1 Introduction

Soil salinization is one of themost serious eco-environmental problems in irrigation areas in arid l and s(Hamid et al., 2011). Soil salinity is the major and most persistent threat toirrigated agriculture(Manoranjan et al., 2001). The global extent of primarysalt-affected soils is about 955×10⁶ hm²,while secondarysalinization affects some 77×10⁶ hm²,with 58% of thesebeing in irrigated areas. Globally,nearly 20% of all irrigated l and issalt-affected(Metternicht and Zinck, 2003). The total area of saline soil inXinjiang,China is about 8.48×10⁶ hm², and 31.1% of theplantation area is threatened by salinization. Soil salinity,together withother soil physical and chemical properties,plays an important role in plantcomposition,productivity, and distribution because of the differences intolerances of plant species to salinity. Excessive salt amounts adverselyaffect soil physical and chemical properties,as well as the microbiologicalprocesses(Abdelbasset et al., 2009). Effects of soil salinity are manifestedin loss of st and ,reduced plant growth,reduced yields, and in severe cases,crop failure. Remedial actions for soilsalinization require reliable information to better set priorities and to choosethe types of action that are most appropriate for combating soil salinization. Studieson the characteristics and spatio-temporal distribution of saline soil can providesuch necessary basic information for the documentation of salinity changes and theanticipation of further degradation(Dennis et al., 2007). Therefore,soil salinization in irrigation areas is a significant issue that istheoretically and practically essential for sustainable development ofagricultural l and in arid regions.

The spatial and temporaldynamics of soil salinization are important issues in soil salinization studies(Zhao et al., 2005; Mehmet and Turgut, 2006). With the increasingconcern of soil salinization in irrigated areas,the monitoring of thespatio-temporal dynamics of soil salinity in an aquifer can act as an early warningdevice in l and degradation. Irrigation of agricultural l and in arid regionsrequires that water be applied in excess of evapotranspiration to prevent saltaccumulation in the root zone(Prendergast et al., 2004). However,applications of excessive water,chemical fertilizers, and inefficient irrigationmethods threaten the sustainability of groundwater(Stigter et al., 2006). The expansion of irrigated agriculture and intensive agricultural activitiesinduce the risk of soil quality degradation.

There are many management toolsthat can help managers to make appropriate decisions given the available soilresources. Geostatistical methods have been proven to be an applicable and reliable tool for better management and conservation of soil,water resources, and sustainable development of any area(Kumar et al., 2005; Reghunath et al., 2005; Theodossiou and Latinopoulos, 2007). For example,Triantafilis et al.(2004)used the non-linear kriging method for mapping the salinity riskof farml and s using saline groundwater as the source of irrigation. Results oftheir study explained the potential of geostatistics to simulate the criticalconditions of salinization. In this study,we applied geostatistics for analyzingthe spatio-temporal variations of soil salinity in the Ili River IrrigationArea(IRIA). Such quantitative analyses are urgently needed,will be useful in l and use management of the IRIA, and can provide a scientific basic for soilsalinity control and sustainable agriculture in the study area.

The IRIA study area is situatedin the northern slope of the Tianshan Mountains and the southern part of the IliRiver,Xinjiang Uyghur Autonomous Region,China. It ranges from 80°54′E to 81°12′E and from 43°44′N to 43°56′N,with an altitude ranging from 560-750 m a.s.l.,witha total area of 424.6 km²(Figure 1). The study area has an aridcontinental climate with an average annual temperature of 8.9 °C,an average annual rainfall of 277 mm, and an average annual evaporation capacity of 1,594 mm. The accumulated active temperature≥10 °C is about 3,100 °C. The annual daily sun duration is 2,100 h. The coldestmonth(January)generally has a mean air temperature of −9.3 °C,while the warmest month(July)usuallyhas an average air temperature of 22.9 °C(Mamattursun et al., 2012).

Figure 1 Locationof the sampling sites

Figure options

Seriphidium transillense,Phragmites australia,Amaranthus retroflexus,Populus albavar. pyramidalis,Elaeagnus oxycarpa, and Hippophaerhamnoides are the main plant species in the study area. Cotton,wheat, and rice are the dominant crops in the region. The peak irrigation period forall crops is June-August,whereas irrigation of winter wheat starts inNovember. Agricultural developments depend on irrigation supplemented bygroundwater resources to certain extent. The irrigation of thesewater-intensive crops and seepage from the canal network are the most likelycauses of changes in groundwater salinity in the study area. A hydrological imbalance,underpinned by improper agricultural practices,has caused the water table torise in the topsoil. This rise of the water table has led to increased soil salinization and degradation of the relatively limited soil resources in the study area(Wang et al., 2008).

