Construction of a hyperspectral estimation model for total nitrogen content in Shajiang black soil

Abstract

AbstractBackgroundThe real‐time non‐destructive estimation of soil total nitrogen content via hyperspectral remote sensing is important for crop fertilization and precision agriculture development.AimsThe study aimed to use hyperspectral technology for the construction of an estimation model using different methods for assessing total nitrogen content in Shajiang black soil.MethodsIn our study, soil samples were obtained from Shangshui County, Henan Province, China. Using hyperspectral data of soil samples, the original spectral reflectance was transformed into logarithmic, reciprocal, and first derivative spectra. Hyperspectral estimation models for assessing the total nitrogen content were built with spectral indices, competitive adaptive reweighted sampling (CARS) algorithms, and a combination of machine learning methods, including the partial least square regression, support vector machine (SVM), and random forest methods.ResultsThe results showed that after the first derivative transformation, the performance of normalized indices constructed with a combination of two bands at 1401 and 776 nm was better. The coefficient of determination (R2), root mean square error (RMSE), ratio of performance to deviation (RPD), and ratio of performance to the inter‐quartile range (RPIQ) values of the model were 0.84, 0.10 g kg–1, 2.25, and 4.04, respectively. Validation was performed using data independent of the modeling samples, and validation R2, RMSE, RPD, and RPIQ values were 0.92, 0.07 g kg–1, 2.56, and 4.19, respectively. The same sample was analyzed using a CARS algorithm for screening feature bands, and the 61 reciprocal reflectance bands selected as the input for the SVM exhibited the best performance. A modeling R2 value of 0.92, RMSE of 0.07 g kg–1, RPD of 3.09, RPIQ of 5.97, and validation R2 of 0.96, RMSE of 0.06 g kg–1, RPD of 3.24, and RPIQ of 8.47 were observed.ConclusionsThe estimation models constructed using the two methods exhibited good ability to determine the total nitrogen content. The accuracy of the SVM model was slightly higher than that of the indices model, but both models could facilitate the rapid estimation of total nitrogen content in Shajiang black soil. The findings can provide a technical reference for the estimation of the levels of soil total nitrogen and other nutrients.