Sciences in Cold and Arid Regions  2016, 8 (6): 507-515   PDF    

Article Information

HanChen Duan, Tao Wang, Xian Xue, CuiHua Huang, ChangZhen Yan . 2016.
Quantitative retrieval of soil salt content based on measured spectral data
Sciences in Cold and Arid Regions, 8(6): 507-515
http://dx.doi.org/10.3724/SP.J.1226.2016.00507

Article History

Received: April 28, 2016
Accepted: July 29, 2016
Quantitative retrieval of soil salt content based on measured spectral data
HanChen Duan, Tao Wang, Xian Xue, CuiHua Huang, ChangZhen Yan     
Key Laboratory of Desert and Desertification, Northwest Institute of Eco-Environment and Resources, Chinese Academy of Sciences, Lanzhou, Gansu 730000, China
Abstract: Choosing the Minqin Oasis, located downstream of the Shiyang River in Northwest China, as the study area, we used field-measured hyperspectral data and laboratory-measured soil salt content data to analyze the characteristics of saline soil spectral reflectance and its transformation in the area, and elucidated the relations between the soil spectral reflectance, reflectance transformation, and soil salt content. In addition, we screened sensitive wavebands. Then, a multiple linear regression model was established to predict the soil salt content based on the measured spectral data, and the accuracy of the model was verified using field-measured salinity data. The results showed that the overall shapes of the spectral curves of soils with different degrees of salinity were consistent, and the reflectance in visible and near-infrared bands for salinized soil was higher than that for non-salinized soil. After differential transformation, the correlation coefficient between the spectral reflectance and soil salt content was obviously improved. The first-order differential transformation model based on the logarithm of the reciprocal of saline soil spectral reflectance produced the highest accuracy and stability in the bands at 462 and 636 nm; the determination coefficient was 0.603, and the root mean square error was 5.407. Thus, the proposed model provides a good reference for the quantitative extraction and monitoring of regional soil salinization.
Key words: spectral reflectance     soil salt content     salinization     multiple linear regression     Minqin Oasis    
1 Introduction

Soil salinization,which is one of the main types of land desertification and degradation,has led to global resource,environmental,and ecological problems(Li and Wu,2002; Hussain et al.,2006). Soil salinization can weaken and even destroy land productivity,reduce crop yields,and threaten the ecological environment and biosphere. It has become a major barrier to agricultural development in arid and semi-arid regions and threatens the security and stability of oases(Dehaan and Taylor,2002; Metternicht and Zinck,2003). Therefore,understanding the nature,scope,and extent of saline soils can assist in the effective governance of soil salinization,facilitating local agricultural production and sustainable development of regional economies.

At present,the application of remote sensing technology in mapping and monitoring soil salinization has shown great accuracy and efficiency,and multispectral remote sensing technology is widely used to monitor soil salinization(Khan and Sato,2001; Khan et al.,2005). However,because of a relatively low spectral resolution,it is difficult to distin- guish the mixed spectral characteristics of complex surfaces,leading to loss of some spectral details and making it impossible to accurately estimate the soil salt content(Pu et al.,2012). In contrast to multispectral technology,hyperspectral technology can generate continuous spectral information of ground surface features on the nanoscale and distinguish subtle spectral differences among objects. Thus,hyperspectral technology makes possible the identification of surface features based on diagnostic spectral absorption characteristics,remote sensing quantitative analysis,and chemical composition research(Zhang et al.,2012b; Xu et al.,2014). Given the advantages of hyperspectral technology,in recent years it has been widely used for the quantitative extraction of soil salinization characteristics. Based on measured spectral data and EO-1 hyperspectral images,Weng et al.(2010)accurately estimated the soil salt content using the salinization index. By analyzing the spectral characteristics of several typical objects in the delta oasis of the Weigan and Kuqa rivers of the Tarim Basin,Zhang et al.(2010)established a ground spectral database. Ding et al.(2012)used saline soil,vegetation,and saline soil spectral reflectance to establish a hyperspectral model for monitoring salinization based on a measured comprehensive spectral index. Zhao et al.(2014)analyzed the spectral features of saline soil by combining spectral reflectance data with hyperspectral satellite images and extracted salinization information using the spectral angle classification method.

