This article was downloaded by: [Northwest A & F University]On: 24 December 2014, At: 20:41Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,37-41 Mortimer Street, London W1T 3JH, UK

Click for updates

Hydrological Sciences JournalPublication details, including instructions for authors and subscription information:http://www.tandfonline.com/loi/thsj20

Spatio-temporal variability of surface soil watercontent and its influencing factors in a desert area,ChinaPingping Zhanga & Ming’an Shaoab

a State Key Laboratory of Soil Erosion and Dryland Farming on the Loess Plateau, NorthwestA & F University, Yangling, Chinab Key Laboratory of Ecosystem Network Observation and Modelling, Institute ofGeographical Science and Natural Resources, Chinese Academy of Sciences, Beijing, ChinaPublished online: 20 Dec 2014.

To cite this article: Pingping Zhang & Ming’an Shao (2014) Spatio-temporal variability of surface soil water content and itsinfluencing factors in a desert area, China, Hydrological Sciences Journal, 60:1, 96-110, DOI: 10.1080/02626667.2013.875178

To link to this article: http://dx.doi.org/10.1080/02626667.2013.875178

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) containedin the publications on our platform. However, Taylor & Francis, our agents, and our licensors make norepresentations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of theContent. Any opinions and views expressed in this publication are the opinions and views of the authors, andare not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon andshould be independently verified with primary sources of information. Taylor and Francis shall not be liable forany losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoeveror howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use ofthe Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematicreproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in anyform to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions

Spatio-temporal variability of surface soil water content and itsinfluencing factors in a desert area, China

Pingping Zhang1 and Ming’an Shao1,2

1State Key Laboratory of Soil Erosion and Dryland Farming on the Loess Plateau, Northwest A & F University, Yangling, Chinamashao@ms.iswc.ac.cn2Key Laboratory of Ecosystem Network Observation and Modelling, Institute of Geographical Science and Natural Resources, ChineseAcademy of Sciences, Beijing, China

Received 9 May 2013; accepted 29 October 2013

Editor D. Koutsoyiannis; Associate editor D. Gerten

Abstract Knowledge of the variability of soil water content (SWC) in space and time plays a key role inhydrological and climatic modelling. However, limited attention has been given to arid regions. The focus of thisstudy was to investigate the spatio-temporal variability of surface soil (0–6 cm) water content and to identify itscontrolling factors in a region of the Gobi Desert (40 km2). The standard deviation of SWC decreasedlogarithmically as mean water content decreased, and the coefficient of variation of SWC exhibited a convexupward pattern. The spatial variability of SWC also increased with the size of the investigated area. The spatialdependence of SWC changed over time, with stronger patterns of spatial organization in drier and wetterconditions of soil wetness and stochastic patterns in moderate soil water conditions. The dominant factorsregulating the variability of SWC changed from combinations of soil and topographical properties (bulk density,clay content and relative elevation) in wet conditions to combinations of soil and vegetation properties (bulkdensity, clay content and shrub coverage) in dry conditions. This study has important implications for theassessment of soil quality and the sustainability of land management in arid regions.

Key words surface soil water; spatial variability; influencing factors; statistical analysis; geostatistics; arid area

Variabilité spatio-temporelle de la teneur en eau superficielle des sols et facteurs d’influence dansune zone désertique en ChineRésumé La connaissance de la variabilité de la teneur en eau du sol (TES) dans l’espace et le temps joue un rôleclé dans la modélisation hydrologique et climatique. Cependant, les régions arides ont fait l’objet d’une attentionlimitée. L’objectif de cette étude était d’étudier la variabilité spatio-temporelle de la teneur en eau superficielle dusol (0–6 cm) et d’en identifier les facteurs de contrôle dans une région du désert de Gobi (40 km2). Les résultatsmontrent que l’écart-type de la TES diminue de manière logarithmique avec la diminution de la teneur en eaumoyenne, et que le coefficient de variation de la teneur en eau moyenne affiche un profil convexe vers le haut. Lavariabilité spatiale de la TES augmente aussi avec la taille de la zone étudiée. La dépendance spatiale de la TESchange au fil du temps, avec des structures d’organisation spatiale plus fortes dans les conditions d’humidité dusol plus sèches et plus humides, et des structures aléatoires dans les conditions intermédiaires. Les facteursdominants qui régissent la variabilité de la TES vont de combinaisons de propriétés pédologiques et topogra-phiques (densité apparente, teneur en argile et élévation relative) dans les conditions humides, à des combinaisonsde propriétés de sol et de végétation (densité apparente, teneur en argile et couverture arbustive) dans lesconditions sèches. Cette étude a des implications importantes pour l’évaluation de la qualité des sols et ladurabilité de la gestion du territoire dans les régions arides.

Mots clefs eau superficielle du sol ; variabilité spatiale ; facteurs d’influence ; analyse statistique ; géostatistique ; zone aride

1 INTRODUCTION

The water content in surface soils plays a critical rolein hydrology, agrology and ecology, and stronglycontrols various natural processes, such as energy

fluxes of the land surface, soil degradation, land–atmosphere interactions, and conservation andrestoration of vegetation (Charpentier and Groffman1992, Wang et al. 2012, Jia and Shao 2013).

96 Hydrological Sciences Journal – Journal des Sciences Hydrologiques, 60 (1) 2015http://dx.doi.org/10.1080/02626667.2013.875178

© 2014 IAHS

Dow

nloa

ded

by [

Nor

thw

est A

& F

Uni

vers

ity]

at 2

0:41

24

Dec

embe

r 20

14

Characterizing the spatio-temporal variability of soilwater content (SWC) is a key challenge for improv-ing hydrological and climatic modelling and predic-tion (Ampofo 2006), and has thus become incre-asingly important in recent years, from the scaleof hillslopes (Famiglietti et al. 1998, Zhu and Shaoet al. 2008, Penna et al. 2009) and small watershedsusing ground-based measurements (Hébrard et al.2006, Hu et al. 2010, Brocca et al. 2012, Takagiand Lin 2012) to national and global scales usingremote sensing (IGPO 1995, Lu and Shi 2012).Remote sensing is promising for the description andmeasurement of surface SWC on large scales, but itsusefulness and interpretation require further testingdue to its relatively low spatial and temporal resolu-tion (Brocca et al. 2007). Data obtained from accu-rate on-site measurements should thus continue to beimportant when studying the variability of SWC(Owe et al. 1982). SWC in arid regions is the keyresource limiting the development of vegetation andis the main constraint to permanently controllingdesertification (Berndtsson et al. 1996). Soil wateralso determines the organization and functioningof the ecosystems in arid regions. Despite its impor-tance to vegetation, soil water has received littleattention in arid environments, especially in the char-acterization of its variability in natural deserts, due tothe difficulty of data collection and the associatedcosts.

The spatio-temporal variability of SWC hasbeen characterized by testing its link to field (spa-tial) means of soil wetness. Knowledge of this rela-tionship is useful for determining the minimumamount of sampling needed to effectively estimatefield means of water content with an acceptableerror, estimating the error if the amount of samplinghas been predefined, and verifying and validatingremotely sensed effects of SWC (Brocca et al.2007). Whether the variability of SWC has a posi-tive or negative relationship with field means ofwater content, however, has not been conclusivelydetermined (Reynolds 1970, Owe et al. 1982, Zhanget al. 2011).

