< Back to previous page

Publication

High-throughput field phenotyping using a drone with RGB imagery. Exploiting the spectral, spatial and temporal dimensions

Book - Dissertation

Altered climatic conditions leading to more frequent and intense abiotic events (e.g. drought or heatwaves) and biotic stresses (pathogens and pests) are going to impact on our food system endangering the access to safe, nutritious and sufficient food. To mitigate the negative effects, the adaptation of crops to future climate has to go also hand in hand with an increase in global crop production in order to achieve the estimated demand for the growing world population. Furthermore, this should be achieved requiring less inputs and increasing efficiency of resources used (i.e. more sustainable production). Different approaches can be taken to accelerate adaptation and increase crop production. This can be done by further optimization of crop management and adjustments of the inputs (precise fertilization, irrigation, …), but especially by the development of improved crop varieties through breeding. Breeding can highly contribute to secure agricultural productivity in this changing environment. However, the current process to adapt a crop to the new growing conditions requires at least a decade of experimental work. The different steps involved in the breeding process are based on the evaluation and selection of genotypes by highly labor intensive measurements and visual assessments of plant characteristics. In order to speed up and improve the breeding process, new technologies are required in different disciplines. In this dissertation the focus is on plant phenotyping, the process of assessing the performance of a genotype for a particular environment. For this purpose, breeding field trials are established to monitor and evaluate the plant performance in the target growth environment. In those trials, a large number of genotypes are established in small plots and a selection of the best and more appropriate ones (to the targeted environment) is carried out based on desirable characteristics/traits (e.g. drought resistance, yield, …). Traits are usually assessed visually, by field measurements or destructive sampling. These methods entail limitations: they are to a certain extent subjective and the measurements take a lot of time. How crops are phenotyped determines the efficiency of the breeding program and the resulting genetic gain. This genetic gain can be improved by upscaling and increasing the accuracy and frequency of the phenotypic evaluation. The breeding process can be accelerated via high-throughput field phenotyping using a sensor mounted on a drone. This allows to collect more objective phenotypic data. Drones allow to fly over the crop at crucial moments in crop development, as frequent as it is needed. Imagery is captured in a short period of time and therefore under the same conditions. To promote integration in breeding programs, the technology should provide high resolution data at an acceptable level of cost and must be user-friendly. Therefore, this PhD research focused on the use of a drone equipped with an RGB camera which can capture imagery at a high resolution. From the imagery, vegetation indices and canopy height models were derived, and variability within and between plots was assessed. The potential of this technology was investigated in two model crops, perennial ryegrass and soybean. Both species have economic and ecological importance, but differ in morphology, growth, management, breeding process and final use. Perennial ryegrass is a fodder crop, monocot, and cultivated for the biomass and is harvested several times during one growing season. Soybean is a grain legume, dicot and grown for the seeds. During the PhD we developed procedures and techniques that can easily be integrated and applied in breeding programs. These techniques were thus tested and validated in running breeding trials. Important breeding traits tackled for perennial ryegrass are yield and persistency. Persistency is a crucial trait to ensure productivity along the years a pasture is exploited. This trait is traditionally assessed by visual scoring. This trait is not easy to evaluate and the assessment is not free from subjectivity. Therefore, as an alternative, we have developed a methodology based on vegetation indices (spectral information) derived from drone-based RGB imagery. Different vegetation indices were tested under different light conditions, revealing that it is possible to evaluate persistency of hundreds of plots with UAV imagery in a more consistent way and achieving a better differentiation between genotypes than visual scoring. Biomass yield, another important trait, is typically monitored by destructive sampling which is laborious and time-consuming. A rising plate meter can be used as a non-destructive method for intermediate measurements. The measurements are influenced by canopy height and density, but are limited in evaluating variability within a plot. In this dissertation, we present a non-destructive method capable of capturing field variability and temporal evolution of crop height used to estimate yield. We tested different models for dry matter yield estimation and all models based on drone data resulted in a lower RMSE compared to a rising plate meter. The best one achieved an improvement from an RMSE of 986 (rising plate meter) to 679 kg/ha (drone). In the case of soybean, the main breeding traits are related to seed yield, maturity (ripening of the seeds), growth and development. Multi-temporal canopy cover and height data can provide information about crop growth and performance. By fitting growth curves per plot/genotype to the multi-temporal data, relevant phenotypic characteristics (e.g. canopy closure, growth rate, senescence, …) were derived. Differences between genotypes were depicted more clearly. Based on the phenotypic characteristics seed yield and maturity were estimated by means of a multiple linear regression (adjusted r2 = 0.49 and 0.80, respectively). This approach resulted in biologically interpretable parameters that were informative for relevant traits. In conclusion, our results demonstrate the potential of RGB imagery captured with a drone in a breeding context, enabling to obtain objective high-throughput phenotypic data from large field trials with high spatial and temporal resolution. The methodologies developed in this dissertation can support or replace traditional breeding methods.
Publication year:2020
Accessibility:Open