In this study,using the Minqin Oasis in Northwest China as the study area,the spectral reflectance of soils with different degrees of salinity was obtained by field measurements,and the relation between the reflectance and soil salt content was analyzed. Finally,sensitive wavebands were screened. Using multiple linear regression analysis,we established a model to quantitatively estimate soil salinity based on measured data. This model facilitates reliable and efficient mon-itoring and evaluation of soil salinization on a large scale.

2 Study area

The Minqin Oasis is located in the lower reaches of the Shiyang River in Gansu Province at latitudes of 38°05′N–39°26′N and longitudes of 101°59'E– 104°12'E(Figure 1). The Minqin Oasis is bordered on the east by the Tengger Desert and on the northwest by the Badain Jaran Desert,which is surrounded by the Gobi Desert on three sides. The total area of Minqin County is 1.6×104 km2,wherein the deserts,eroded mountains,and salinas account for 91%,and the oasis accounts for only 9%(Gao et al.,2006; Pang et al.,2014). The study area is arid and has a typical temperate continental climate; its annual average temperature is 7.6 °C,annual average rainfall is 110 mm,and annual average evaporation is 2,644 mm(about 24 times the amount of precipitation)(Chen et al.,2014).

Figure 1 Study area and distribution of sampling sites

The soil types in the study area include sandy,gray-brown desert,meadow,and meadow-boggy soils. The natural vegetation,which is divided into desert and meadow types,is of the typical extremely arid desert steppe type(Yue et al.,2011; Zhang et al.,2014). Minqin is a typical oasis agricultural county; however,the influence of drought,shortage of water resources,and strong evaporation result in salt accumulation in low-lying land. This soil salinization seriously restricts the sustainable development of oasis agriculture and the social economy.

3 Methodology 3.1 Collection of soil samples and measurement of salt content

In this study,field sampling was conducted on October 2–10,2014(after the autumn harvest and before autumn plowing and winter irrigation). We collected 42 soil sampling units from the surface(0~10 cm)of typical soil salinization regions. The dimensions of a sampling unit were 100m×100m,and three samples(forming a triangular shape)were obtained from each unit,resulting in a total of 126 soil samples. The coordinates of the center position of each sampling unit were simultaneously recorded via GPS. After natural drying and grinding,the soil samples were filtered using a 1-mm aperture sieve and mixed with deionized water(5:1 by mass)to extract the salt content. The mean salt content of the three soil samples in each sampling unit was considered as the measured value of a sampling unit. Errors and abnormal values caused by human operation and instruments were eliminated. In total,we obtained 39 data points from the sampling units(117 soil samples),29 of which were used to construct the model,and the remaining 10 were used to test the model's accuracy.

According to the classification standard of soil salinization of Minqin County(Chen et al.,2013),and the distribution of the field survey sampling points,we classified the soil salinization degree by choosing the soil salt content as the main factor(Table 1). It can be seen from Table 1 that the degree of soil salinization was divided into five types in the study area: non-salinization,slight salinization,moderate salinization,severe salinization,and saline. The sampling points were mainly distributed in abandoned land,unused land within the oasis,and the border region between the oasis and desert,all of which have a certain representativeness.

Table 1 Classification of soil salinization
Salinization intensityTotal salt content (g/kg)Electric conductivity EC1:5 (ds/m)
Non-salinization0.0~1.00.0~0.35
Slight salinization1.0~3.00.35~0.91
Moderate salinization3.0~5.00.91~1.48
Severe salinization5.0~10.01.48~2.90
Saline>10.0>2.90
3.2 Field spectral data acquisition

The spectral data were acquired using an ASD Field Spec. Hand Held 2 spectrometer(Analytical Spectral Devices,Boulder,CO,USA)with a spectral range of 325~1,075 nm,spectral resolution of <3 nm,sampling interval of 1.4 nm,and wavelength accuracy of ±1 nm.

The field spectral measurements and soil sampling were simultaneously performed on the same ground. Field spectra were obtained on clear,cloudless days with low wind speeds and sufficient,stable sunlight intensity. To reduce the influence of changes in the solar altitude angle on the spectral measurement results,the measurement period was 10:00–14:00 local time,because during this period the field of view was directly illuminated by the sun and the atmospheric transmittance was the highest. The view angle of the probe(25°)was perpendicular to the surface(20 cm)during the measurements. Before measurement,the influence of dark current was removed,and the instrument was optimized based on a white reference panel. The spectral data were collected 10 different times at each measurement point. After removing any abnormal curves,the arithmetic average value was determined as the spectral reflectance value of a sampling point. The average value of the spectral reflectance of the three sampling points in each unit was determined as the spectral reflectance of a sasampling unit to reduce the difference among sampling points in the same area.

