Field-scale estimation of sugarcane leaf nitrogen content using vegetation indices and spectral bands of Sentinel-2: Application of random forest and support vector regression

Abstract

Nitrogen is an essential factor for assessing the quality of sugarcane during the growing season, as deficiency of this component significantly reduces crop yield . The Kjeldahl method is the most common approach to measuring sugarcane nitrogen, but it is a laborious, costly, and time-consuming process. Conversely, multispectral satellite imagery can provide timely, cost-effective, and large-scale information on the nitrogen content in sugarcane fields. The current study applied random forest (RF) and support vector regression (SVR) models to estimate sugarcane leaf nitrogen using vegetation indices and spectral bands of Sentinel-2. In-situ data was taken from 45 farms (1125 ha) in a sugarcane production agro-industrial complex in southwest Iran. The global environmental monitoringindex (GEMI), chlorophyllindexgreen (Clgreen), and Sentinel-2 red-edge position index (S2REP) were found to be the most important variables related to sugarcane nitrogen. The coefficient of determination (R2) for RF and SVR was 0.59 and 0.58, respectively, and the corresponding root mean square error (RMSE) was 0.08 and 0.09, respectively. Despite the similar performances of the two models, RF showed higher accuracy; however, to improve the results, the use of multi-temporal data for model calibration is recommended.