12Spectral analysis is a useful tool for the rapid and accurate prediction of soil properties. Our ]) and sodium adsorption ratio (SAR). Three methods were applied, i.e., stepwise multiple 15 linear regression (SMLR), partial least squares regression (PLSR), and support vector machine 16 (SVM). Estimation models for four soil properties were developed using three different spectral 17 processing and transformation techniques, i.e., reflectance (R e ), logarithm of reciprocal R e (LR), and 18 standard normal variable of R e (SNV) were used. A total of 36 models were established. Of these, (RPD) were all smaller than the 1.4 threshold. However, the models for SAR~R using PLSR (R The prediction of the physico-chemical properties of degraded soils is very important for crop 37 growth, low quality land improvement, and environmental sustainability (Childs and Hank, 1975; 38 Metternicht and Zinck, 2003 method to quantitatively characterize soil chemical properties (Ben-Dor et al., 1995; Aldabaa et al., 48 2015), and offers both cost and statistical power advantages (Cohen et al., 2005). Reflectance 49 spectroscopy analysis of ground objects is a good means of predicting soil properties, and of different soil properties (Islam et al., 2003; Davies, 2005; Cambule et al., 2012).
52There have been a number of studies relating soil physico-chemical properties to 53 reflectance-related features (Mahesh et al., 2014; Kim et al., 2014). Lü et al. (2013) and other physical-chemical properties (Li et al. 2012; Zhang et al., 2016).
63Since reflectance data are numerous and complex, the modeling approaches are crucial to 64 achieve an accurate estimate of soil properties. Zhang and Li (2016) conducted in the region, and has achieved good results in improving soil quality.
117The soils in the region were characterized as typical grey desert soils (Luo, 1985). The samples
118were collected from the 22nd of April to the 15th of May, 2013. Plant residues present on the soil 119 surface were carefully removed before sampling. A total of 78 samples were randomly collected 120 from within a 2 to 3 km range. The sampling depth was from 0 -10 cm, and the sampling location 121 is shown in Figure 1.Each soil sample was put into a cloth bag and transported to the laboratory. All of the soil 124 samples were air-dried, ground, and passed through a 2 mm sieve. The soil moisture of all the the R e data (Buddenbaum and Steffens, 2012). LR is used to reduce the effects of noise, light, and 153 background on R e . SNV is adopted to eliminate the effects of solid particle size and optical path 154 change of surface scattering on R e . SNV is calculated by (Barnes et al., 1989):where µ is mean value of R e , and σ is standard deviation of R e .
157
Model establishment and validation
158Soil samples were randomly divided into a calibration set for model establishment, and a
171The model establishment using the SMLR method was performed using SPSS 17.0 software, The leave-one-out cross-validation was chosen as the validatio...