Figure 1 shows the geographical location of the soil sampling points. The coordinatessystem used in Figure 1 is the Universal Transverse Mercator(UTM). The datumof this system is the World Geodetic System of 1984(WGS 1984)upon whichGlobal Positioning System(GPS)measurements were made. A total of 544 soilsamples from 56 sampling sites were taken in March(136 samples),June(136 samples), and September(136 samples)of 2009. The soil samples were taken r and omly, and accordingto the spatial variation parameters of soil salinity,the total numbers of soilsamples were suitable for a spatial analysis.

Soil samples collected in thefield were analyzed for chemical constituents,such as electrical conductivity(EC),anions(HCO₃⁻,Cl⁻,CO₃²⁻, and SO₄²⁻), and cations(Na⁺,K⁺,Ca²⁺, and Mg²⁺). Soil salinity and EC were measured usingdigital meters immediately after sampling. The concentrations of Mg²⁺,CO₃²⁻, and HCO₃⁻ were determined byvolumetric titrations,AgNO₃ was used to estimate Cl⁻,a flamephotometer was used to measure Na⁺ and K⁺ ions, and EDTAtitrations were used to measure SO₄²⁻ and Ca²⁺.The accuracy of the chemical analysis was verified by calculating ion-balanceerrors,where the errors were generally around 10%.

Geostatistical methods can be used to ideally describe the spatial variability of the environment and reveal the spatial heterogeneity and spatial patterns of natural phenomena. Analyses using the semivariogram model and the kriging interpolation are the most common geostatistical analysis methods(Goovaerts,1997).

The main tool in geostatisticsis the semivariogram,which expresses the spatial dependence betweenneighboring observations(Yang et al., 2005). The semivariogram,γ(h),can be defined as one-half the variance of the difference between the attributevalues at all points separated by h,as follows:

where γ(h)is thesemi-variance,h is the sampling distance,Z(x)indicatesthe magnitude of the variable, and N(h)is the total number ofpairs of attributes that are separated by a distance h.

Prior tothe geostatistical estimation,we require a model that enables us to compute avariogram value for any possible sampling interval. A variogram model can beused to indicate both the structural and r and om aspects of a variable,such assoil salinity. The most commonly used variogram models are the spherical model,the exponential model, and the Gaussian model(Goovaerts,1997; Wang,1999;Hamid et al., 2011). The most appropriate semivariogram model wasselected by comparing these models by the following parameters: mean error,rootmean square,average st and ard error,root mean square st and ardized error,determinationcoefficient, and sum of the residual squares(Hu and Lu, 2009). ArcGIS 9.2 wasused for choosing the most appropriate semivariogram model.

Kriging interpolation is an exact interpolation estimator used to find the best linear unbiased estimate. It is based on certain mathematical models and statistical models, and it is used to derive the weight coefficients from the measured values ofthe nearby measurement points, and then to predict them(Pucci and Murashige, 1987; Hamid et al., 2011).Kriging weight coefficients are calculated by using the semivariogram figure reflecting the spatial structure of the data. They are determined not only by thesemivariogram figures and the distances to the prediction points,but also bythe spatial relations of the measured values of the nearby measurement points(Yang et al., 2008). The spatial variability of the related points in the study areacan be estimated by using the monitoring data of the sampling sites and thelocation relations between the sampling sites and the semivariogram model.

Descriptive statistics,including minimum values,maximum values,mean values,st and ard deviation(St.D), and coefficient of variation(CV)for soil salinity from 136 samplingpoints,are summarized in Table 1.

Table 1shows that the soil salinity varied widely,with a maximum value of 28.10 g/kg and a minimum value of 0.10 g/kg for the 0-3 cm soil layer,a maximum value of 8.70 g/kg and a minimum value of 1.37 g/kg for the 3-20 cm soil layer,a maximum value of 6.20 g/kg and a minimum value of 1.09 g/kg for the 20-40 cm soil layer, and a maximum value of 4.60 g/kg and a minimum value of 0.10 g/kg for the 40-60 cm soil layer. The mean values of soil salinity were 5.05,1.37,1.09, and 0.88 g/kg for the 0-3,3-20,20-40, and 40-60 cm soil layers,respectively. The overall mean top-soil layer salinity value of 5.05 g/kg falls into the category of moderately saline soil(Rhoades et al., 1992). The abundance of the major ions was in the following order: Cl⁻ > SO₄²⁻> Ca²⁺ > Na⁺ > K⁺ > Mg²⁺> CO₃²⁻ > HCO₃⁻ for the 0-3 cmsoil layer,SO₄²⁻ > Cl⁻ > Na⁺> Ca²⁺ > Mg²⁺ > K⁺ > CO₃²⁻= HCO₃⁻ for the 3-20 cm soil layer,SO₄²⁻> Cl⁻ > Ca²⁺ > Na⁺ > Mg²⁺> K⁺ > CO₃²⁻ = HCO₃⁻for the 20-40 cm soil layer, and SO₄²⁻ > Na⁺> Cl⁻ > Ca²⁺ > Mg²⁺ > K⁺ =CO₃²⁻ = HCO₃⁻ for the 40-60 cm soillayer. The CVs of soil salinity were fairly high. This could have been due touneven crop growth and non-uniform management practices,resulting in markedchanges in soil salinity over small distances. The CVs of anions and cations inthe soil also exhibited remarkable variability,ranging from a maximum value of650% to a minimum value of 50%. The salinity and main ions of the 0-3 cm soillayer were higher than the salinity and main ions of the 3-20 cm,20-40 cm, and 40-60 cm soil layers. The salinity of the surface soil layer(0-3 cm)accountedfor 60.19% of the salinity of 0-60 cm soil layer,indicating that the soilsalinity was distributed mainly in the top soil layer.

