modeling. Remote Sensing of Environment, 127, 341–356.

Wang, J.R., & Choudhury, B.J. (1981). Remote sensing of soil moisture content, over barefield at 1.4 GHz frequency. Journal of Geophysical Research, Oceans, 86(C6),5277–5282.

Wen, J., & Su, Z.B. (2003a). A time series based method for estimating relative soil mois-ture with ERS wind scatterometer data. Geophysical Research Letters, 30(7), 1397.http://dx.doi.org/10.1029/2002GL016557.

Wen, J., & Su, Z.B. (2003b). The estimation of soil moisture from ERS wind scatterometerdata over the Tibetan Plateau. Physics and Chemistry of the Earth, 28(1), 53–61.

Wen, J., Su, Z.B., & Ma, Y.M. (2003). Determination of land surface temperature and soilmoisture from Tropical Rainfall Measuring Mission/Microwave Imager remote sens-ing data. Journal of Geophysical Research, 108(D2). http://dx.doi.org/10.1029/2002JD002176.

Wigneron, J.P., Chanzy, A., de Rosnay, P., Rüdiger, C., & Calvet, J.C. (2008). Estimating theeffective soil temperature at L-band as a function of soil properties. IEEE Transactionson Geoscience and Remote Sensing, 46, 797–807.

Wigneron, J.P., Chanzy, A., Kerr, Y.H., Lawrence, H., Shi, J.C., Escorihuela, M.J., et al. (2011).Evaluating an improved parameterization of the soil emission in L-MEB. IEEETransactions on Geoscience and Remote Sensing, 49(4), 1177–1189.

Wigneron, J.P., Kerr, Y., Waldteufel, P., Saleh, K., Escorihuela, M.J., Richaume, P., et al.(2007). L-band Microwave Emission of the Biosphere (L-MEB) model: Description

and calibration against experimental data sets over crop fields. Remote Sensing ofEnvironment, 107(4), 639–655.

Wigneron, J.P., Laguerre, L., & Kerr, Y.H. (2001). A simple parameterization of the L-bandmicrowave emission from rough agricultural soils. IEEE Transactions on Geoscienceand Remote Sensing, 39(8), 1697–1707.

Wigneron, J.P., Schwank, M., Baeza, E.L., Kerr, Y., Novello, N., Millan, C., et al. (2012). Firstevaluation of the simultaneous SMOS and ELBARA-II observations in the Mediterra-nean region. Remote Sensing of Environment, 124, 26–37.

Xie, H., Ye, J., Liu, X., & Chongyi, E. (2010). Warming and drying trends on the Tibetan Pla-teau (1971–2005). Theoretical and Applied Climatology, 101(3–4), 241–253.

Yang, K., Qin, J., Zhao, L., Chen, Y.Y., Tang, W.J., Han, M.L., et al. (2013). A multiscale soilmoisture and freeze–thaw monitoring network on the Third Pole. Bulletin of theAmerican Meteorological Society, 94(12), 1907–1916.

Yang, K., Ye, B., Zhou, D., Wu, B., Foken, T., Qin, J., et al. (2011). Response of hydrologicalcycle to recent climate changes in the Tibetan Plateau. Climatic Change, 109(3–4),517–534.

Zeng, J.Y., Li, Z., Chen, Q., & Bi, H.Y. (2014). A simplified physically-based algorithm for sur-face soil moisture retrieval using AMSR-E data. Frontiers of Earth Science, 8(3),427–438.

Zeng, J.Y., Li, Z., Chen, Q., & Bi, H.Y. (2015). Method for soil moisture and surface temper-ature estimation in the Tibetan Plateau using spaceborne radiometer observations.IEEE Geoscience and Remote Sensing Letters, 12(1), 97–101.






http://dx.doi.org/10.1029/2002GL016557



http://dx.doi.org/10.1029/2002JD002176

http://dx.doi.org/10.1029/2002JD002176





























remote sensing of environment -...

Documents