3.3 Spectral data processing

Because of the difference in energy responses among spectrometer bands,a spectral curve has "burr" noise; thus,optimizing the spectral curve is necessary. In this study,the nine-point weighted moving average method was used to smooth spectral reflectance data(Peng et al.,2013).

Previous studies(e.g.,He et al.,2006; Ma,2014)demonstrated that the appropriate mathematical transformation of a spectral reflectance can effectively limit the influence of low-frequency noise on a target spectrum,eliminate the correlation among bands,and enhance the spectral difference among features. It is easy to identify a high-correlation band with soil salinity,which can improve the inversion accuracy. In addition to the analysis of the original spectral reflectance(R),various mathematical transformations have been employed: root mean square($\sqrt{R}$),logarithm(lgR),reciprocal(1/R),logarithmic reciprocal(1/lgR),reciprocal logarithm(lg(1/R)),first-order differential(R′),first-order differential of root mean square((R)′),first-order differential of logarithm((lgR)′),first-order differential of reciprocal((1/R)′),first-order differential of logarithmic reciprocal((1/lgR)′),and first-order differential of reciprocal logarithm((lg(1/R))′).

Correlation analysis helps identify bands that are sensitive to soil properties and establish a regression model for soil properties(Lei et al.,2014). Based on the measured spectral reflectance and its transformation forms as feature vectors,the correlation coefficient ri between each band's feature vector and soil salt content was obtained,and the characteristic bands of the salt content in the soil were screened. A prediction model was then established using a curve regression method based on the following formula:

${{r}_{i}}=\frac{\sum\limits_{i=1}^{n}{\left( {{R}_{ij}}-\overline{{{R}_{j}}} \right)\left( S{{C}_{i}}-\overline{SC} \right)}}{\sqrt{\sum\limits_{i=1}^{n}{{{\left( {{R}_{ij}}-\overline{{{R}_{j}}} \right)}^{2}}\sum\limits_{i=1}^{n}{{{\left( S{{C}_{i}}-\overline{SC} \right)}^{2}}}}}}$ (1)

where ri is the correlation coefficient between spectral reflectance,its transformation forms,and soil salt content; n is the total number of sampling units; j is the band number; Rij is the reflectance value of a spectrum and its transformation forms in the j band of the ith sampling unit; $\overline{{{R}_{j}}}$ is the average value of the spectral reflectance and transformation forms of n units in the j band; SCi is the soil salt content in the ith unit; and SC is the average value of the soil salt content in n units.

3.4 Model construction and testing

In this study,we randomly selected 29 soil samples from a total of 39 samples to establish a curve regression model; the other 10 samples were used for model validation. The total root mean square error(RMSE)was used to evaluate the accuracy of the estimation model; a relatively smaller RMSE value indicates a relatively higher model precision and more reliable simulation results. RMSE is defined as follows:

$RMSE=\sqrt{\sum\limits_{i=1}^{n}{{{\left( y_{i}^{*}-{{y}_{i}} \right)}^{2}}}}$ (2)

where n is the total number of sampling units,$y_{i}^{*}$ is the soil salt content predicted by the model,and yi is the measured soil salt content.

4 Results 4.1 Analysis of spectral characteristics of saline soil

Figure 2 shows the reflectance curves of soils with different degrees of salinity. Although some differences were observed,the overall shapes of the reflectance curves of these soils were consistent. The spectral reflectance increased with wavelength; the reflectance increased sharply in the wavelength range of 400~600 nm and gently in the range of 600~900 nm. With the exception of a weak absorption band in the vicinity of 751 nm,no obvious peak and trough were observed in the spectral curves. In the wavelength range of 350~1,000 nm,the spectral reflectance did not increase with the soil salt content. In contrast,as the degree of soil salinization increased,the reflectiv-ity gradually decreased,and the spectral reflectance of the slightly salinized soil was the highest. This result was attributed to the serious effect of soil moisture on the spectral curves. In the study area,the soil water content is high,and salt accumulates because of strong evaporation,resulting in soil salinization.

