Food production must increase to keep up with population growth and become more sustainable to lessen the environmental footprint. There is a need for more productive wheat cultivars and site-specific crop management. The PhenoCrop project will tailor crop phenotyping technologies to Norwegian conditions and develop tools for use in plant breeding and precision farming. The project is coordinated by the Norwegian University of Life Sciences (NMBU) in close collaboration with plant breeders at Graminor, the Norwegian Agricultural Extension Service (NLR), the drone service provider Unmanned and selected wheat farmers. International collaboration is established with world-leading expertise on statistical modeling (CIMMYT), and application of deep learning for image analysis in plant breeding (University of Minnesota).
A robotic platform will be adapted for close-up imaging in field trials while drones fitted with multispectral and hyperspectral cameras will be used to gather crop canopy data in a fast and efficient manner. Phenotyping protocols will be developed and optimized for Norwegian field conditions, and specific algorithms applied to estimate important agronomic traits and predict yield performance.
In this project we make use of field trials both at Vollebekk research station in Ås (conducted by NMBU) and Staur research farm in Stange (conducted by Graminor). In the first field season we have performed weekly flights with multispectral camera at both sites to gather data that can be used to improve the yield prediction models that were developed in the vPheno project. In one of the field trials at Vollebekk we established extra field trial plots that were harvested just after anthesis (Zadoks 65) to develop models for prediction of biomass that can be used for estimating yield potential early in the season. In 2021 we have also focused on testing out flights with different cameras at different heights and different angles to optimize the data gathering for estimation of specific traits, with special focus on plant height and heading date.
At Vollebekk in 2021 we used our existing field phenotyping robot to gather high resolution close-up images from field trials that we contribute towards an international consortium (Global Wheat Head Challenge) for improving deep learning algorithms for automated recognition and counting of wheat heads. These models will be further tested on high resolution drone images taken at low flight altitude to see if we can use drone images for head detection. This is connected to our ambition of using image-based phenotyping to detect and quantify Fusarium head blight (FHB) infections in wheat, which is a particular challenging disease to evaluate by visual scoring.
In the spring of 2021 we took advantage of field trials of winter wheat with clear variability in winterkill to compare drone-based and visual assessment of winter survival. Not surprisingly, a simple index based on the proportion of vegetation pixels to total pixels in multispectral drone images turned out to be more reliable and less time-consuming than visual estimation. This side-work is now being written up for publication.
2021 was also a pilot season for testing out yield prediction in farmers? fields. We conducted drone imaging with multispectral camera in wheat fields at five selected farms in southeastern Norway. Data analysis is ongoing to compare the multispectral information with the actual yield variability captured by the yield mapper at the farmers? combine harvesters. A long-term goal for this work is to develop models for early estimation of yield variability that the farmer can use for site-specific fertilization and other precision technologies to achieve higher yield and quality stability.
Improved cultivars and crop management is needed to increase productivity and sustainability of agriculture. Advances in sensor technologies and genomics offer possibilities for faster breeding of new cultivars and more site-specific crop management. PhenoCrop seeks to increase the sustainability and productivity of Norwegian wheat production by developing new high-throughput phenotyping tools to be used in plant breeding and precision agriculture. A robotic platform will be adapted for close-up imaging in field trials while drones fitted with multispectral and hyperspectral cameras will be used to gather crop canopy data in a fast and efficient manner. Specific algorithms including machine learning models will be applied to estimate important agronomic traits and predict yield performance. Deep learning object detection methods will be applied to develop image-based automated solutions for Fusarium head blight (FHB), which is a particular challenging disease to evaluate by visual scoring. A database system with user-friendly tools for efficient data management and visualization will be developed. Big data in breeding and agriculture will be utilized by applying novel concepts of combining crop performance in large-scale plant breeding nurseries across years and locations with yield mapping in farmers’ fields coupled with multispectral drone imaging data. This will help us to disentangle the genotype-by-environment interactions, develop better tools for predicting performance of new cultivars and helping farmers to apply site-specific crop management to increase yield stability, productivity and profitability. The work will be carried on in close cooperation with plant breeders (Graminor), farmers, extension service (NLR), and drone service providers (Unmanned), and in collaboration with world-leading expertise on statistical modeling (CIMMYT), and application of deep learning for image analysis in plant breeding (University of Minnesota).