The variability and distribution pattern of sur-face SWC in time and space can be controlled bymany factors, such as soil properties, topographicalcharacteristics and vegetation (Qiu et al. 2001, Huet al. 2008, Zhao et al. 2011, Takagi and Lin 2012).However, identifying the relative significance ofthese individual factors is challenging because oftheir mutual or multiple influences on SWC. Many

studies have indicated that the spatial pattern ofSWC, especially in arid regions, cannot be accu-rately predicted by any single factor. In addition,the dominant factors may change with the chosensites and depend on the conditions of soil wetnessresulting from seasonal wetting and drying (Graysonet al. 1997, Famiglietti et al. 1998, Western et al.1999, Pan and Wang 2009). For example,Famiglietti et al. (1998) explored the possible fac-tors influencing the spatial pattern of SWC alongthe profile of a hillslope. In wetter conditions, theheterogeneity of soil properties (porosity, hydraulicconductivity) significantly contributed to the spatialvariability of SWC, but in drier conditions, topogra-phical characteristics (relative elevation and aspect)largely determined the variability of SWC. Westernet al. (1999) showed that the indices associated withthe lateral flow of water and those related to evapo-transpiration were the most important predictors ofdistributional patterns of SWC in wet and dry con-ditions, respectively. Hawley et al. (1983) demon-strated that the variability of SWC was mainlydriven by topography, but the explanatory powertended to decrease with seasonal variations of thevegetation cover. Gómez-Plaza et al. (2001) foundthat some local controls (e.g. soil texture and slopeangle) were the main regulators of the distributionof SWC in non-vegetated areas, whereas the con-trols operating on SWC in vegetated areas werethose linked to the properties of the vegetation.The factors controlling the dynamics of SWCremain quite uncertain, and much more research isneeded in a variety of environments and on a vari-ety of scales.

Soil water content and its complex spatial dis-tribution are very important in arid regions. Wetherefore studied the spatio-temporal variability ofsurface soil (0–6 cm) water content in a typicalfenced region of the Gobi Desert where precipitationis the only source of soil moisture. The main goalsof the study were: (a) to quantify the spatio-tem-poral variability of the surface SWC, (b) to under-stand the relative roles of elevation, soil propertiesand vegetation in controlling the variability of SWCunder different conditions of soil wetness, (c) todetermine the amount of sampling needed to esti-mate field means of water content at a predefinedlevel of statistical significance, and (d) to determinethe relationship between the variability of SWC andthe size of the sampling area using a re-samplingmethod.

Spatio-temporal variability of surface soil water content 97

Dow

nloa

ded

by [

Nor

thw

est A

& F

Uni

vers

ity]

at 2

0:41

24

Dec

embe

r 20

14

2 MATERIALS AND METHODS

2.1 Description of the study area

The study was conducted in a region of the GobiDesert (39°24′–39°28′N; 100°08′–100°11′E) in thecentral reaches of the Heihe River basin in GansuProvince, northern China (Fig. 1(a), (b)). The eleva-tion ranges from 1390 to 1470 m a.s.l., and the watertable is at a depth of nearly 11–13 m (Yu et al. 2012).The area is characterized by low and highly variablerainfall, and has an arid desert climate. The averageannual rainfall and air temperature are 117 mm year-1

and 7.6°C, respectively. Rainfall in brief summershowers contributes 65% of the total annual precipi-tation. The mean annual pan-evaporation is approxi-mately 2390 mm year-1; this is 20 times greater thanthe annual precipitation. The zonal soil is classifiedas grey-brown desert soil and is derived from dilu-vial-alluvial materials. The soil profile is sandy intexture, low in nutrients and has a small amount ofgravels in the surface and subsoil horizons. Theabove-ground plant cover is discontinuous and canbe described as patches of sub-shrubs surrounded bybare areas. The study area has been fenced and isprotected from grazing for the purpose of revegeta-tion and reclamation. The dominant plant speciesare Nitraria sphaerocarpa Maxim. and Reaumuria

soongorica (Pall.) Maxim., and the accompanyingplant species are mainly Kalidium gracile Fenzl,Allium mongolicum Rgl., Bassia dasyphylla (Fisch.and Mey.) Kuntze, Halogeton arachnoideus Moq.,Suaeda microphylla (Mey.) Pall., Caragana brachy-poda Pojark., Salsola ruthenica Iljin, Asterothamnuscentrali-asiaticus and Sympegma regelii Bunge.

2.2 Soil sampling and data collection

A regular sampling grid (500 m × 500 m) of 187sampling points was established over a 40-km2

(5 km × 8 km) experimental area (Fig. 1(c)). Eachsampling point was positioned with a portable GarminGPS receiver (with a resolution of 3 m), and eachpoint was marked by a wooden stick. The volumetricwater content of the surface soil (0–6 cm) was mea-sured with FDR probes (Type ML2x, Delta-TDevices) approximately every two weeks from 15April to 15 October 2012. A total of 13 measurementcampaigns were conducted. Measurements weretaken in the same sequence to reduce the possibleinfluence of sampling time. The probes were handledcarefully to avoid touching the gravel, which caninfluence the measurements. To further reduce thepossible influence of micro-scale variability, we aver-aged three measurements for each point and date. The

Fig. 1 (a and b) Location of the study site in northern China and (c) elevation contours and the sampling grid in the studysite.

98 Pingping Zhang and Ming’an Shao

Dow

nloa

ded

by [

Nor

thw

est A

& F

Uni

vers

ity]

at 2

0:41

24

Dec

embe

r 20

14

reliability of the Theta Probe was enhanced by cali-brating it with a site-specific gravimetric assessment.Calibration was performed at three times of year(May, August and October) to allow for variations insoil wetness. The calibrations had a root mean squarederror of approximately 1.09%, indicating adequacy.

A quantitative survey of the vegetation was per-formed in mid-August 2012, the period of vigorousgrowth of the vegetation. Quadrats of 20 m × 20 mfor shrubs and 2 m × 2 m for herbaceous specieswere established at each sampling point to investigatespecies richness and the coverage of herbs andshrubs. Species richness was defined as all occur-rences of species in all quadrats. The coverage ofthe herbs and shrubs was calculated as a percentageof the area.

At each sampling point, disturbed soil samplesof the surface layer (0–5 cm) were collected with asoil auger from five randomly selected positions.These samples were then pooled to produce onerepresentative sample. The samples were air-dried,weighed and then sieved to 2 mm to separate thegravel (>2 mm) from the fine soil (<2 mm). Theformer was reweighed to determine the gravel con-tent. The soil component was separated into twoparts. One was analysed for particle size by laserdiffraction with a Mastersizer 2000 (Malvern). Theother was crushed and passed through a 0.25-mmsieve for determining the amount of soil organiccarbon (SOC) with the dichromate-oxidation method(Nelson and Sommer 1982). Undisturbed soil coresamples (100 cm3) were collected from the surfacelayer (0.5–5.5 cm) at each point for measurement ofsaturated hydraulic conductivity (KS), using the con-stant hydraulic head method (Klute and Dirksen1986), and of soil bulk density (BD), based on thevolume of each original soil core and the total weightof the soil after oven drying at 105°C for 48 h.

2.3 Statistical methods

To clarify the descriptions, we will first define theterms used herein:

– ‘micro-site’ is the ground location at which ameasurement was taken (three measurementswere done at each point);

– ‘point’ is the mean location of a group of micro-sites; and

– ‘sampling day’ represents an individual day onwhich several measurements were taken.

A descriptive statistical analysis was first appliedto determine the overall trends and variation of theSWC. Let θijk be the SWC at micro-site i and pointj on sampling day k. The spatial mean at point j onsampling day k is then calculated as:

θjk ¼ 1

np

Xnpi¼1

θijk (1)

where np is the number of micro-sites at point j. Thespatial mean for each sampling day k, θk , and thetemporal mean at each point j, θj, can thus begiven by:

θk ¼ 1

n

Xni¼1

θjk (2)

θj ¼ 1

m

Xmk¼1

θjk (3)

where n and m are the number of sampling points anddays, respectively.

The coefficient of variation of each sampling dayin space, CVk, and each sampling point in time, CVj,can consequently be defined as:

CVk ¼ σkθk

¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1

n�1

Pnj¼1 θjk � θk

� �2qθk

(4)

CVj ¼ σjθj

¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1

m�1

Pmk¼1 θjk � θj

� �2qθj

(5)

where σk and σj are the standard deviations in spaceand time, respectively.

For each sampling day, the coefficient of varia-tion of the sampling point, CVk

point, was calculated asthe mean of the coefficients of variation determinedfor each point as:

CVpointk ¼ 1

n

Xnj¼1

CVjk (6)

where CVjk is the coefficient of variation of thesampling point j on sampling day k.

