Spatial-temporal heterogeneity of yield, protein concentration, and leaf area index in grassland agroforestry systems can be modeled from UAV-borne imagery

The introduction of alley-cropped trees in grassland is known to improve carbon sequestration, soil microbial biodiversity, and nutrient availability. Many studies examined the spatio-temporal effects of trees on grassland plant parameters using transect points; however, they relied on few point measurements with high uncertainty in large parts of the agroforestry systems (AFS). Dry matter yield (DMY), crude protein concentration (CP), and leaf area index (LAI) have rarely been studied on a sizeable areal scale in grassland AFS. Unoccupied-aerial-vehicle (UAV) 􀀀 borne imagery can be utilized to model DMY, CP, and LAI at extensive coverage and high spatial resolution. This study employed UAV-borne red, green, blue (RGB) and multispectral (MS) sensors to model DMY, CP, and LAI from 3D point clouds, reflectance information, vegetation indices, and texture variables. Data were captured during five dates from spring to summer 2020 in an undisturbed first grassland growth in a grassland alley cropping AFS located in Central Germany with two different grassland mixtures. DMY, CP, and LAI could be modeled with good accuracies (DMY: RMSEval = 0.42 t ha􀀀 1, R2val = 0.82; CP: RMSEval = 1.91 %, R2val = 0.74; LAI: RMSEval = 0.35, R2val = 0.73). The models were used to produce maps, highlighting the spatio-temporal heterogeneity of the plant parameters studied. After transforming raster data into point data, the effect of grassland mixtures on DMY, CP, and LAI was proven to be significant using mixed models. The study illustrated that UAVborne imaging can be utilized as a method for spatio-temporal analysis of grassland AFS at high spatial resolution, enabling spatial coverage of entire AFS to incorporate the whole picture of fine-scale spatio-temporal heterogeneity into the statistical analysis.