UAV-based 3D models of olive tree crown volumes for above-ground biomass estimation

Variation in trees crown volume can be assumed as useful tool to estimate crop response to agronomical management practices and to provide information for a correct mass balance of the crop. In that direction, innovative technologies applied in agriculture, such as UAV high resolution images and structure from motion (SfM) algorithms, provide a fast and cost-effective solution to perform accurate trees volume reconstruction. In this study, the crown volume of 15 olive trees (Olea europaea L., 'Leccino') was estimated by means of two UAV flight campaigns before and after a pruning treatment. High resolution RGB images were acquired at 30m flight quote with a 20MP camera angled at 45° with respect to the vertical at the ground and flights direction. Image processing was performed by means a 2D and 3D methods. The tree volume was calculated as the sum of polygons based on pixel surface, related to canopy height model (VCHM) in a 2D approach and as the volume that envelopes the set of 3D dense points cloud (with 3 different resolution levels V3Dhigh, V3Dmedium, and V3Dlow) in the 3D approach. Three levels of pruning (in terms of biomass removed) have been applied. The removed biomass volumes, computed as differences between crown volumes before and after cutting, were correlated to pruned biomass and the ANOVA of the linear regression have been conducted. The accuracy of the estimation was evaluated by means of a training/validation routine.The relation between VCHM and pruned biomass reported high relative root mean square errors (RRMSE) values suggesting a scarce efficiency of the method in biomass estimation. On the contrary, V3Dlow lead to the lowest RRMSE (22%) assuring a good prediction of biomass. Moreover, the low requested resolution in V3Dlow, (affecting imagery process duration) represents an additional encouraging issue of this approach.