With an understanding of the spatial standarddeviation (σk), we can quantify the minimum amountof sampling needed for estimating the field mean

Spatio-temporal variability of surface soil water content 99

Dow

nloa

ded

by [

Nor

thw

est A

& F

Uni

vers

ity]

at 2

0:41

24

Dec

embe

r 20

14

water content at a given level of confidence, whichcan be described by:

N ¼ λ2α;fσkkμ

� �2

(7)

where λα;f is the limiting value of the t-distributionwithin a level of confidence 1 – α (α is the level ofsignificance) and with f (f = N – 1) degrees of free-dom, µ is the mean value of the SWC (%) and k isthe permitted relative error.

One-sample Kolmogorov-Smirnov (K-S) testswere used to examine the normality of the data.Semivariograms were then constructed to evaluatethe spatial dependence of SWC (David 1977).Correlation analysis was conducted for each day toevaluate the possible relationships between SWC andrelative elevation, soil properties and vegetation.Stepwise linear regression analysis was subsequentlycarried out to detect the relative significance of thesefactors influencing the dynamics of the SWC. Alldescriptive statistical analyses used the programSPSS 16.0. The geostatistical analysis was performedwith GS+ software (version 7.0, Gamma DesignSoftware).

3 RESULTS AND DISCUSSION

3.1 Temporal dynamics of SWC

The mean SWC had apparent seasonal changes andwas tightly linked to rainfall (Fig. 2). The mean watercontent was higher in the rainy season (summer) thanin dry seasons (spring and autumn) but never roseabove 10%. Rain is the main source of soil moisturein this area because of the deep water tables. Because

this area is characterized by low precipitation andhigh evapotranspiration, the soil is subjected todrought most of the time.

The mean SWC generally increased after a sig-nificant rainfall and decayed rapidly during dry per-iods. Relatively larger rainfalls always correspondedto higher SWCs, but this relationship was largely dueto the timing of the sampling relative to the periodsof rain. The higher SWCs occurred immediately aftera rainfall event (e.g. 31 July and 15 August), with alag following the rain. As seen in Fig. 2, the twomost significant rainfall events occurred on 5 June(15.8 mm) and 27 June (14.4 mm), but did not giverise to the highest SWCs, which were on 18 June and30 June, long after the rains. Rain wets the soil easily,and the water content accordingly increases quickly.The amount of water in the surface soil thendecreases rapidly due to drainage to the sublayerand due to high evapotranspiration. The water con-tent of the surface soil is to a very large extentinfluenced by the timing, amount and intensity ofthe rain. The influence of prior conditions of SWC,though, should also be considered. A wetter antece-dent condition can produce greater water contentwhen comparing rainfall events with the sameamount and intensity (Pan et al. 2008). This phenom-enon was not very evident in this study because ofthe low frequency of data collection.

3.2 Statistical analysis of SWC

3.2.1 Statistical parameters The variability ofSWC in space and time was surveyed by calculatingthe temporal mean of CV of the spatial SWC, CVk,and the spatial mean of CV of the time series ofSWC, CVi. The value of CVk ranged from 26.57%

0

3

6

9

12

15

18

17-A

pr

2-M

ay

17-M

ay

1-Ju

n

16-J

un

1-Ju

l

16-J

ul

31-J

ul

15-A

ug

30-A

ug

14-S

ep

29-S

ep

14-O

ct

Time (Date)

Mea

n so

il w

ater

con

tent

(%

)

0

10

20

30

40

50

60

Pre

cipi

tatio

n (m

m)Precipitation

Mean

Fig. 2 Daily precipitation depth (mm) and mean soil water content (SWC, %) in the surface (0–6 cm) layer vs time.

100 Pingping Zhang and Ming’an Shao

Dow

nloa

ded

by [

Nor

thw

est A

& F

Uni

vers

ity]

at 2

0:41

24

Dec

embe

r 20

14

to 63.78% (Table 1), with a mean of 46.09%, indicat-ing that SWC in this region was moderately variable.Based on the results of Hills and Reynolds (1969),CVk should remain above 5% for any specific field.In our study, all values were larger than 25%, whichmay have been due to the low amount of SWC in thisarea. In contrast, the mean CVi was 60.42% (rangingfrom 40.61% to 97.9%), which was obviously largerthan the mean CVk, suggesting that the temporalvariability of SWC should receive more attentionthan spatial variability in studies of SWC (Broccaet al. 2010). Coarse spatial sampling at a high fre-quency may thus be a good option for modellinghydrological processes in this region.

Statistical parameters were tested for the fieldmean water content, θk , to determine if the variabilityof surface soil water was related to the conditions ofwetness. Figure 3(a) shows increasing standarddeviation (σk) with increasing mean water content.This result is in accordance with previous studiesby Famiglietti et al. (1998), Western et al. (1998)

and Li et al. (2013), but is contrary to the studies ofMeyles et al. (2003) and Brocca et al. (2007), whostated that the relationship between σk and meanwater content is linked to climatic conditions; inhumid regions, σk is larger when the soil is dry,while in semi-arid regions, σk increases as the soilbecomes wetter. Vereecken et al. (2007) demon-strated that the change of σk with soil wetness waslargely determined by the ability of soil to retainwater and by the associated spatial variability.

An inherent characteristic of σk, though, is itsstrong dependence on θk , which usually restricts itsapplication when comparing datasets with widelyvarying mean water contents. To overcome this diffi-culty, the relative variance, i.e. CVk, can be used,which can remove the influence of the differentmeans. The CVk exhibited a general convex upwardrelationship with θk , with its highest value at amoderate level of wetness, as shown in Fig. 3(b).The variability of SWC is mainly regulated bythe wilting point in extremely dry conditions, the

Table 1 Statistical properties of the data for surface soil (0–6 cm) water content during the observation period (15 April–15October 2012).

Date Minimum (%) Maximum (%) Mean (%) σ (%) CV (%) Skewness Kurtosis K-S test

16 Apr 1.56 8.32 3.37 1.50 44.41 1.51 1.66 0.00030 Apr 1.16 4.84 2.15 0.82 38.31 1.44 1.63 0.00015 May 1.41 7.73 3.06 1.40 45.65 1.48 1.61 0.00029 May 0.83 4.99 1.94 0.93 47.87 1.44 1.63 0.00018 Jun 1.31 6.64 2.53 1.00 39.70 1.58 2.68 0.00030 Jun 1.62 13.95 4.63 2.47 53.38 1.40 1.78 0.00021 Jul 1.46 19.90 4.94 2.75 55.57 2.00 5.50 0.00031 Jul 4.33 22.16 8.90 4.60 51.70 1.38 2.08 0.02715 Aug 1.83 30.08 6.93 4.42 63.78 1.87 4.76 0.0013 Sep 1.58 12.90 2.71 1.35 49.96 3.75 19.99 0.00016 Sep 0.94 6.95 2.42 1.01 41.60 1.73 3.83 0.0032 Oct 0.77 6.84 2.28 0.93 40.83 1.75 4.11 0.00216 Oct 0.27 4.91 2.07 0.55 26.36 1.34 5.32 0.000

σ: standard deviation; CV: coefficient of variation.

y = 2.7069Ln(x) – 1.3762

R2 = 0.9549

0.0

1.0

2.0

3.0

4.0

5.0

0 2 4 6 8 10

Mean water content (%)

Sta

ndar

d de

viat

ion

(%)

0

16

32

48

64

80

0 2 4 6 8 10

Mean water content (%)

Coe

ffici

ent o

f var

iatio

n (%

)(a) (b)

Fig. 3 Relationships between mean soil water content (SWC, %) and (a) standard deviation (%) and (b) coefficient ofvariation (%) in the study area.

Spatio-temporal variability of surface soil water content 101

Dow

nloa

ded

by [

Nor

thw

est A

& F

Uni

vers

ity]

at 2

0:41

24

Dec

embe

r 20

14

hydraulic conductivity in mid-range conditions andthe soil porosity in wet conditions (Lawrence andHornberger 2007). In moderate conditions, smallpatches of quickly drying soil (generally character-ized by coarse texture) may coexist with patches thatremain wet (generally characterized by fine texture),leading to heterogeneous conditions of wetness (Hillsand Reynolds 1969). This situation, though, is notcommon, as demonstrated by the above studies thatindicated a universal negative relationship between

CVk and θk over all conditions of wetness. However,this negative relationship may be due to the limitedperiods of measurement or to the low frequencies ofmeasurement in these studies, which would limit thedata to an incomplete range of soil wetness. Therelatively wetter climates of their study areas couldfurther restrict the range of SWCs.

