| Different soil horizons contained rich information about salt dynamics, migration, and accumulation of soil profile, established the soil salt prediction model based on hyperspectral reflectance had a certain reference value in quantitative management, monitoring and forecasting. Taken the west bank of Boston lakeside oasis as the study area and determined five kinds of typical surface cover soil profile (0～100cm) about tamarisk land, cultivated land, unused land, forest land, and reed land as the research object, divided the types of soil salinity profile and soil grass-roots level according to the characteristics of salt content and accumulation in soil profile, and divided into soil-salinity-class and saline soil type at the different layers, used the descriptive statistical method to contrastive analysised the characteristics of salinity and Hyperspectral reflectance at the different layers. Based on four kinds of mathematical methods about first order differential, second order differential, logarithmic transformation and inverse logarithm increase to enhance the soil original hyperspectral reflectance after pretreatment and empirical mode decomposition (EMD). Extract the band which has the max-relativity between the content of salt, Ca2﹢,Mg2﹢,Na﹢＋K﹢,SO42-,CL- and the original hyperspectral reflectance and enhanced hyperspectral reflectanceas in soil profile as the optimal band, used the method of partial least-squares regression (PLSR) to set up the forecasting model about the content of salt, Ca2+,Mg2+,Na++K+,SO42- and CL- basing on the hyperspectral reflectanceas(both prediction model through the confidence level was more than 95% of F test), tested the accuracy of model and applied the model in sample region, preliminarily established soil profile spectral library of tamarisk land, cultivated land, unused land, forest land, and reed land. The conclusion included the following four aspects:(1)According to the characteristics of salt content in soil profile clustering Tamarisk and unused soil salt section was divided into table type, the types of Cultivated land, Forest land and Reed land was divided into the average. According to the effective thickness of soil layer within the Numbers of 1,2,3,4,5, named ultra-thin layer 1, thin soil layer 2, thin layer of soil, middle and thick layer of soil five at the grass-roots level of different surface cover soil profile,used the uppercase letters T, F, W, L, P to named Tamarisk land, Cultivated land, Unused land, Forest land and Reed land respectively. T1 and W1 salinization of grade, W2 and P1 salinization rank as weak salinization, the rest of the grass-roots level of salinization are very light.T1 and W1 as sulfate-chloride salinization soil, T2, L3, P1 as sulfate soil salinization, the rest of the grassroots are chloride-sulfate soil salinization. HCO3- in soil salt section as a whole was small, the average content of each layer was 0.06 g/kg. CL" in soil salt section average content of each laye r was 0.23 g/kg, the tamarisk soil salt content in the profile has obvious vertical variation, the average content of each layer was 0.39 g/kg, vertical variation coefficient was 113.03%. SO42- in the tamarisk maximum content of soil salt section, the average content of each layer was 0.79 g/kg, it has table together characteristics. Ca2+ and Mg2+ in the tamarisk soil salt content were higher in profile, the average content of each layer were 0.15 g/kg and 0.08 g/kg, the variation coefficient were 52.40% and 66.28% respectively. Na+ and K+ in the tamarisk salt section, the average content of each layer was 1.011g/kg, the variation coefficient was 59.51%. (2) T2’s reflectance was highest in tamarisk soil profile within the band of 3501800 nm, T3’s reflectance was minimum, In the range of 350～1800 nm, T3’s reflectance was highest, T1’s reflectance was minimum. The reflectance characteristics of farmland soil profile was:The difference of wavelength range within the band of 350～1800 nm was small, F2’s reflectance was maximum and the F5 was minimum within 1900～2500 nm.The reflectance features of Wasteland soil profile were:Within the band of 350～1800 nm, the largest reflectance was W5, and the minimum reflectance was W2, the reflectance of W4 was maximum in the range of 1900～2500, W3 and W1’s reflectance were lower.In the range of 350～1800 nm in reed soil profile,P1’s reflectance was biggest and P5’s reflectance was minimum, within the band of 1900～2500 nm, P2’s reflectance was biggest and P4’s reflectance was minimum. The reflectance features of woodland soil profile were:the difference of reflectance were small within 350～1800 nm, within the band of 1900～2500 nm,L3’s reflectance was highest and L4 was minimal.T2, L3 and P1 were sulfate salinization soil,and had highest reflectance within the band of 350～1800 nm in the soil profile of Tamarisk, forest land, reed.T1 and W1 were sulfate-chloride salinization soil,the reflectance of T1 and W1 were minimum in the range of 1900～ 2500 nm in Tamarisk and desert vegetation soil profile.(3)Partial least squares regression analysis method was adopted to establish the salt content in soil profile, the regression model of the contents of salinity、Ca2+、Mg2+ Na++K+ and SO42- by hyperspectral reflectance had a higher accuracy overall. The concentration prediction model expression of Ca2+ was:Y=0.151-31.091R(490)+ 22.829R(500)-1.544R(510)-1,344R(520)-18.815R(530)+33.176R(840)(R2=0.91,RMSE=0.23);The concentration prediction model expression of Mg2+was Y=-0.061-1.921R(660) +1.032R(670)+6.117R(690)-6.834R(700)+2.345R(730)-0.684R(780),(R2=0.89,RMSE=0.18);The concentration prediction model expression of SO42- was Y=0.064-13.388 R(1150)+38.61R(1180)+3.884R(1190)-35.105R(1230)-53.14R(1240)+59.598R(1260) (R2=0.85, RMSE=0.26)The concentration prediction model expression of Na++K- was Y= 0.046-23.489R(1690)+24.294R(1720)+4.342R(1730)+19.988R(1750)-20.202R(1760)-4.406R(1770) (R2=0.79,RMSE=0.31)The prediction model of salt content was Y=-0.294+74.427R(1610)-129.135R(1650)+45.701R(1660)-54.053R(1670)+59.247R(1680)+6.179R(1700) (R2=0.81, RMSE=0.31)(4) The results of sample region application showed that the contents of salinity、 Ca2+、Mg2+、Na++K+ and SO42- exist linear correlation regression model predicted and the measured values, the determination coefficient were 0.63,0.80,0.74,0.80 and 0.74 。 The prediction model has feasibility and suitability, it can provide theoretical basis for the research about regional soil salt quantitative inversion. |
PDF Full Text Request