Estimation of nitrogen content across grass communities at Telperion Nature Reserve using Sentinel-2

Abstract

Grass nitrogen is the main indicator of forage conditions in a rangeland environment. The main objectives of the research were to map the quality and quantity of common grass communities and to predict Nitrogen (N) content across different grass communities. A machine-learning algorithm of Support Vector Machines (SVM) was tested in the mapping of grass quality and quantity. An overall accuracy of 72.68% was achieved for the mapping analysis which demonstrated the capability of the Sentinel-2 10m resolution in discriminating the spectral properties of different grass communities. The foliar nitrogen was predicted using univariate regression, stepwise multiple linear regression (SMLR), multivariable regression methods, partial least square regression (PLSR) and random forest (RF). Foliar N was predicted using multivariate regression models; the best model was selected based on the highest coefficient of determination (R2) value, and the low root mean square error (RMSE). The best RF model for foliar N estimation was based on the simple ratio (SR) index because the model attained the highest prediction accuracy of 35%. The study demonstrates the applicability of Sentinel-2 MSI utility in mapping and estimation of leaf N at a landscape scale .The results of both regression models (univariate and multivariate) such as random forest and partial least squares indicated that the inclusion of the Sentinel-2 MSI red edge bands provides an opportunity to accurately map and estimate leaf bio-chemical composition using remote sensing techniques.