Within-field correlation between satellite-derived vegetation indices and grain yield of wheat

INTRODUCTION: In recent times, satellite-derived vegetation indices were widely used in the field of agriculture especially in the assessment of crop damage and crop progress as well as in clarification of spatial variability of yield. These types of analyses play an important role in the estimation of the health condition of each crop during its growth and provide an opportunity for timely decision making.

OBJECTIVES: This research aimed to inspect the correlation coefficients, during the crop growth stages, between vegetation indices (VIs) derived from Sentinel-2 imagery and grain winter wheat yield derived from yield monitoring and select the most promising indices for monitoring crop growth and yield estimation.

METHOD / DESIGN: The satellite images in 10m resolution were selected based on crop growth stages, from the end of tillering phase (beginning of March 2019) until the full ripening (end of June 2019). For the analysis, the BBCH-scale for cereals was used. Yield observations were performed at harvest on five fields in one season and twelve VIs were calculated across 10 growth stages. To designate their correlation and dependence, a statistical comparison of the VIs and yield was made. The Pearson’s and Spearman’s correlation coefficients were calculated, and their statistical significance was tested using p-value (at p=0.01, p=0.05).

RESULTS: According to the crop growth stages, the highest correlation coefficients were detected from the early boot stage (BBCH 41) until the middle of development of the fruiting stage (BBCH 73 – early milk). In that period the correlation coefficients varied from 0.39 to 0.84 depending on the field. Based on the location, the highest correlation coefficient values for all 12 indices were recorded for the parcel named C-6 (April 15), and the lowest values for the parcel named C-10 (June 29). Most of the indices showed statistically significant dependence (at the p<0.01 and p<0.05 significant levels) on the yield in the first five growth stages except the chlorophyll vegetation index (CVI) for the parcel named C-11 (p=0.21, p=0.39).

CONCLUSIONS: To conclude, the last growth stage named ripening showed the lowest values both for correlation coefficient and statistical significance which means that VIs also had low values because the reflectance is weak in this growth stage and wheat is about to be harvested. In the first five stages, VIs showed significantly high spectral reflectance values since in this period the leaf is full of chlorophyll pigments. Analyzing the correlation coefficient in different stages of wheat growth, we look at the current state of crops and have the opportunity to take appropriate measures in time to increase yields or save inputs at specific locations.