The relationship between σk and θk can be welldescribed logarithmically as: σk ¼ 2:7069 lnðθkÞ�1:3762, which can quantify the amount of sampling

needed to determine θk at given confidence levelsand relative errors. The sampling number as a func-tion of θk for predefined confidence levels of 95%and 90% and relative errors of 5, 10 and 15% isshown in Fig. 4. The required amount of samplinggenerally exhibited higher values for the 95% thanfor the 90% confidence level and declined slightly asthe allowable relative error increased from 5% to15%. In accordance with other studies, these resultsindicated that many more samples are needed in

relatively drier conditions for estimating θk in thestudy area. The maximum number of samples neededcoincided with a water content of approximately4.5%. The number of sampling points in our study

(187) was sufficient to meet the requirement of arelative error of ±10% at a 95% confidence level,which had a maximum sampling number of approxi-mately 150.

3.2.2 Variability of SWC and size of areaEstimating the variability of SWC over a wide vari-ety of scales has significant implications for develop-ing an effective scheme of data acquisition andmonitoring. For each sampling date, the coefficientof variation of the sampling point (three measure-ments at each point), CVk

point, was calculated as themean of the coefficients determined for each pointand was compared with CVk of the entire area. Thevalue of CVk

point ranged from 7.34% to 27.99% andaveraged 17.12%, which was considerably lowerthan CVk. This comparison suggests an increasingtrend of variability as the size of the investigatedarea increases. As displayed in Fig. 5, the coefficientof variation for the entire study area had a significantlinear relationship with the coefficient at the pointscale (i.e. CVk and CVk

point), and the determiningcoefficient was as high as 0.78, implying that thechange of spatial variability at the point scale followsa trend similar to that of the entire area: when thelatter is high, so is the former.

To further detect the tendency of the variabilityof SWC to change with the size of the study area, aseries of sampling-point allocations of different sizeswere re-sampled using all sampling points (n = 187)in the study area (5 km × 8 km). For convenience, theeast−west and north−south spans of the re-samplingarea were assigned integral multiples of 1 km togenerate five re-sampling options for the west−east

0

100

200

300

400

500

600

0 5 10 15 20 25 30Mean water content (%)

Num

ber

of s

ampl

es /4

0 km

2

95% confidence level

90% confidence level

± 10%

± 5%

± 15%

Fig. 4 Number of samples (per 40 km2) for accurate measurement of field mean SWC (%) at 95% and 90% confidencelevels and for relative errors of ±5%, ±10% and ±15% in the study area.

102 Pingping Zhang and Ming’an Shao

Dow

nloa

ded

by [

Nor

thw

est A

& F

Uni

vers

ity]

at 2

0:41

24

Dec

embe

r 20

14

span and eight options for the south−north span. Therandom combination of these two sets of options thusyielded 40 potential re-sampling methods. Some ofthese methods, though, were the same within thestudy area, so a final total of 24 types of re-samplingareas were obtained (see Table 2). The variability ofSWC for an area was first computed by averaging theCVk values of all possible re-sampling scenarios withthe same area for each sampling date, and the result-ing data were then averaged over the observationperiod. The averaged spatial CVk clearly increasedwith the size of area and could be well parameterizedas a power function of the size (Fig. 6(a)). Thefactors influencing the variability were scale depen-dent. As the size increases, the causes of variation,such as vegetation, topographical parameters andparental material, may become increasingly complexand heterogeneous, leading to greater variability. Thefractal power parameter of the function was 0.0838,nearly half the value obtained by Brocca et al.

(2012), who discussed the changes of spatial CVk

with sizes of between 1 m2 and 250 km2. The differ-ent increases of spatial CVk with size may be attrib-uted to the characteristics of the different regions orto the different methods used to obtain the results.The conclusions of Brocca et al. (2012) were mainlyderived from the information reported in previouslypublished studies. The spatial CVk of the differentsizes was determined for different times, so the finalresults are not likely to be very accurate due to thechanges in spatial CVk over time.

We also discovered that the spatial CVk had aconvex upward pattern with the field mean watercontent independent of the point scale or the othersize scales, confirming that the relationships betweenthe two statistical parameters were characterized bysimilar behaviours at different scales. The variabilityof the spatial CVk over time, however, differed atdifferent scales. As seen in Fig. 6(b), the coefficientof variability of the CVk over time significantlydecreased with increasing size. This relationshipmay be due to the dependence on scale of the factorsinfluencing the variability of surface water content.How they change over time is also dependent onscale.

3.3 Geostatistical analysis of SWC

The coefficients of skewness and kurtosis, togetherwith the non-parametric K-S test of the observations(Table 1), indicated that the SWCs for all dates werenot normally distributed, so we logarithmically trans-formed the data. The transformed data for all datespassed the K-S test at the 0.05 significance level andconsequently could be used for geostatistical ana-lyses. The semivariograms of the SWCs were opti-mally fitted by exponential models; the structuralparameters are given in Table 3. All models were

-

8

16

24

32

0 20 40 60 80Coefficient of variation of the entire area, CVk (%)

Coe

ffici

ent o

f var

iatio

n of

the

sam

plin

gpo

int,

CV

kpoin

t (%

)

Fig. 5 Relationship between the spatial coefficients ofvariation computed for the entire area, CVk (%), and theaverage of the coefficients computed for each samplingpoint, CVk

point (%), in the study area.

y = 33.541x0.0838

R2 = 0.9121

30

34

38

42

46

50

0 10 20 30 40 50

Area (km2)0 10 20 30 40 50

Area (km2)

Spa

tial c

oeffi

cien

t of v

aria

tion,

CV

k (%

)

y = –0.0415x + 22.739

R2 = 0.4362

19

20

21

22

23

24

Coe

ffici

ent o

f var

iatio

nof

CV

k (%

)

(a) (b)

Fig. 6 Relationship between the size of the study area (km2) and (a) the coefficient of variation of SWC (%) and (b) thecoefficient of variation of CVk (%).

Spatio-temporal variability of surface soil water content 103

Dow

nloa

ded

by [

Nor

thw

est A

& F

Uni

vers

ity]

at 2

0:41

24

Dec

embe

r 20

14

satisfactory, with R2 varying from 0.572 to 0.917,suggesting that the theoretical models well repre-sented the space structure feature of the surfaceSWC in the study area.

The nugget effect, C0, representing the variationat zero distance, ranged from 0.007 to 0.0648, per-haps due to an undetected sampling error, a finer-scale variability or an inherent random heterogeneity.The sill values, C0 + C, representing the total varia-tion, ranged from 0.105 to 0.375. The differences inC0 and C0 + C over time markedly influenced thedegree of soil-water heterogeneity under differentwater conditions. Both C0 and C0 + C had convexupward relationships with the field mean water con-tent, θk (Fig. 7(a), (b)). Soil water after a significantrainfall can move easily on the soil surface, so thedistribution of SWC is relatively uniform over the

study area. However, irregular upward evapotran-spiration and downward infiltration during periodsof drying become increasingly significant forces forthe distribution of soil water, increasing its spatialheterogeneity. When the SWC falls below the levelat which C0 or C0 + C is maximal, evapotranspirationis then water limited, and the distribution pattern ismainly controlled by the wilting coefficient. The wilt-ing coefficient is distributed quite uniformly acrossthe study area and thus leads to a reduction in spatialheterogeneity. Pan et al. (2007) and Gao et al. (2011)also demonstrated that the spatial heterogeneity ofSWC peaked in moderate conditions of wetness andthen decreased independent of further soil drying orwetting. The degree of spatial heterogeneity of SWCis determined by the percentage of total variationexplained by the systemic variation, i.e. C/(C0 + C).This ratio varied with soil wetness, with values ran-ging between 0.731 and 0.933 (Fig. 7(d)), indicatingthat SWC was strongly to moderately spatially depen-dent. A strong spatial dependence is generally causedby intrinsic factors such as soil parental material andtopographical parameters.