Previous studies showed that thesoil salinity of different l and use/cover types varies due to the differenteffects of groundwater withdrawal,irrigation, and transpiration(Mamattursun et al., 2010). Therefore,we analyzed the change of soil salinization indifferent l and use/cover types(Table 2).

Table 2shows that the soil salinity of grassl and varied from 0.10 g/kg to 8.70 g/kg,while the salinity of forestl and varied from 0.20 g/kg to 1.10 g/kg and the salinity of cropl and varied from 0.10 g/kg to 4.00 g/kg. The mean values of soil salinity were 2.50,0.60, and 1.10 g/kg for grassl and ,forestl and , and cropl and ,respectively. This indicates that the grassl and was slightly salinized. The abundance of the major ions of grassl and was inthe following order: Cl⁻ > SO₄²⁻ > Na⁺> Ca²⁺ > Mg²⁺ > K⁺ > CO₃²⁻= HCO₃⁻. The abundance of the major ions of forestl and was in the following order: SO₄²⁻ > Na⁺> Cl⁻ > Mg²⁺ > Ca²⁺ = K⁺ =CO₃²⁻ = HCO₃⁻. The abundance of themajor ions of cropl and was in the following order: SO₄²⁻> Na⁺ > Ca²⁺ > Cl⁻ > Mg²⁺> K⁺ > CO₃²⁻ = HCO₃⁻.The CVs of soil salinity for different l and use types were relatively high,which could have been due to non-uniform management practices,resulting inmarked changes in soil salinity over small distances.

The changes in soil salinity in March,June, and September are shown in Table 3. The soil salinity in March varied from0.10 g/kg to 10.30 g/kg,while it varied from 0.10 g/kg to 8.70 g/kg in June and from 0.06 g/kg to 6.17 g/kg in September. The mean values of soil salinity were 1.61,1.37, and 1.17 g/kg for March,June, and September,respectively. The CVs of soil salinity in different periods were also relatively high,possibly due to irrational agriculturalactivities,resulting in marked changes in soil salinity over small distances.

The results of spatialstructure analysis of soil salinity are summarized in Table 4. By fitting avariogram model to the data,it was found that the spherical model was the mostsuitable variogram model to describe the salinity of the 0-3 cm and 3-20 cmsoil layers in March and June, and the 3-20 cm soil layer in September; the exponentialmodel was the most suitable variogram model to describe the salinity of the 0-3cm soil layer in September.

The spatial heterogeneity of avariable in a variogram model can be represented by the values of nugget,sill, and range. Nugget and sill characterize the r and om aspect of the variable,whereasrange characterizes the structural aspect. The ratio of nugget variance to sillvariance could be regarded as a criterion to classify the spatial dependence ofgroundwater levels and salinity. If the ratio is less than 25%,the variable hasstrong spatial dependence; between 25% and 75%,the variable has moderatespatial dependence; and greater than 75%,the variable shows only weak spatial dependence(Li,1998).

In our study,the nugget effectof less than 1.1 for soil salinity indicated the existence of a strong spatialauto-correlation for these elements(Yang et al.,2008). The low nuggeteffect reflects the fact that the variation of soil salinity was highly spatiallystructured, and also that there was little or no variability of soil salinityin shorter distances of the range values. This led us to the conclusion thatthe fitted semivariogram model well represented the spatial structure of variationof soil salinity. In our study the nugget-to-sill ratios of March,June, and September were less than 75%,indicating that the soil salinity in those monthshad a strong spatial dependence. The nugget-to-sill ratios of soil salinity alsorevealed that the spatial heterogeneity of soil salinity was caused mainly bythe spatial structure, and the soil salinity was jointly affected by the spatialstructure and the stochastic factors. The spatial correlation of the soilsalinity might be caused by the spatial structure,such as natural factorsincluding the terrain,l and forms,climate, and soil types. The ranges of soilsalinity of March,June, and September simulated in the semivariogram model variedfrom 0.93 km to 2.98 km,indicating that our sampling density was suitable forthe study area. However,the ranges were similar among the measurement periodsdespite the differences in irrigation intensity between the measurement periods.

