Remote estimation of crop canopy parameters by statistical regression algorithms for winter rapeseed using Sentinel-2 multispectral images

Abstract

Еstimation of crop canopy parameters is important task for remote sensing monitoring of agriculture and constructing strategies for within-field management. The main objective of this study is to evaluate the retrieval from Sentinel-2 images by parametric and non-parametric statistical models several crop canopy parameters for monitoring before winter and after winter rapeseed crop in real farming conditions of North East Bulgaria. For the calibration of the models in-situ data from three field campaigns is used. For most of the studied parameters models with good accuracy were identified, except for aboveground fresh biomass. The best identified model for vegetation fraction (RMSEcv=0.14%) and plant density (RMSEcv=9 nb/m2) were parametric models with three band vegetation index (3BSI-Tian) and linear fitting function for the first, three band vegetation index (3BSI-Verreslt) and polynomial for the second parameter. For aboveground dry biomass (RMSEcv=52 g/m²), mean plant height (RMSEcv=4cm) and nitrogen concentration in fresh biomass (RMSEcv=2%) the best models were non-parametric, Gaussian Processes Regression for the first parameter and Variational Heteroscedastic variant of the Gaussian Processes Regression for the other two.

Code and data availability

The paper reports in-situ rapeseed canopy measurements and Sentinel-2 based regression modeling, but provides no public deposit of its field data, imagery subsets, models, or analysis code. ARTMO is a third-party generic tool, and the Copernicus/FAOSTAT links are generic data sources, not paper-specific assets.

No evidence-backed public reproduction asset is currently recorded.