The range value, A, also changed significantlywith mean SWC (Fig. 7(c)). These changes werelikely due to the control of the distribution of soilmoisture by different hydrological processes underdifferent conditions of wetness. The sampling con-ducted immediately following a heavy rainfall event(e.g. 31 July and 15 August) presented relatively highrange values (7185 and 8646 m) (Table 3). Theseresults agree with the findings of Grayson et al.(1997) and Brocca et al. (2007) that SWC underwet conditions can exhibit much better distributionsof spatial continuity. The high range under wet

Table 2 Details of the re-sampling areas and samplingmethods.

Re-samplingarea (km2)

Sampling method Re-samplingarea (km2)

Samplingmethod

1 1 × 1* 15 3 × 5; 5 × 32 1 × 2; 2 × 1 16 2 × 8; 4 × 43 1 × 3; 3 × 1 18 3 × 64 1 × 4; 4 × 1; 2 × 2 20 4 × 5; 5 × 45 1 × 5; 5 × 1 21 3 × 76 1 × 6; 2 × 3; 3 × 2 24 3 × 8; 4 × 67 1 × 7 25 5 × 58 1 × 8; 2 × 4; 4 × 2 28 4 × 79 3 × 3 30 5 × 610 2 × 5; 5 × 2 32 4 × 812 2 × 6; 3 × 4; 4 × 3 35 5 × 714 2 × 7 40 5 × 8

* The digit before the multiplication sign represents the west–east sam-pling distance (km), and the digit after it represents the south–northsampling distance (km).

Table 3 Parameters of the semivariogram models for the surface soil (0–6 cm) water contentduring the observation period. The semivariograms were optimally fitted by exponential models.

Date Nugget, C0 (%2) Sill, C + C0 (%

2) C/(C0 + C) Range, A (m) R2

16 Apr 0.036 0.137 0.734 1503 0.57230 Apr 0.007 0.105 0.933 1359 0.62215 May 0.019 0.146 0.869 1485 0.65329 May 0.013 0.165 0.919 1476 0.66018 Jun 0.008 0.116 0.932 2085 0.78530 Jun 0.065 0.241 0.731 4368 0.81821 Jul 0.025 0.220 0.891 1269 0.75531 Jul 0.037 0.174 0.790 8646 0.91715 Aug 0.059 0.375 0.843 7185 0.8463 Sep 0.010 0.122 0.919 2874 0.69416 Sep 0.021 0.151 0.858 6057 0.8572 Oct 0.023 0.140 0.839 6234 0.84816 Oct 0.026 0.108 0.760 9342 0.824

R2: coefficient of determination.

104 Pingping Zhang and Ming’an Shao

Dow

nloa

ded

by [

Nor

thw

est A

& F

Uni

vers

ity]

at 2

0:41

24

Dec

embe

r 20

14

conditions may be attributed to spatially uncorrelatedsampling errors and can play a relatively more sig-nificant role in these wet conditions (Brocca et al.2012). As the soil dried, even with sporadic light rainduring the drying period, the surface SWC tended tobe much more independent and was characterized bya stochastic pattern of water content (Fig. 7(c)). Forexample, on 30 June, three days after the last largerainfall event (14.4 mm on 27 June), the rangedecreased to 4368 m. On 18 June, 13 days after thelast heavy rainfall (15.8 mm on 5 June, but with lightintervening rain), the range was only 2085 m(Table 3). When the soil dries enough for the wiltingpoint to control the distribution of water content,however, the range can again increase. As displayedin Table 3, the range increased from 2874 to 9342 mbetween 3 September and 16 October.

3.4 Factors regulating the spatial pattern ofSWC

Correlation analysis was conducted for all dates toevaluate the possible relationships between SWC andpotentially associated properties (relative elevation,soil properties and vegetation). The coefficients arepresented in Table 4 and Table 5 is the correlationmatrix of these associated properties. Significant cor-relations were observed between some of theseproperties.

3.4.1 Relative elevation Relative elevation isan important topographical property that always var-ies jointly with some other soil and topographicalproperties (e.g. the specific contributing area, claycontent, etc.), which can significantly regulate thepattern of surface soil water by its effect on the lateralredistribution of soil water (e.g. infiltration and run-off). Relative elevation was negatively correlated withsurface SWC in this study (Fig. 8), in agreement withother studies (Qiu et al. 2001, Hébrard et al. 2006)that highlighted a declining trend of SWC withincreasing relative elevation. The magnitude of thenegative correlation generally increased with anincrease in mean water content (Fig. 8). Accordingto Pan et al. (2008), rain can increase the relativesignificance of elevation on the spatial pattern of sur-face SWC, because elevation becomes the dominantfactor governing the redistribution of soil water imme-diately after a rainfall event. The effect of relativeelevation, though, can quickly be replaced by strongevapotranspiration soon after a rainfall event. Thedegree of association between SWC and relative ele-vation in drier conditions when redistribution is minoror absent may thus decrease (Grayson et al. 1997).However, relative elevation also correlated stronglywith soil and vegetation properties (Table 5), so theincreasingly weaker correlations as soil dries are alsolikely to be the consequence of the joint contributionsof soil and vegetation properties.

0.00

0.02

0.03

0.05

0.06

0.08

0 2 4 6 8 10Mean water content (%)

Nug

get,

C0

(%2 )

0.00

0.08

0.16

0.24

0.32

0.40

0 2 4 6 8 10Mean water content (%)

Stil

l, C

+C0

(%2 )

0

2000

4000

6000

8000

10000

0 2 4 6 8 10

Mean water content (%)

Ran

g, A

(m

)

0.50

0.60

0.70

0.80

0.90

1.00

0 2 4 6 8 10

Mean water content (%)C

/ C+C

0

(a)

(c) (d)

(b)

Fig. 7 Relationship between mean SWC (%) and (a) the geostatistical parameters nugget, C0 (%2), (b) sill, C + C0 (%

2), (c)range, A (m) and (d) C/(C + C0) in the study area.

Spatio-temporal variability of surface soil water content 105

Dow

nloa

ded

by [

Nor

thw

est A

& F

Uni

vers

ity]

at 2

0:41

24

Dec

embe

r 20

14

3.4.2 Soil properties Soil BD, soil texture,gravel content, SOC concentration and KS havebeen considered as the main soil properties regulatingthe distribution of SWC. Table 4 shows remarkablecorrelations between water content and most of thesoil properties. These correlations confirm the impor-tance of the spatial heterogeneity of soil properties tothe variability of SWC.

In our study, BD and soil texture were the mostcrucial properties regulating the distribution of sur-face SWC. The SWC was highly inversely correlatedwith BD and sand content but was positively cor-related with silt and clay contents for all dates(Table 4). The BD strongly reflects soil structureand the distribution of large pores and thus plays amajor role in soil saturation and hydraulic conductiv-ity. Soil texture can influence hydraulic propertiesand the ability of soil to retain water and can hencehave a particularly significant effect on the distribu-tion pattern of SWC. These results are supported byother studies (Zhao et al. 2010, Gao and Shao 2012),which also found strong correlations between SWCand soil texture and BD. Increased SOC contentgenerally leads to a better-aggregated soil structureand a lower BD, and therefore encourages anenhanced capacity of soil to retain water, which canconsiderably reduce the loss of water (mainly byevapotranspiration and deep percolation).