Figure 2 Spectral curves of soil reflectance with different degrees of salinity

The correlation analysis in this study indicated a significantly positive correlation between soil salinity and soil water content(Figure 3); the correlation coefficient was 0.852,which was significant at the 0.01 level. These results indicate that the higher the soil moisture,the greater the soil salinity,and vice versa. Previous studies showed that the soil in the study area contains a large amount of high-moisture-absorbing salt(MgCl2)(Qi et al.,2010; Zhang and Qi,2013). Thus,under natural conditions,the water in the surrounding environment can be easily absorbed,increasing soil moisture and resulting in a negative correlation between the spectral reflectance and soil salt content. In general,the spectral reflectance of a salinized soil in the visible and near-infrared bands is higher than that of a non-salinized soil.

Figure 3 Relation between soil salinity and soil water
4.2 Screening of sensitive bands of salt content in soil

There are significant differences between the spectral reflectance and soil salinity,and the sensitive bands of soil salinity can be screened from measured soil spectral curves. Thus,it is necessary to quantitatively analyze the relation between the soil spectral reflectance and soil salinity. To identify the sensitive bands of soil salinity,we calculated the correlation coefficient between the soil spectral reflectance,its transformation,and soil salt content(Figure 4). Since the reflectance signal had more noise at wavelengths below 400 nm and above 900 nm,we selected a wavelength range of 400~900 nm to analyze the correlation between the soil spectral reflectance and soil salinity. As shown in Figure 4,the correlation coefficient changed with wavelength,and both positive and negative correlations were observed between spectral reflectance,its transformation,and soil salt content. Soil salt content with spectral reflectance(R),root mean square($\sqrt{R}$),and logarithm(lgR)was negatively correlated,whereas that with reciprocal(1/R),logarithmic reciprocal(1/lgR),and reciprocal logarithm(lg(1/R))was positively correlated. Soil salt content with first-order differential(R′),first-order differential of root mean square((R)′),first-order differential of logarithm((lgR)′),first-order differential of reciprocal((1/R)′),first-order differential of logarithmic reciprocal((1/lgR)′),and first-order differential of reciprocal logarithm((lg(1/R))′)were both positively and negatively correlated. After differential transformation,the correlation coefficient showed a fluctuating trend around the 0 axis,and the fluctuation range of the correlation coefficient was significantly greater than that before differential transformation.

Figure 4 Correlation coefficients of spectral reflectance and its different transforms with soil salinity

From the correlation coefficient peak values and their corresponding bands between the spectral reflectance,reflectance transformation,and the soil salt content,we can see that(Table 2)the correlation coefficients of the peak band between the spectral reflectance(R),root mean square($\sqrt{R}$),logarithm(lgR),first-order differential of reciprocal((1/R)′),first-order differential of reciprocal logarithm((lg(1/R))′),and the soil salt content was negative; the correlation coefficients of the first three variables was significantly correlated at the 0.05 level,and that of the last two variables were significantly correlated at the 0.01 level.

Table 2 Peak values and the corresponding bands of correlation coefficients between spectral different transformations and soil salinity
TransformationWavelength
(nm)
Correlation
coefficient (r)
R898−0.392*
897−0.388*
$\sqrt{R}$898−0.385*
897−0.383*
lgR898−0.378*
897−0.374*
1/R8980.359*
8970.359*
1/lgR8980.402**
8970.397**
lg(1/R)8980.378*
8970.374*
R464−0.524**
6360.481**
($\sqrt{R}$)′460−0.436**
5750.444**
(lgR)′4620.453**
6360.516**
(1/R)′603−0.502**
636−0.559**
(1/lgR)′4600.440**
5540.466**
(lg(1/R))′462−0.453**
636−0.550**
Note: * Significantly correlated at the 0.05 level; ** significantly correlated at the 0.01 level.

