| Agricultural production and the ecological environment in the coastal region of China are seriously affected by soil salinization.High-precision dynamic monitoring of soil salinization is essential for accurately obtaining the current status of soil ecology in the coastal region.With the remarkable advantages of high spectral resolution,integration of maps,and a high amount of information,hyperspectral remote sensing provides important technological support for the quantitative estimation of soil salinity.As an important part of the soil,moisture is an almost inevitable spectral interference factor in soil salt estimation,and it is also one of the important reasons for the reduction of its estimation accuracy.The quantitative estimation of soil salinity based on the removal of moisture interference is of great theoretical and practical significance to the ecological restoration of coastal saline soil and the improvement of soil ecological conditions.This paper takes the coastal saline region of the Yellow River Delta as the research area.In a controlled laboratory environment,145 surfaces(0-20 cm)of saline soil samples after drying,grinding and sieving were re-wetted to simulate the different gradients of soil moisture(1-50%)and obtain measured hyperspectral data.A hyperspectral estimation method that can weaken or eliminate the interference of soil moisture was proposed,which is named the "moisture resistance" estimation method of soil salinity.The soil spectral characteristics under soil moisture and soil salt stress were analyzed,and the hyperspectral index models and statistical models of "moisture resistance" for estimating the soil salinity were constructed.The main contents and conclusions were as follows:(1)The quantitative relationship between soil moisture,salinity,and soil spectrum was untangled.Soil moisture in the coastal region is characterized by high content and high variability,which limits the research of saline soil spectra with different moisture content and reduces the accuracy of soil salinity estimation.Using reflectance data of saline soils with different moisture gradients,soil spectral analysis under moisture and salinity stress was carried out.The results showed that moisture was the dominant factor for the decrease of spectral reflectance of saline soil with salt content between 0.56-5 g/kg.When the moisture content increases to 50%,the moisture content in the soil pores tends to be saturated,and the soil reflectance begins to increase.Under the same moisture content,as soil salt content continuously increases above 10g/kg,salt begins to change the trend of soil reflectance changing with soil moisture content.Further analysis indicates that moisture conceals part of the spectral changes caused by soil salt,which is an important reason for the reduction of the accuracy of soil salt quantitative estimation.The spectra around 840 nm,1915 nm,and 2025 nm still maintained a high sensitivity to soil salinity under moisture interference and were regarded as soil-salt sensitive bands resistant to soil moisture interference,which provided new ways for the "moisture resistance" estimation of soil salinity in the coastal region.(2)The hyperspectral index model and statistical model for the "moisture resistance" estimation of soil salinity were established.Given the practical problem of reduced precision of soil salt estimation due to moisture interference in the coastal region,the hyperspectral index model was proposed.Based on the original spectrum and Savitzky-Golay and First Derivative(SG-FD)preprocessed spectrum,the model was constructed in the form of spectral reflectance(R),difference index(D),ratio index(SR),and normalization index(ND),and the "moisture resistance" mechanism was explained by Spearman correlation analysis.The results showed that the d SR and d ND models based on SG-FD preprocessed spectrum had the best accuracy in estimating soil salinity,and the optimal wavelengths of the index were both around 1901 nm and 1992 nm.The "moisture resistance" mechanism of the d SR and d ND models was as follows: on the one hand,the correlation between soil spectrum and soil salt content was significantly enhanced after SG-FD spectral pretreatment.On the other hand,the index model effectively extracted and utilized the soil salt-sensitive wavelengths located at 840 nm,1435 nm,1915 nm as well as near the 2025 nm edge with "moisture resistance" potential.The model provides a new method for the "moisture resistance" estimation of soil salinity in the coastal region.In addition to the index model,the "moisture resistance" statistical model for soil salinity estimation was also proposed.The model was constructed based on PDS spectrum converted by piecewise direct standardization(PDS),OSC spectrum filtered by orthogonal signal correction(OSC),and unprocessed SG-FD spectrum combined with partial least square regression(PLSR).The "moisture resistance" mechanism was explained by combining Spearman correlation analysis and model variable importance for projection(VIP)analysis.The results showed that the order of soil salinity estimation performance of the statistical model was OSC-PLSR >PLSR >PDS-PLSR.The "moisture resistance" mechanism of the OSC-PLSR model was as follows: on the one hand,the correlation between the spectrum filtered by OSC and soil salt content was improved.On the other hand,the OSC-PLSR model precisely refined the important wavelengths at 1410 nm,1500 nm,and 2100 nm,while improving the consistency between the wavelengths with high soil salt correlation and the important wavelengths of the model.The model provides another new method for the "moisture resistance" estimation of soil salinity in the coastal region.Facing the requirements of soil salinity "moisture resistance" estimation in the coastal saline region of the Yellow River Delta,the most suitable model method was determined.The performance of the index model and the statistical model in estimating soil salinity under the same moisture content were compared and analyzed.The results showed that the statistical model of OSC-PLSR was more stable and reliable than the index model.The reason is that the model was constructed based on the OSC spectrum,which realizes the filtering and moisture removal directly at the spectral level. |
PDF Full Text Request