Empowering growth: building the green Infrastructure for plant phenotyping

In recent years, technological progress has been made in plant phenomics (major improvements concerning imaging and sensor technologies). High-throughput plant phenotyping platforms now produce massive datasets involving millions of plant images concerning hundreds of different genotypes at different phenological stages in both field and controlled environments. Networks of sensors also measure environmental conditions in real time. The ongoing robotization of experimental processes foreshadows an explosion in the volume and complexity of the data produced by the different research facilities.

Various initiatives have helped to structure the European phenotyping landscape (EMPHASIS, EPPN) and enable researchers to use facilities, resources and services for plant phenotyping across Europe. However, among these services, the data services needed to be improved. Thus there was a need to build a federated and interoperable e-infrastructure allowing researchers to share and analyse phenotyping data.

The Scientific objectives were:

• analyse raw data (thousand of images) and extract variables of agronomic interest;
• combine data coming from several sources including large genetic and environmental variabilities and run meta analysis in order to predict the response of genotypes and species to current and future climate scenarios;
• structure the data (raw and processed data) in a way that it can be shared and reanalyzed by a wide scientific community: using a standardised, unique and unambiguous identification of objects involved in experiments, enriching datasets with knowledge and metadata enabling the reuse of data and meta-analyses;
• evaluate online deep learning and iterate with shared annotation and training/testing.

The Computing objectives were:

• Data management:

• Data integration between infrastructures/platforms (and other systems);
• Data management and integration of heterogeneous multi-dimensional data from multiple sources; 
• Data export to computing and modelling platforms;
• Dataset compatibility and interoperability between infrastructures;
• Production pipeline for creation of shared, and FAIR-compliant datasets
• Connection over BrAPI, https://brapi.org/

• Data analysis:

• access to a robust infrastructure for storage and data processing of large datasets;
• Integration of innovative computing services provided by the EOSC portal (including deep learning);
• Use cases of computer vision including all main informational tasks such as classification, object recognition, segmentation and regression.

• Data (or services) publishing:

• Increase visibility of our services by publishing them into the EOSC portal.
• for our published datasets, we are looking for a long-term dataset repository like dataverse.

The implementation of the data management part of this infrastructure is based on a storage layer based on the open-source Phenotyping Hybrid Information System PHIS (Neveu et al, 2019; www.phis.inrae.fr). PHIS is an ontology-driven Information System designed for Plant Phenomics, that allows to enrich datasets with knowledge and metadata enabling the reuse of data and meta-analyses. It interoperates and integrates data into external resources (e.g. modelling platforms or external databases) and provides FAIR data. PHIS is designed to store, organise and manage highly heterogeneous (e.g. images, spectra, growth curves) and multi-spatial and temporal scale data (leaf to canopy level) originating from multiple sources (field, greenhouse). PHIS has already been deployed in several French platforms for high throughput plant phenotyping including field and controlled conditions within the PHENOME-EMPHASIS project.