Using Sentinel-2 Satellite Images to Estimate Traits of Forage Grasslands

Sammanfattning: In this project, regression models based on data from field measurements and spectral information extracted from satellite imagery were used to estimate traits of forage grasslands; dry matter yield, canopy average height and total leaf chlorophyll. Four fields at SLUs Röbäcksdalen field station were sampled on 22 occasions and a total of 198 samples, including measurement of the highest plant, canopy height, leaf chlorophyll content, canopy spectral reflectance and biomass were collected. Two regression methods, partial least squares (PLS) and support vector machines (SVM), were used to build regression models using different subsets of the available spectral information. Model calibration was performed with 2/3 of the dataset and model validation was performed with the remaining 1/3 of the dataset. It was shown that the models built with SVM outperformed the models built with PLS, during both calibration and validation as well as for all different traits and subsets of spectral information. Field measurement and regression model results were discussed and limitations, their significance and possible improvements were considered. It was concluded that using spectral information from satellite images is a promising approach for estimation of traits in the field and could be used to build tools as a tool to support farmers’ decision making.