Ordinary kriging was applied for estimation of soil salinity in March,June, and September across the study area(Figure 2). This figure shows that higher soil salinitywas distributed in the central parts and lower soil salinity was distributed in the marginal parts of the study area. Soil salinity tended to increase from the marginal parts to the central parts across the study area. The soil salinity in the central parts of the study area exceeded 28 g/kg and the maximum value reached 28.10 g/kg. The soil salinity in the marginal parts was low; its minimum value was about 0.10 g/kg.

Figure 2 Spatial distribution of soil salinity obtained by the kriging method

Figure options

This research showed that geostatisticalanalysis methods are very useful for estimating changes in the spatio-temporal dynamics of soil salinization at the local level.In many cases,geostatistical analysis methods may be the most economically feasibleway to gather regular soil salinization information over large areas. Resultsof our study indicated that inthese spatial distribution maps of soil salinity the similarity of the spatial patternsof the soil salinity in each measurement period was quite striking: soilsalinity was higher in the central and in the marginal parts of the region evenin March,when no or rare irrigations occurred. Interpolation and comparison ofthe soil salinity maps in each measurement period showed that the spatialdistribution of soil salinity was very similar to those of the averaged March,June, and September measurements.

Agricultural wells are denserin central parts of the study area than in the northern parts, and theydirectly exert much pressure on groundwater levels and salinity. Furthermore,thereare more agricultural canals in the central parts of the area than in the southernparts,which also directly exert much pressure on the groundwater level. Therefore,the groundwater level in the central parts of the region is not as deep aselsewhere. Because of drought,soil water evaporates,leaving salt in the soil and groundwater. Also,higher-EC groundwater used forirrigation contributes salt to the soil,causing soil salinization. The groundwater table in the central partsof the study area has now almost reached the ground surface in some areas,causing an advanced stage of soil salinization. This indicates that more attention should be paid tothese areas to prevent future problems.

Reasonable irrigation rates - well-adjusted to the water dem and s of crops- and suitable changes in l and use shouldnot result in an unbalance of the regional water quantity. However,the waterresources in the study area have been used for agriculture for a long time,mostly for irrigation by traditional flooding(although some cropl and s have adoptednew irrigation techniques). As a result,water has been lost in the transportationprocess and excess water has been infiltrated,leading to a rise of the groundwaterlevel to near the surface,becoming the origin of soil salinization. Bio-drainagecan be enhanced by cultivating salt-tolerant trees and shrubs that have a highrate of evapotranspiration. Vegetation such as Populus euphratica,Halocnemumstrobilaceum,Halogeton arachnoideus,Halostachys caspica,Haloxylonammodendron,T. taklamakanensis, and Tamarix arceuthoidesare suitable to be grown and have great benefits to control soil salinization inthe area(Mamattursun et al., 2010). Biological drainage has proven tobe very effective in lowering shallow groundwater tables and facilitating someleaching of salts from the surface layers of salinized soils. Enhancing the integrated use of surfacewater and groundwater will optimize the use of water resources.

With an extensive field surveyof soil salinity in 2009, and with many interviews of local experts,we haveshown that the use of geostatistics can be an effective means of acquiringinformation on soil salinization changes. In an arid zone oasis such as the IRIA,extensive field-based survey methods can bedifficult and expensive to implement due to restricted accessibility. However,in such areas a limited amount of field sampling combined with geostatisticalanalysis methods can produce reasonably accurate large-scale information atrelatively little cost.

In thisresearch,we examined the characteristics and spatio-temporal dynamics of soilsalinity in the IRIA,where the entire irrigated area is suffering from variousdegrees of soil salinization. The observed spatial distribution of soilsalinity in the IRIA revealed that central parts of the area were at risk of gravel and degradation due to higher groundwater salinity and deeper groundwaterlevels. The higher soil salinity in the study area was also associated withimproper human activities,specifically mismanaged agricultural practices.

It is concluded that application of geostatistics can produce better insight into soilsalinization and can lead to valuable solutions for those critical conditionswhich endanger soil resources. Applying the kriging method is also helpful indetecting the problematic areas and is a good tool for soil resources management.Hence,management efforts in the appropriate use of soil and water resources insalinized areas are very important for sustainable agriculture. The futuresustainable agriculture of salinized areas is highly dependent on the presentmanagement of the soil and water resources.

The authors are grateful to the anonymousreviewers for their critical reviews and comments on drafts of this manuscript.This research was funded by the National Natural Science Foundation of China(Nos.41201032,41561073, and U1138302).