Compared with the soil properties mentionedabove, gravel content and KS had much weakereffects on the distribution of SWC. The relationshipbetween KS and SWC did not even reach a significantlevel on some drier days (Table 4). Variations instone content and KS in this area may thus at besthave a much lower influence on the distribution ofSWC. Correlations with soil texture were largelyunaffected by soil wetness, which were nearly con-stant over the complete range of soil wetness (Fig. 8).This result agrees with that of Gómez-Plaza et al.(2001), who found that the role of soil texture in thedistribution of SWC did not vary with seasonal var-iations of soil wetness condition. The strong correla-tion with SWC thus suggests that soil texture isimportant in both wet and dry conditions. In contrast,the correlations between SWC and other soil proper-ties (BD, gravel content, KS and SOC concentration)were affected by soil wetness. The correlations withBD, gravel content and KS were stronger in wetconditions, and the correlation with SOC concentra-tion was stronger in dry conditions. These resultssuggest that the relative significance of these soilproperties in regulating the distribution of SWCT

able4Pearson

’scorrelationcoefficientsbetweensurfacesoil(0–6

cm)water

contentand

topographical,soilandvegetatio

npropertiesduring

theobservationperiod.*

and**

indicate

that

thecorrelationisstatistically

significantat

probability

levelsof

5%and1%

,respectiv

ely.

16Apr

30Apr

15May

29May

18Jun

30Jun

21Jul

31Jul

15Aug

3Sep

16Sep

2Oct

16Oct

Relativeelevation

−0.35

1**

−0.35

8**

−0.35

3**

−0.34

6**

−0.39

9**

−0.43

2**

−0.25

3**

−0.52

2**

−0.43

0**

−0.26

2**

−0.40

9**

−0.38

9**

−0.29

8**

BD

−0.63

5**

−0.66

1**

−0.66

4**

−0.64

1**

−0.65

1**

−0.68

9**

−0.59

6**

−0.72

0**

−0.62

4**

−0.53

2**

−0.60

9**

−0.59

1**

−0.55

6**

Gravel

−0.20

7**

−0.22

7**

−0.21

5**

−0.22

0**

−0.21

7**

−0.23

1**

−0.15

6**

−0.25

9**

−0.21

3**

−0.12

6**

−0.16

5**

−0.16

5**

−0.14

9**

Sand

−0.55

6**

−0.55

0**

−0.56

2**

−0.53

4**

−0.60

8**

−0.62

1**

−0.46

6**

−0.58

0**

−0.54

2**

−0.47

4**

−0.55

1**

−0.54

6**

−0.51

3**

Silt

0.53

6**

0.54

0**

0.54

7**

0.52

4**

0.59

7**

0.60

8**

0.45

2**

0.58

4**

0.53

0**

0.44

6**

0.53

9**

0.52

6**

0.50

1**

Clay

0.55

9**

0.54

6**

0.56

1**

0.53

0**

0.60

3**

0.61

6**

0.46

7**

0.57

3**

0.54

0**

0.48

5**

0.54

8**

0.54

9**

0.51

2**

SOC

0.42

0**

0.43

2**

0.42

3**

0.42

0**

0.48

3**

0.46

8**

0.38

3**

0.42

1**

0.41

8**

0.42

0**

0.47

1**

0.48

4**

0.48

2**

KS

−0.18

9**

−0.17

8**

−0.19

3**

−0.19

3**

−0.14

6*−0.18

6**

−0.19

1**

−0.19

7**

−0.18

9**

−0.13

0−0.14

0−0.13

8−0.17

6*Species

richness

−0.25

6**

−0.26

8**

−0.26

9**

−0.26

0**

−0.22

4**

−0.24

0**

−0.15

0*−0.32

1**

−0.23

6**

−0.12

5−0.20

1**

−0.18

8**

−0.12

8Shrub

coverage

−0.10

9−0.10

7−0.12

3−0.116

−0.08

6−0.07

1−0.13

90.03

7−0.06

1−0.15

8*−0.05

7−0.08

6−0.10

8Herbcoverage

−0.04

9−0.04

5−0.02

5−0.04

6−0.05

9−0.03

2−0.00

30.08

40.15

6*−0.09

4−0.07

5−0.09

8−0.08

8

BD:bulk

density

;SOC:soilorganiccarbon

concentration;

KS:soilsaturatedhydraulic

conductiv

ity.

106 Pingping Zhang and Ming’an Shao

Dow

nloa

ded

by [

Nor

thw

est A

& F

Uni

vers

ity]

at 2

0:41

24

Dec

embe

r 20

14

Relative elevation

y = –0.0199x – 0.2959R2

= 0.3328

–0.55

–0.47

–0.39

–0.31

–0.23

–0.150 2 4 6 8 10

BD

y = –0.0118x – 0.5848

R2 = 0.2349

–0.80

–0.72

–0.64

–0.56

–0.48

–0.400 2 4 6 8 10

Gravel

y = –0.0085x – 0.1648

R2 = 0.2125

–0.40

–0.32

–0.24

–0.16

–0.08

0.000 2 4 6 8 10

Sandy = –0.0033x – 0.5342

R2 = 0.0252

–0.80

–0.72

–0.64

–0.56

–0.48

–0.400 2 4 6 8 10

Silt

y = 0.0052x + 0.5137

R2 = 0.0537

0.40

0.48

0.56

0.64

0.72

0.80

0 2 4 6 8 10

Clay

y = 0.0027x + 0.5355

R2 = 0.0187

0.40

0.48

0.56

0.64

0.72

0.80

0 2 4 6 8 10

SOC

y = –0.0064x + 0.4641

R2 = 0.1747

0.20

0.28

0.36

0.44

0.52

0.60

0 2 4 6 8 10

KS

y = –0.0057x – 0.1518

R2 = 0.2426

–0.40

–0.32

–0.24

–0.16

–0.08

0.000 2 4 6 8 10

Species richness

y = –0.0122x – 0.179

R2 = 0.2223

–0.40

–0.32

–0.24

–0.16

–0.08

0.000 2 4 6 8 10

Shrub coverage

y = 0.0155x – 0.1468

R2 = 0.4765

–0.30

–0.22

–0.14

–0.06

0.02

0.100 2 4 6 8 10

Herb coverage

y = 0.0298x – 0.1388

R2 = 0.757

–0.20

–0.12

–0.04

0.04

0.12

0.200 2 4 6 8 10

Mean soil water content (%)

Cor

rela

tion

coef

ficie

nt

Fig. 8 Correlation coefficients between SWC and relative elevation, soil properties (gravel, sand, silt, clay and SOCcontent, BD and KS) and vegetation (shrub coverage, herb coverage and species richness) plotted against mean SWC (%).

Table 5 Pearson’s correlation coefficients among topographical, soil and vegetation properties. * and ** indicate that thecorrelation is statistically significant at probability levels of 5% and 1%, respectively.

Relativeelevation

BD Gravel Sand Silt Clay SOC KS Speciesrichness

Shrubcoverage

Herbcoverage

Relative elevation 1.000 0.375** 0.386** 0.469** −0.471** −0.456** −0.267** −0.116 0.397** −0.336** −0.201**BD 1.000 0.409** 0.659** −0.671** −0.634** −0.413** 0.146* 0.307** −0.058 −0.004Gravel 1.000 0.448** −0.458** −0.429** −0.196** 0.145* 0.402** −0.230** −0.176*Sand 1.000 −0.984** −0.990** −0.679** 0.176* 0.336** −0.234** −0.004Silt 1.000 0.949** 0.677** −0.167* −0.329** 0.229** 0.017Clay 1.000 0.665** −0.179* −0.333** 0.232** −0.006SOC 1.000 −0.158* −0.115 0.022 −0.108KS 1.000 0.211** 0.023 0.078Species richness 1.000 −0.140 −0.092Shrub coverage 1.000 0.325**Herb coverage 1.000

BD: bulk density; SOC: soil organic carbon concentration; KS: soil saturated hydraulic conductivity.

Spatio-temporal variability of surface soil water content 107

Dow

nloa

ded

by [

Nor

thw

est A

& F

Uni

vers

ity]

at 2

0:41

24

Dec

embe

r 20

14

may vary with wetness. The BD can be reasonablyexpected to influence the distribution of SWC imme-diately after a significant rainfall via its relationshipwith soil porosity and hydraulic conductivity.Moreover, the presence of gravel on the soil surfacecan further promote to some degree the percolation ofsoil water to lower layers (Williams et al. 2003). Assoil dries, evaporation and plant transpiration gradu-ally become the dominant hydraulic processes, andSOC and clay content may then become more impor-tant in the distribution of SWC because of theirhigher capacities to retain water.

