Estimating The Soil Particle Size Based On Spectral Reflectance

Soil particle size,as one of the most important soil properties,has been recognized as an important factor affecting soil structure,soil hydraulic properties and plant nutrient availability.In addition,it’s very meaningful to understand the temporal and spatial variation of soil particle size for surface infiltration and wind and water erosion in arid and semi-arid regions.Therefore,it is necessary to estimate soil particle size at different regional scales.However,in the current estimation of soil particle size,semi-empirical and empirical methods are mainly based on single-band reflection information.The heterogeneity between different types of soil in the region or the change of incident conditions will change the reflection,which may be an influencing factor for the estimation of soil particle size based on the reflection information.In order to improve the effectiveness of estimating soil particle size from reflectance spectrum data,it is necessary to explore the use of reflectance information of different types of soil under different lighting conditions to establish a relationship with soil particle size and improve the estimation accuracy.Therefore,it is necessary to eliminate the influence of soil types and incident angles for estimating of soil particle size accurately.Based on hyperspectral reflectance data,this paper analyzes the relationship between soil particle size and spectrum,and confirms the influence of soil type and incident angle on the relationship between spectral reflectance and soil particle size.In order to eliminate the influence of soil types and light conditions on soil particle size estimation,this study constructed four spectral indices for estimating soil particle size.The results show that the simple difference spectral index SDSI(R₉₀₀-R₂₁₅₀),proposed based on the spectral reflectance of different types of soil,has a strong relationship with different types of soil particle size compared with other single-band and spectral indices（R²=0.88）and has no incident angle dependence.The SDSI index also has high estimation accuracy（RMSE=0.12mm,RRMSE=6.2%）when estimating the different types of soil particle size under different incident angles in the field.The results demonstrate that the SDSI index is insensitive to soil type and incident zenith angle under both laboratory and field measurement conditions.In addition,the two bands R₉₀₀ and R₂₁₅₀ that make up the SDSI index are located in the atmospheric window bands of most satellites.This study selected three satellites Landsat8（B5-865nm,B7-2200nm）,Sentinel-2A（B8a-865nm,B12-2190nm）and Landsat5（B4-840nm,B7-2223nm）,which are commonly used for surface monitoring.Through spectral resampling,the SD（Sipmple Difference）difference spectral index is constructed based on the satellite band instead of the two bands in the SDSI index(Landsat8:(B5-B7(R₈₆₅-R₂₂₀₀)),Sentinel-2A:(B8a-B12(R₈₆₅-R₂₁₉₀)),Landsat5:(B4-B7(R₈₄₀-R₂₂₂₃))),established the relationship with soil particle size,based on Landsat8,Sentinel-2A and Landsat5 satellites.There was a strong correlation between SD index and soil particle size,and R² was greater than 0.85.The exponential estimation model calculated based on the modeling dataset of laboratory was applied to the validation dataset of the field,and good estimation results were also obtained（Landsat8:RMSE=0.19mm,RRMSE=9.8%;Sentinel-2A:RMSE=0.19mm,RRMSE=9.8%;Landsat5:RMSE=0.19mm,RRMSE=9.7%）.It also proves that the proposed SDSI index can not only eliminate the influence of different types of soil and incident angles,but also has high soil particle size estimation accuracy after resampling it to satellite multispectral.In addition to the above methods,this study also used the Partial Least Square Regression（PLSR）method to establish a soil particle size estimation model.The results show that the estimation model established by this method based on the original reflectance spectra measured by different types of soil particle size also has a strong correlation with soil particle size（R²=0.88）,and the prediction performance of the model is also well（RMSE=0.18 mm）.In summary,this study established particle size estimation models for different types of soil based on spectral index and partial least squares regression,and found the following results:The estimation model established by spectral index and partial least squares regression eliminates the effects of soil type and incident angle have good estimation performance for soil particle size.In comparison,the newly proposed SDSI index not only uses fewer bands,the calculation method is simple,and when it is applied to the field measurement data set,it can accurately reflect the changing trend of soil particle size.Therefore,SDSI index(R₉₀₀-R₂₁₅₀)is a simple and more effective method for estimating soil particle size,which is conducive to the rapid characterization of soil texture distribution and provides a theoretical basis for field measurement of soil particle size.