The potential of UAV-borne imagery for assessing the productivity, quality and botanical composition of grasslands

Abstract:

Knowledge on the productivity, forage quality and botanical composition is important for the management of grasslands. Remote sensing (RS) is a promising tool for estimating field data, however, the applicability of RS prediction models depends on the variability of underlying calibration data. Major aims of the studies presented here were (i) to build prediction models for aboveground biomass (AGB), symbiotic N fixation, forage quality and botanical composition based on unmanned aerial vehicle (UAV)-borne imaging spectroscopy and (ii) to generate maps using the best models obtained.

Multispectral sensors mounted on an UAV were used to assess AGB and symbiotic N fixation for two legume-grass mixtures through a whole vegetation period based on UAV multispectral information. The prediction models covered different proportions of red clover and alfalfa (0–100% legumes) to represent the variable conditions in practical farming. Accuracies were substantially improved by the inclusion of texture features. The best model was generated for the whole dataset by random forest (RF) modelling with an rRMSE of 10%. For N fixation accuracy of the best model was based on RF including texture (rRMSEP = 18%), which was not consistent with crop specific models.

A UAV with a hyperspectral camera on board was utilised to acquire spectral images from eight grasslands which largely differed in terms of vegetation type and cutting regime, and crude protein (CP) and acid detergent fibre (ADF) concentration of the forage was assessed at each cut. Support vector regression provided the highest accuracy for CP (nRMSEp = 10.6%), while cubist regression model proved best for ADF (nRMSEp = 13.4%).

UAV-borne RGB and thermal imaging, as well as photogrammetric canopy height modelling, were applied to map lupine coverage with object-based image analysis. Images were segmented by unsupervised parameter optimisation into image objects representing lupine plants and grass vegetation. Image objects obtained were classified using RF based on objects’ attributes. The models yielded a mean prediction accuracy of 89 %, and 0.78 mean kappa statistics. The maximum difference in lupine area between classified and digitised lupine maps was 5 %.