3.4.3 Vegetation properties Vegetation regu-lates the pattern of SWC mainly by interceptingrain, shading the soil surface, extracting soil waterfor transpiration, increasing the infiltration of soilwater and reducing soil temperatures. Also, plantlitter and root residues accumulate on the soil surfaceand thereby prolong the periods of increased levels ofsoil water. Species composition affects interspecificcompetition for the limited water resources and maythus play a significant role in the evaporation andstorage of SWC.

Surprisingly, we found no obvious dependenceof SWC on the coverages of shrubs and herbs (lowcorrelation coefficients, as shown in Table 4), excepton a few days. Other studies have indicated that thedynamics of SWC should be strongly influenced byvariations in vegetation coverage (Reynolds 1970,Hawley et al. 1983, Bhark and Small 2003). Therelatively weak dependence of SWC on vegetationcoverage in the present study might be ascribed to thevery limited amount of data collected (one time inmid-August). The correlations based on these limiteddata cannot represent the entire season. As shown inTable 4, SWC was significantly correlated with herbcoverage on 15 August, near the date when vegeta-tion data were collected. On the other hand, SWCwas significantly and negatively correlated with spe-cies richness (Table 4). The presence of more specieswill increase the intensity of interspecific competitionfor the scarce water and will consequently acceleratethe loss of soil water.

The correlation coefficients for the vegetationproperties displayed a clearly linear function forfield mean water content (Fig. 8), likely due to thedifferential regulation of soil-water distribution byvegetation at different stages of growth. The effectof transpiration was illustrated by the negative corre-lation between SWC and shrub coverage under dryconditions. In contrast, shrub coverage was positively

correlated with soil water under wet conditions, per-haps caused by the interception of rainfalls. Thiswould happen, but is not likely if just a small rainfallevent occurs, because the high temperature andabsence of wind would enable rapid evaporationand prevent the rain from reaching the soil surface(Bhark and Small 2003). The herbs in this region aremainly annuals, such as B. dasyphylla, H. arachnoi-deus and S. ruthenica, which prefer locations withhigh water contents. Herb coverage was thus posi-tively correlated with field mean water content(Fig. 8). Interspecific competition also increaseswith higher SWC, as shown by the strong correlationbetween species richness and SWC (Fig. 8).

3.5 Stepwise linear regression analysis

The distribution of SWC was not regulated by asingle factor but was governed in more complicatedand comprehensive ways because many of the con-trolling factors were interdependent. Stepwise linearregression analysis was performed to find the factorsthat could best predict the dynamics of SWC. Toinvestigate how combinations of factors changedwith the seasonal variations of soil wetness, we sepa-rated the measurement campaigns into three groupsbased on the conditions of wetness: wet (θk > 5%),medium (θk = 3–5%) and dry (θk < 3%).

Table 6 shows that the factors regulating thedistribution of SWC varied with wetness, althoughBD was the most dominant factor in all threeconditions. Under wet conditions, relative elevationand clay content were also important factors influ-encing the pattern of SWC. Under dry conditions,clay content and shrub coverage became moreimportant through the ability to retain water andto transpire, respectively. Stone content and KS alsoreached significant levels, but played minor rolesin the variability of SWC compared to other soilproperties. These results have led us to the conclu-sion that the major factors controlling the dynamicsof SWC changed from combinations of topographi-cal and soil properties under wet conditions tocombinations of soil and vegetation propertiesunder dry conditions.

4 CONCLUSIONS

This study surveyed the variability of surface soil(0–6 cm) water content in space and time, and iden-tified the factors influencing the distribution of SWC

108 Pingping Zhang and Ming’an Shao

Dow

nloa

ded

by [

Nor

thw

est A

& F

Uni

vers

ity]

at 2

0:41

24

Dec

embe

r 20

14

in a 40-km2 area of the Gobi Desert. The spatialvariability of SWC increased during drying after awet period, peaked at specific SWCs and thendecreased during further drying. The minimumamount of sampling needed coincided with a watercontent of approximately 4.5%. A re-samplingmethod further indicated that the spatial variabilityof SWC increased with the expansion of area andcould be well parameterized as a power function ofthe size of the sampling area. Geostatistical analysisindicated that SWC exhibited a variable spatialdependence with time. It had a stronger spatial pat-tern of organization following significant rainfallsand in extremely dry conditions but had a stochasticpattern under moderate conditions of wetness.Correlation analysis implied that the dynamics ofSWC in this area were linked to relative elevation,soil properties and vegetation. However, the domi-nant factors influencing the dynamics of SWC largelydepended on the average state of SWC. Stepwiseregression analysis further indicated that SWC wasstrongly associated with topographical and soil prop-erties (BD, relative elevation and clay content) underwet conditions, but soil and vegetation properties(BD, clay content and shrub coverage) controlledthe variability of SWC under dry conditions. Thisstudy is applicable to the description of soil-waterdistribution in arid regions, which may have impor-tant implications for sampling design, hydrological

modelling, environmental protection and sustainabil-ity of land use.

Funding This work was supported by grants fromthe Major Programme of the National NaturalScience Foundation of China (no. 91025018).

REFERENCES

Ampofo, E.A., 2006. Soil moisture dynamics in coastal savanna soilsin the tropics under different soil management practices.Hydrological Sciences Journal, 51 (6), 1194–1202.doi:10.1623/hysj.51.6.1194

Berndtsson, R. et al., 1996. Soil water and temperature patterns in anarid desert dune sand. Journal of Hydrology, 185, 221–240.doi:10.1016/0022-1694(95)02987-7

Bhark, E.W. and Small, E.E., 2003. Association between plant cano-pies and the spatial patterns of infiltration in shrubland andgrassland of the Chihuahuan Desert, New Mexico. Ecosystems,6, 185–196. doi:10.1007/s10021-002-0210-9

Brocca, L., et al., 2007. Soil moisture spatial variability in experi-mental areas of central Italy. Journal of Hydrology, 333, 356–373. doi:10.1016/j.jhydrol.2006.09.004

Brocca, L., et al., 2010. Spatial-temporal variability of soil moistureand its estimation across scales. Water Resources Research, 46,W02516. doi:10.1029/2009WR008016

Brocca, L., et al., 2012. Catchment scale soil moisture spatial-tem-poral variability. Journal of Hydrology, 422–423, 63–75.doi:10.1016/j.jhydrol.2011.12.039

Charpentier, M.A. and Groffman, P.M., 1992. Soil moisture varia-bility within remote sensing pixels. Journal of GeophysicalResearch, 97, 18987–18995. doi:10.1029/92JD00882

David, M., 1977. Geostatistical ore reserve estimation. Amsterdam,The Netherlands: Elsevier.

Table 6 Stepwise linear regression of surface soil (0–6 cm) water content with selectedproperties under different wetness conditions.

Wetness Property Standardizedcoefficient

t P AdjustedR2

Wet BD −0.476 −7.949 0.000 0.664Relative elevation −0.327 −5.999 0.000Clay content 0.299 4.711 0.000Shrub coverage −0.180 −3.740 0.000Gravel content 0.181 3.436 0.001KS −0.103 −2.208 0.029

Medium BD −0.484 −7.450 0.000 0.605Clay content 0.324 4.709 0.000Shrub coverage −0.243 −4.656 0.000Relative elevation −0.201 −3.407 0.001Gravel content 0.182 3.196 0.002KS −0.108 −2.130 0.035

Dry BD −0.453 −6.889 0.000 0.595Clay content 0.348 5.037 0.000Shrub coverage −0.251 −4.743 0.000Relative elevation −0.222 −3.817 0.000Gravel content 0.159 2.769 0.006

BD: bulk density; KS: soil saturated hydraulic conductivity; t: Student’s t; P: significance level; R2: coefficient ofdetermination. The values in bold font represent the three most important factors influencing the pattern of SWC.

