Estimation of soil total nitrogen using the synthetic color learning machine (SCLM) method and hyperspectral data

Abstract

Hyperspectral remote sensing is a potentially feasible, nondestructive and rapid tool for monitoring soil total nitrogen (TN) content. Nevertheless, soil color often decreases spectral reflectance and severely affects the accuracy of TN estimations. To improve TN estimation accuracy, the synthetic color learning machine (SCLM) method was presented in this study. This method combines machine learning concepts with rapid decoloring functions. Soil samples were collected from Renqiu, Cangzhou and Fengfeng, all located in Hebei Province, China. The soil spectra were measured using an ASD FieldSpec 3 spectrometer in a dark room, and the TN content was determined using the Kjeldahl method. Based on the 66 synthetic soil color values acquired via reflectance spectra, 1254 TN models were established. The optimal model was obtained when the synthetic soil color was 0.1R + 0.5G + 0.4B and the fuzzy coefficient parameter was 7. This model was treated as the ultimate model of the SCLM method, and it produced better performance outcomes (R2c/R2v = 0.881/0.803, RMSEc/RMSEv = 0.096 g/kg/0.135 g/kg, MREc/MREv = 6.162%/8.994%) than the partial least squares regression (PLSR) model. Therefore, the SCLM method can potentially improve the estimation accuracy of TN by rapid and effective decoloring. This study determined that further investigation into estimating TN using unmanned aviation vehicle hyperspectral images is warranted, despite the challenges inherent in the soil remote sensing field.