However,not all transformation forms were able to improve the sensitivity of the spectral reflectance to the soil salt content. On the contrary,some transformations reduced the correlation between the reflectance and the soil salt content,such as the root mean square($\sqrt{R}$),logarithm(lgR),reciprocal(1/R),and reciprocal logarithm(lg(1/R)); the correlation coefficient between them and the soil salt content was smaller than that between the original spectral reflectance and the soil salt content. The correlation coefficients of the peak band between the reciprocal(1/R),logarithmic reciprocal(1/lgR),reciprocal logarithm(lg(1/R)),first-order differential of logarithm((lgR)′),first-order differential of logarithmic reciprocal((1/lgR)′)and soil salt content were positive; the correlation coefficients of the peak band of reciprocal(1/R)and reciprocal logarithm(lg(1/R))show significantly correlated at the 0.05 level,whereas the rest were significantly correlated at the 0.01 level. Among them,the soil salt content and the first-order differential of logarithm((lgR)′)had the most close correlations; the corresponding correlation coefficients were 0.453 and 0.516,respectively. The correlation coefficients of the peak band between the first-order differential(R′),first-order differential of root mean square(($\sqrt{R}$)′)and the soil salt content were both positive and negative,but all were significantly correlated at the 0.01 level. These results indicate a linear relation between soil salt content and the spectral indices of the corresponding points.

After the differential transformation of the spectral reflectance,the subtle information contained in the original spectral data was amplified,significantly enhancing the correlation coefficient between the salt content and spectral reflectance. On this basis,we determined that the corresponding bands of the correlation coefficient's peak value of the spectral different transformations were the sensitive bands to the soil salt content(Table 2). Based on the correlation analysis between the spectral reflectance and the soil salt content,the sensitive bands of the soil salt content were located between the visible and near-infrared wave bands. This conclusion is consistent with the results of previous research(Gu et al.,2011; Zhang et al.,2011).

4.3 Construction of the model to predict soil salt content and verification of its accuracy

According to the correlation coefficients between spectral reflectance,its different transformations,and soil salinity,the corresponding bands of the correlation coefficient peak values were selected. We applied the multiple linear regression method to construct an inversion model of soil salinity based on different independent variables(Table 3). The accuracy of the model was verified using the field-measured soil salinity data,and the optimal inversion model of soil salinity was determined based on the measured spectral data.

Table 3 Comparison of multiple linear regression models with different dependent variables
TransformationsRegression modelsR2RMSE
RY=58.920+547.859X897-681.785X8980.4026.357
$\sqrt{R}$Y=108.861+477.875X897-641.779X8980.4006.342
lgRY=−39.007+529.099X897-640.679X8980.4206.320
1/RY=−37.872-63.069X897+80.335X8980.4136.294
1/lgRY=58.703-92.389X897+113.571X8980.3956.406
lg(1/R)Y=−39.007-529.099X897+640.679X8980.4206.317
RY=8.287+3909.788X464-8014.214X6360.5335.828
($\sqrt{R}$)′Y=1.174-13988.856X460+14019.538X5750.4136.281
(lgR)′Y=0.321+5190.923X462+8405.077X6360.5785.605
(1/R)′Y=3.197-652.603X603-1399.609X6360.5385.635
(1/lgR)′Y=30.896+3591.978X460+2551.676X5540.5435.549
(lg(1/R))′Y=0.793-4892.492X462-6747.222X6360.6035.407
Note: R2 is the determination coefficient; RMSE is the root mean square error.

By comparison and analysis the soil salt content inversion models(Table 3),the sensitive bands of the soil salt content of R,$\sqrt{R}$,lgR,1/R,1/lgR,and lg(1/R)were mostly located around 897 and 898 nm,and other forms after differential transform were mostly located around 462 and 636 nm. The first-order differential transformation model based on the logarithm of the reciprocal of saline soil spectral reflectance had the highest determination coefficient(R2 = 0.603),the lowest RMSE(5.407),and the most reliable stability. Followed by the first-order differential of the logarithm model,its determination coefficient R2 and RMSE were 0.578 and 5.605,respectively. The logarithmic reciprocal model had the lowest determination coefficient(R2 = 0.395),the highest RMSE(6.406),and the worst reliability. Therefore,the first-order differential transformation model based on the logarithm of the reciprocal of saline soil spectral reflectance((lg(1/R))′)was used to retrieve and extract the soil salinity in the study area.

We considered the 10 soil sample units that were not involved in constructing the model as true values in order to verify the stability and prediction ability of the model. The determination coefficient(R2)was used to test the stability of the model,and the total RMSE was used to test the model's predictive capability; the predicted values of soil salinity were obtained from the first-order differential transformation model based on the logarithm of the reciprocal of saline soil spectral reflectance,and then the regressive relation was established between the predicted and measured values. The scatter diagram of the predicted and measured values is shown in Figure 5. It can be seen that the determination coefficient(R2)of the test samples was 0.806,and the total RMSE was 3.662,satisfying the accuracy requirement of this study. Therefore,compared with the other models in Table 3,the first-order differential transformation model based on the logarithm of the reciprocal of saline soil spectral reflectance had a better accuracy and higher stability; thus,it can better achieve the quantitative extraction and retrieval of the soil salt content in the study area.