Spatio-temporal variability of surface soil water content 109

Dow

nloa

ded

by [

Nor

thw

est A

& F

Uni

vers

ity]

at 2

0:41

24

Dec

embe

r 20

14

Famiglietti, J.S., Rudnicki, J.W., and Rodell, M., 1998. Variability insurface moisture content along a hillslope transect: RattlesnakeHill, Texas. Journal of Hydrology, 210, 259–281. doi:10.1016/S0022-1694(98)00187-5

Gao, X.D., et al., 2011. Soil moisture variability along transects overa well-developed gully in the Loess Plateau, China. Journal ofHydrology, 87, 357–367.

Gao, L. and Shao, M., 2012. Temporal stability of shallow soilwater content for three adjacent transects on a hillslope.Agricultural Water Management, 110, 41–54. doi:10.1016/j.agwat.2012.03.012

Gómez-Plaza, A., et al., 2001. Factors regulating spatial distribution ofsoil water content in small semiarid catchments. Journal ofHydrology, 253, 211–226. doi:10.1016/S0022-1694(01)00483-8

Grayson, R.B., et al., 1997. Preferred states in spatial soil moisturepatterns: Local and nonlocal controls. Water ResourcesResearch, 33, 2897–2908. doi:10.1029/97WR02174

Hawley, M.E., Jackson, T.J., and McCuen, R.H., 1983. Surface soilmoisture variation on small agricultural watersheds. Journal ofHydrology, 62, 179–200. doi:10.1016/0022-1694(83)90102-6

Hébrard, O., et al., 2006. Spatio-temporal distribution of soil surfacemoisture in a heterogeneously farmed Mediterranean catch-ment. Journal of Hydrology, 329, 110–121. doi:10.1016/j.jhydrol.2006.02.012

Hills, R.C. and Reynolds, S.G., 1969. Illustrations of soil moisturevariability in selected areas and plots of different sizes. Journalof Hydrology, 8, 27–47. doi:10.1016/0022-1694(69)90029-8

Hu, W., et al., 2008. Soil water content temporal-spatial variability ofthe surface layer of a Loess Plateau hillside in China. ScientiaAgricola, 65, 277.

Hu, W., et al., 2010. Watershed scale temporal stability of soil watercontent. Geoderma, 158, 181–198. doi:10.1016/j.geoderma.2010.04.030

International GEWEX Project Office, 1995. Global Soil WetnessProject, 47 pp.

Jia, Y.H. and Shao, M.A., 2013. Temporal stability of soil waterstorage under four types of revegetation on the northern LoessPlateau of China. Agricultural Water Management, 117, 33–42.doi:10.1016/j.agwat.2012.10.013

Klute, A. and Dirksen, C., 1986. Hydraulic conductivity of saturatedsoils. In: A. Klute, ed. Methods of Soil Analysis. Madison, WI:ASA and SSSA, 694–700.

Lawrence, J.E. and Hornberger, G.M., 2007. Soil moisture variabilityacross climate zones. Geophysical Research Letters, 34,L20402. doi:10.1029/2007GL031382

Li, H.D., et al., 2013. Spatio-temporal variability of soil moisture andits effect on vegetation in a desertified aeolian riparian ecotoneon the Tibetan Plateau, China. Journal of Hydrology, 479, 215–225. doi:10.1016/j.jhydrol.2012.12.002

Lu, H. and Shi, J.C., 2012. Reconstruction and analysis of temporaland spatial variations in surface soil moisture in China usingremote sensing. Chinese Science Bulletin, 57, 2824–2834.doi:10.1007/s11434-012-5011-8

Meyles, E., et al., 2003. Runoff generation in relation to soil moist-ure patterns in a small Dartmoor catchment, SouthwestEngland. Hydrological Processes, 17, 251–264. doi:10.1002/hyp.1122

Nelson, D.W., and Sommers, L.E., 1982. Total carbon, organiccarbon and organic matter. In: A.L. Page, R.H. Miller, andD.R. Keeney, eds. Methods of soil analysis, Part 2, Chemicaland Microbiological Properties. Madison, WI: ASA andSSSA, 534–580.

Owe, M., Jones, E.B., and Schmugge, T.J., 1982. Soil moisturevariation patterns observed in Hand County, South Dakota.

Journal of the American Water Resources Association, 18,949–954. doi:10.1111/j.1752-1688.1982.tb00100.x

Pan, Y.X., et al., 2007. Variability characteristic of surface soilmoisture content in sand areas covered by different vegetationtypes. Journal of Soil and Water Conservation, 21, 106–109.(in Chinese with English abstract).

Pan, Y.X., et al., 2008. Spatial variability of surface soil moisturecontent in a re-vegetated desert area in Shapotou, NorthernChina. Journal of Arid Environments, 72, 1675–1683.doi:10.1016/j.jaridenv.2008.03.010

Pan, Y.X. and Wang, X.P., 2009. Factors controlling the spatialvariability of surface soil moisture within revegetated-stabi-lized desert ecosystems of the Tengger Desert, NorthernChina. Hydrological Processes, 23, 1591–1601. doi:10.1002/hyp.7287

Penna, D., et al., 2009. Hillslope scale soil moisture variability in asteep alpine terrain. Journal of Hydrology, 364, 311–327.doi:10.1016/j.jhydrol.2008.11.009

Qiu, Y., et al., 2001. Spatial variability of soil moisture content andits relation to environmental indices in a semi-arid gully catch-ment of the Loess Plateau, China. Journal of AridEnvironments, 49, 723–750. doi:10.1006/jare.2001.0828

Reynolds, S.G., 1970. The gravimetric method of soil moisturedetermination Part III An examination of factors influencingsoil moisture variability. Journal of Hydrology, 11, 288–300.doi:10.1016/0022-1694(70)90068-5

Takagi, K. and Lin, H.S., 2012. Changing controls of soil moisturespatial organization in the Shale Hills Catchment. Geoderma,173–174, 289–302. doi:10.1016/j.geoderma.2011.11.003

Vereecken, H., et al., 2007. Explaining soil moisture variability as afunction of mean soil moisture: A stochastic unsaturated flowperspective. Geophysical Research Letters, 34, L22402.doi:10.1029/2007GL031813

Wang, Y.Q., et al., 2012. Regional spatial pattern of deep soilwater content and its influencing factors. HydrologicalSciences Journal, 57 (2), 265–281. doi:10.1080/02626667.2011.644243

Western, A.W., et al., 1999. Observed spatial organization of soilmoisture and its relation to terrain indices. Water ResourcesResearch, 35, 797–810. doi:10.1029/1998WR900065

Western, A.W., Blöschl, G., and Grayson, R.B., 1998. Geostatisticalcharacterisation of soil moisture patterns in the Tarrawarracatchment. Journal of Hydrology, 205, 20–37. doi:10.1016/S0022-1694(97)00142-X

Williams, A.G., et al., 2003. Soil moisture variability and land use ina seasonally arid environment. Hydrological Processes, 17,225–235. doi:10.1002/hyp.1120

Yu, S.W., et al., 2012. Determination of water sources of Gobi plantsby D and 18O state isotopes in middle reaches of the HeiheRiver. Journal of Desert Research, 32, 717–723. (in Chinese).

Zhang, J.G., et al., 2011. Spatial variability and patterns of surfacesoil moisture in a field plot of karst area in southwest China.Plant, Soil and Environment, 57, 409–417.

Zhao, Y., et al., 2010. Controls of surface soil moisture spatialpatterns and their temporal stability in a semi-arid steppe.Hydrological Processes, 24, 2507–2519.

Zhao, Y., et al., 2011. Factors controlling the spatial patterns of soilmoisture in a grazed semi-arid steppe investigated by multi-variate geostatistics. Ecohydrology, 4, 36–48. doi:10.1002/eco.121

Zhu, Y.J. and Shao, M.A., 2008. Variability and pattern of surfacemoisture on a small-scale hillslope in Liudaogou catchment onthe northern Loess Plateau of China. Geoderma, 147, 185–191.doi:10.1016/j.geoderma.2008.08.012

110 Pingping Zhang and Ming’an Shao

Dow

nloa

ded

by [

Nor

thw

est A

& F

Uni

vers

ity]

at 2

0:41

24

Dec

embe

r 20

14