Study On Spectral Characteristics Of Salinized Soil Based On Controlled Experiment

Soil salinization is one of the common forms of land degradation in arid and semi-arid areas.It is also one of the main limiting factors that affect the social and economic development and obstacle to the sustainablieties of agriculture development.Therefore,studying the essential causes of salinization and prevent the deterioration of soil is the main research topics for scholars all over the world.However,due to the fact that the traditional methods used in the past are time consuming,high costing and limited sampling,they cannot meet the needs of research.Remote sensing technology with macroscopic real-time observation potential has effectively solved these limitatioms and has been widely applied.Even though the multy spectral remote sensing thchnology make it possible to qualitively monitoring soil salinization problems on timely and widely,but its law spectral resolution unable to quantitavely analyse the composition of soil properties.The hyperspectral remote sensing thechniques chnarechterized with high spectral resolutionahs solved these limitations and getting popularities in recent years.However,the field measured hyperspectral data has often affected by the surrounding environment,such as wind speed,sun zenith angle,soil roughness,soil moisture,soil particles and the weather conditions,and resulting in inaccuracies of the soil spectral data.Considering these practical problems,the effects of moisture and salt content on soil spectral properties were analyzed under laboratory controlled experiments.First,the controlled soil samples with different soil water content?110 samples,water content ranges from 0 to36%?,different salt content degrees?total 75 samples,salt content degrees are ranges from 0 to 50 g/kg?and different salt ingredients?including NaCI,CaCI₂,MgCI₂,Na₂CO₃,NaHC₃ and Na₂SO₄;40 samples for each,240 soil samples in total?was prepared based on salt content,soil salt ions,soil moisture values of soil samples collected in Ugan-Kuqa river oasis.Then,the soil spectres of those samples was measured indoor condition and the soil spectral data were processed using different mathematical transformation methods including differential,cubic,square root,logarithmic and continuum removing.The correlation was analyzed between these processed spectral data and soil salt and water contant.The best sensitive bands were selected and soil salt and water content estimation models were established using partial least square regression and principle content regression methods.The main results of this study are as follows:?1?Through the analysis of the salinity,electrical conductivity and eight major ions of the soil in the Weiku oasis,it is found that the degree of salinization of the oasis is different,the maximum salt content is 69.85 g/kg,and the minimum salt content is 0.15g/kg,the average value of soil salt content is 19.89 g/kg,belong to heaviely salinization.The proportions of Na⁺and Ca²⁺ions in the positive ions are86.85%and 8.90%,the proportions of Cl^-and SO₄^2-ions in the anions are 64.59%and32.10%.It can be conlcluded that the main salty ions salinized soil are Na⁺,Ca²⁺,Cl^-and SO₄^2-.After the correlation analysis between the eight major ions,found that the highest correlation coefficient between Cl^-and Na⁺ions is 0.89,and the correlation coefficient of SO₄^2-ions and Ca²⁺is 0.68.It is indicate that there is a certain correlation between cation and anion.?2?Compared with the soil spectral curves obtained in the indoor and field,it is found that the spectral data obtained in the field are more disturbed by the external influence than indoor measurement,which greatly affects the extraction of accurate spectral information.Correlation analysis between salt content and original spectrum shows that there is a certain correlation between reflectance and soil salinity.For the soil spectral curves of different types of salt,there is no significant difference in the spectral reflectance curves of the six types of salt,but the specific location of the absorption Valley and the reflection peak is quite different.These differences are mainly reflected in the correlation between soil reflectance and salt content.?3?Several mathematical transformation forms of soil spectral reflectance of different degrees of salt are carried out.The best bands for soil salinity are 540nm,1390 nm,1430nm,2010nm,2270 nm,2350nm,1684nm,1921nm,and 1928nm.In the PLSR and PCR models established by these bands,the best modeling accuracy and prediction accuracy are be seen in the original reflactence data with second ordered transformation and the R² of the model is 0.811,the RMSE is 2.456,and the RPD is 2.72.The accuracy of the models established by PLSR is higher than that of PCR model.?4?For different types of spectral characteristics,it is found that the first and second order differential methods are very favorable to reflect the best sensitive waveband.After correlation analysis,it is found that the most sensitive bands for NaCI are 618nm,622nm,627nm,1342nm and 2260nm;highly correlated bands for CaCI₂ are 680nm,974nm,1753nm,1914nm,1922nm and so on.The most sensitive bands for Na₂SO₄ are 1985nm,2196nm and so on;the sensitive bands for Na₂CO₃ are794nm,1368nm,1434nm,2078nm,2104nm,etc.In the estimation models,the models which established by using second ordered differential reflactence data is the best among the all models.The R² and RPD values of soil NaCI content prediction model are 0.894 and 2.95 respectively.For Na₂SO₄,the R² and RPD values are 0.86and 2.47,respectively.The R² and RPD of Na₂CO₃ are 0.78 and 2.07 respectively;the values of R² and RPD of NaHCO₃ are 0.86,2.47;R² and RPD values for MgCI_2,CaCI₂ are0.86,2.71 and 0.88,2.93,respectively.?5?For the soil moisture prediction model,the partial least squares regression prediction model using three transformed bands,including R`1880,logr1932,?logr?`1897,as variables shows the highest accuracy of the R2 value of the modeling and validation data sets of 0.946 and 0.922,and the RPD value of the modeling and validation data sets is up to 4.243 and 3.517.