Figure 5 Scatterplot of the measured and estimated values of the curve regression model
5 Discussion

Previous research has shown that the correlation between soil salt content and soil spectral reflectance under artificial light is significantly higher than that under natural light. The spectral reflectance increases with soil salinity(Rao et al.,1995; Farifteh et al.,2008; Zhao et al.,2012). The quantitative inversion model established in this paper is mainly based on the field-measured hyperspectral data,the purpose of which is to provide a scientific basis for quantitative monitoring of soil salinization by using hyperspectral remote sensing. However,we found that,due to the influence of the sampling number,local climate,soil type,soil moisture content,soil texture,and many other factors,there are some differences between the soil spectral curves(Elvidge et al.,1990; Zhang et al.,2012a). Although the accuracy of the soil salt content estimation model based on the reflectance of artificial light is higher than that based on natural light(Zhang et al.,2012c),it is not the actual reflection of the land surface. It is difficult to establish the corresponding relationship between the estimation model and the remote sensing images. So,for monitoring the soil salinization quantitatively,spectrum analysis for modeling under natural light is more applicable to connect the satellite images,which helps achieve rapid and accurate monitoring on the regional scale. Therefore,compared with the reflectance of artificial light in the laboratory,the soil salt content estimation model based on the spectral reflectance of natural light is more practical.

During our research,we found that there were some wavelengths that had significant correlations with the soil salt content. However,due to the existence of collinearity effects between bands,the high-autocorrelation bands were removed to avoid the impact of collinearity effects on the results of the study. Only two sensitive bands that had higher correlation and lower autocorrelation with the soil salt content were selected to establish the soil salinity estimation model. Moreover,the linear regression analysis method used in this study has some limitations: it requires the independent and dependent variables to have a good linear relation. If there is no linear relation between them,this method will not be able to accurately estimate the salinity of the soil. Therefore,we need to develop a method based on a combination of linear and nonlinear relations to model soil salinity and to make the quantitative monitoring methods more accurate. Also,the number of samples we used for model establishment and verification was too limited,which may have induced adverse influences to the results. The relatively small sampling number may have lessened the simulation accuracy and created an error in the result,thereby restricting the model extent to only a large regional scale. In future research we will increase the number of soil samples as much as possible,using different soil types and textures,so as to improve the accuracy and applicability of the model.

6 Conclusions

In this paper,based on field-measured soil spectral reflectance and their 11 transformation forms,the coupling relations between the soil salt content and the spectral reflectance,as well as its transformation forms,were analyzed. Based on the results,we constructed a quantitative model to predict the soil salt content and verified the model using measured soil salt content data. The main conclusions are as follows:

1)The spectral curves of soils with different degrees of salinity were consistent as a whole,and the spectral reflectance increased with wavelength. Spectral reflectance increased sharply in the wavelength range of 400~600 nm and gently in the range between 600 and 900 nm. Due to the effects of soil moisture,the soil spectral reflectance decreased with increasing degree of soil salinization; thus,spectral reflectance and soil salt content were significantly negatively correlated. By analyzing the spectral characteristics of the saline soil,we found that the spectral reflectance in the visible and near-infrared regions was higher for salinized soil than that for non-salinized soil.

2)Both positive and negative correlations were found between the soil spectral reflectance,its transformation forms,and the soil salt content. After the differential transformation of spectral reflectance,the correlation coefficient between soil salt content and spectral reflectance was significantly enhanced. The sensitive bands of the soil salt content of the different spectral transformation forms were determined based on the corresponding bands of the correlation coefficient peak values,which were all located between the visible and near-infrared wavelengths.

3)By comparing and analyzing the inversion models of the soil salt content based on different independent variables,we found that the optimal sensitive bands of the first-order differential transformation model based on the logarithm of the reciprocal of saline soil spectral reflectance were the bands at 462 and 636 nm; the correlation coefficient was maximized for these bands,and the RMSE was minimized. The inversion model was the optimum model for the estimation of soil salt content. We used the measured soil salinity data to test the stability and predictive ability of the model; the results showed that the determination coefficient(R2)of the test samples was 0.806,and the RMSE was 3.662. Thus,compared with other regression models,the first-order differential transformation model based on the logarithm of the reciprocal of saline soil spectral reflectance had a better accuracy and higher stability,and this model can better predict the soil salt content in the study area.

