Crop type mapping and LAI, fAPAR, and fCover prediction in winter wheat fields with Sentinel-2 data

Abstract

The aim of this study is to map winter wheat fields in designated study area in Bulgaria and to analyze the spatial variation of various biophysical parameters of winter wheat crops in these fields using multispectral satellite imagery. The study uses Sentinel-2 data for image classification and to predict the Leaf Area Index (LAI), fraction of absorbed photosynthetically active radiation (fAPAR), and fraction of vegetation cover (fCover) of the crops. The study area is situated in the Danubian plain in Bulgaria, a heavily cultivated agricultural region in Europe. To distinguish winter wheat fields from other agricultural fields, classification techniques were applied using two methods, Support Vector Machines (SVM) and Random Forest (RF), to identify the winter wheat fields in different phonological phases during the growing season. Both classification methods performed with similar accuracy and showed high accuracies in classifying winter wheat using Sentinel-2 images at various phonological phases (F1 > 93%; tillering; F1 > 95%; stem elongation; F1 > 94%; anthesis). To predict the LAI, fAPAR, and fCover dynamics in the winter wheat fields, regression models were used, calibrated with vegetation indices and in situ data. Maps displaying the within-field variation of LAI, fAPAR, and fCover were created for two growth stages: tillering and stem elongation.