Estimation of Soil Organic Carbon in Plow Layer of Northern Henan Plain Based on Sentinel-2 Data
To accurately understand the content and spatial distribution of soil organic carbon(SOC)in the plow layer of Plain,in this paper,actual SOC data were used from Huaxian County,integrated with Sentinel-2 reflectance spectra and spectral indices,and data of meteorological parameters,soil properties,terrain and human activity factors,and then the predictive accuracy of the Elastic Net(EN),Partial Least Squares(PLS),and Random Forest(RF)models were quantitatively evaluated and compared under various factor combinations,the optimal factors and key variables that influence SOC estimation,as well as the spatial variation patterns were revealed.The results showed that in terms of model accuracy and stability,RF was best,followed by EN and PLS.Specially,RF model achieved the highest precision with the best combination of factors including reflectance spectra,spectral indices,meteorological data,and soil properties,with the coefficient of determination(R2)of 0.63,the ratio of performance to inter-quartile range(RPIQ)of 2.75,and the mean absolute percentage error(MAPE)of 5.86.Moreover,the 12th band with a central wavelength of 2 190 nm(B12),potential evapotranspiration,soil bulk density,and the normalized difference soil index(NDSI)cumulatively accounted for 69.63％of the importance in RF Model,becoming the key variables in RF model and playing a decisive role in SOC estimation.The range of SOC in the study area was between 6.15 and 11.08 g/kg,with an average of 8.16 g/kg,lower than the national average of 22.28 g/kg.Spatially,the area with higher SOC content was mainly located in the northeastern part,while the area with lower content was primarily located in the southwestern part.
国家科技期刊平台
NSTL
万方数据