Acknowledgments:

This study was financially supported by the National Natural Science Foundation of China(No. 41401109),the Foundation for Excellent Youth Scholars of CAREERI,CAS(No. Y551D21001),and the Open Fund Project of the Key Laboratory of Desert and Desertification,CAS(No. Y452J71001). The authors would like to thank Enago(www.enago.cn)for the English language review.

Reference
Chen LJ, Feng Q, Cheng AF, 2013. Spatial distribution of soil water and salt contents and reasons of saline soils' development in the Minqin Oasis. Journal of Arid Land Resources and Environment, 27(11): 99–105. doi: 10.3969/j.issn.1003-7578.2013.11.017.(inChinese)
Chen XS, Gao BB, Wang XJ, et al, 2014. Changes and status of desertification land in Minqin, Gansu, China. Journal of Desert Research, 34(4): 970–974. doi: 10.7522/j.issn.1000-694X.2014.00077.(inChinese)
Dehaan RL, Taylor GR, 2002. Field-derived spectra of salinized soils and vegetation as indicators of irrigation induced soil salinization. Remote Sensing of Environment, 80(3): 406–417. doi: 10.1016/S0034-4257(01)00321-2
Ding JL, Wu MC, Liu HX, et al, 2012. Study on the soil salinization monitoring based on synthetical hyperspectral index. Spectroscopy and Spectral Analysis, 32(7): 1918–1922. doi: 10.3964/j.issn.1000-0593(2012)07-1918-05.(inChinese)
Elvidge CD, 1990. Visible and near-infrared reflectance characteristics of dry plant materials. International Journal of Remote Sensing, 11(10): 1775–1795. doi: 10.1080/01431169008955129
Farifteh J, Meer F, Meijde M, et al, 2008. Spectral characteristics of salt-affected soils: A laboratory experiment. Geoderma, 145(3-4): 196–206. doi: 10.1016/j.geoderma.2008.03.011
Gao ZH, Li ZY, Wei HD, et al, 2006. Quantitative monitoring of vegetation cover change by using remotely sensed data over Minqin Oasis, Gansu. Geographical Research, 25(4): 587–595. doi: 10.3321/j.issn:1000-0585.2006.04.004.(inChinese)
Gu XH, Wang K, Pan YC, et al, 2011. Remote sensing quantitative retrieval of farmland organic matter based on HJ1A-HIS image. Geography and Geo-Information Science, 27(6): 69–73.
He T, Wang J, Lin ZJ, et al, 2006. Spectral features of soil organic matter. Geomatics and Information Science of Wuhan University, 31(11): 975–979.
Hussain N, Al-Rawahy SA, Rabee J, et al, 2006. Causes, origin, genesis and extent of soil salinity in the Sultanate of Oman. Pakistan Journal of Agricultural Sciences, 43(1/2): 1–6.
Khan NM, Rastoskuev VV, Sato Y, 2005. Assessment of hydrosaline land degradation by using a simple approach of remote sensing indicators. Agricultural Water Management, 77(1-3): 96–109. doi: 10.1016/j.agwat.2004.09.038
Khan NM, Sato Y, 2001. Monitoring hydro-salinity status and its impact in irrigated semi-arid areas using IRS-1B LISS-II data. Asian Journal of Geoinform, 1(3): 63–73.
Lei L, Tiyip T, Ding JL, et al, 2014. Study on the soil salinization monitoring based on measured hyperspectral and HSI data. Spectroscopy and Spectral Analysis, 34(7): 1948–1953. doi: 10.3964/j.issn.1000-0593(2014)07-1948-06.(inChinese)
Li FQ, Wu ZR, 2002. A study on soil salinization early-warning in semi-arid area: As a sample in west Jilin province. Bulletin of Soil and Water Conservation, 22(1): 57–59.
Ma C, 2014. Research on soil salinization using remote sensing of HJ-1A hyper spectral images. Journal of Arid Land Resources and Environment, 28(2): 180–184.
Metternicht GI, Zinck JA, 2003. Remote sensing of soil salinity: potentials and constraints. Remote Sensing of Environment, 85(1): 1–20. doi: 10.1016/S0034-4257(02)00188-8
Pang GJ, Wang T, Sun JH, et al, 2014. Quantitative analysis of soil salinity content with hyperspectra data in Minqin, Gansu, China. Journal of Desert Research, 34(4): 1073–1079. doi: 10.7522/j.issn.1000-694X.2014.00045.(inChinese)
Peng J, Xiang HY, Wang JQ, et al, 2013. Inversion models of soil water content using hyperspectral measurements in fields of the arid region farmland. Agricultural Research in the Arid Areas, 31(2): 241–246.
Pu Z, Yu RD, Yin CY, et al, 2012. Optimal hyperspectral indices for soil salt content estimation on typical saline soil in arid areas. Bulletin of Soil and Water Conservation, 32(6): 129–133.
Qi HP, Weng YL, Zhao DY, et al, 2010. A study of salinity characteristics and spectral characteristics of salt affected soil in Caka-Gonghe basin. Remote Sensing for Land & Resources,: 4–8.
Rao BRM, Sankar TR, Dwivedi RS, et al, 1995. Spectral behavior of salt-affected soils. International Journal of Remote Sensing, 16(12): 2125–2136. doi: 10.1080/01431169508954546
Weng YL, Gong P, Zhu ZL, 2010. A spectral index for estimating soil salinity in the yellow river delta region of China using EO-1 Hyperion data. Pedosphere, 20(3): 378–388. doi: 10.1016/S1002-0160(10)60027-6
Xu YJ, Ma HC, Peng SY, 2014. Study on identification of altered rock in hyperspectral imagery using spectrum of field object. Ore Geology Reviews, 56: 584–595. doi: 10.1016/j.oregeorev.2013.07.004
Yue DX, Du J, Gong J, et al, 2011. Dynamic analysis of farmland ecosystem service value and multiple regression analysis of the influence factors in Minqin Oasis. Acta Ecological Sinica, 31(9): 2567–2575.
Zhang F, Tiyip T, Ding JL, et al, 2010. The reflectance spectroscopy of geo-objects in Xinjiang on delta oasis of Weigan and Kuqa Rivers. Journal of Infrared and Millimeter Wave, 29(3): 190–195. doi: 10.3724/SP.J.1010.2010.00190
Zhang F, Tiyip T, Ding JL, et al, 2012a. Spectral reflectance properties of major objects in desert oasis: a case study of the Weigan-Kuqa River Delta Oasis in Xinjiang, China. Environmental Monitoring and Assessment, 184(8): 5105–5119. doi: 10.1007/s10661-011-2326-x
Zhang F, Tiyip T, Ding JL, et al, 2012b. The study on the reflectance spectral characteristic of salt-affected soil in typical oasis of the middle reaches of Tarim River. Progress in Geography, 31(7): 921–932.
Zhang F, Xiong FM, Luan FM, et al, 2011. Characteristics of field measured spectral response to alkalinization soil. Journal of Infrared and Millimeter Wave, 30(1): 55–60.
Zhang F, Xiong HG, Ding JL, et al, 2012c. Characteristics of laboratory-field measured spectra responding to alkalinized soil and conversion. Transactions of the Chinese Society of Agricultural Engineering, 28(5): 101–107. doi: 10.3969/j.issn.1002-6819.2012.05.017.(inChinese)
Zhang JM, Qi WW, 2013. Soil salt components and their spectral characteristics in Minqin Oasis of Northwest China. Chinese Journal of Ecology, 32(10): 2620–2626.
Zhang M, Gao TP, Zhu YR, et al, 2014. Spatial distribution characteristics of soil salinity in habitat soil of Halogeton glomeratus in Minqin Oasis. Soils, 46(4): 756–760. doi: 10.13758/j.cnki.tr.2014.04.028.(inChinese)
Zhao ZL, Tiyip T, Sun Q, et al, 2014. Soil spectrum characteristics and information extraction of salinization: A case study in Weigan-Kuqa Oasis in Xinjiang. Progress in Geography, 33(2): 280–288. doi: 10.11820/dlkxjz.2014.02.014.(inChinese)
Zhao ZL, Tiyip T, Zhang F, et al, 2012. Spectral characteristics of soil salt content in typical oasis of Tarim River's middle reaches. Journal of Natural Disasters, 21(5): 72–78.