Initial evaluation of models to predict methane production based on VIS-NIR signatures captured in the field

Characterizing the genotypes conserved in a Genebank is crucial to promoting their use in crop breeding programs or incorporating outstanding varieties directly into agri-food systems. Since the 1970s, the forage collection of the Alliance Bioversity International & CIAT has conserved approximately 22,000 accessions of tropical forages across 690 species. These are primarily legume species, with 93% of the accessions belonging to the Fabaceae family and 7% to the Poaceae family. Traditionally, genebank characterizations focus on capturing basic morphological and phenological traits to support genetic quality evaluations and inform decisions on regeneration and seed production. However, due to the high demand for specialized personnel, these characterizations often do not explore traits related to productivity, nutrition, or environmental impact. High-throughput phenotyping is increasingly robust, featuring more accurate sensors and models to assess productivity traits. For example, for forages, drone technology is being refined to measure plant height and biomass accumulation, while near-infrared reflectance (NIR) spectral signature analysis is widely used to quantify forage nutritional quality, significantly reducing the costs of nutritional assessments, which are traditionally expensive in laboratory settings. In this study, 300 forage accessions sown in the first half of 2024 are being evaluated under the environmental conditions of Palmira, at the Alliance Bioversity International & CIAT campus. These accessions correspond to legume species adapted to tropical lowland conditions, characterized by their tolerance to low soil fertility and high aluminum levels, common challenges in many tropical regions. The study involves capturing hyperspectral signatures using the ASD QualitySpec, a portable spectrometer that facilitates field-based measurements. These measurements are taken when the plants reach at least 20% flowering, ensuring the representativeness of nutritional and functional data. The spectral data are analyzed using a Random Forest artificial intelligence model, which identifies complex patterns and determines the spectral regions that contribute most to the nutritional variability of the samples. Additionally, this study explores the correlation between spectral signatures and the mitigation of methane emissions measured under in vitro conditions, leveraging known regions of the NIR spectrum associated with anti-methanogenic compounds such as tannins, flavonoids, and saponins. Models are also employed to assess the functional diversity of the accessions, applying hierarchical clustering to reveal similarities between accessions based on key spectral regions. This approach establishes functional relationships, evaluates environmental benefits, and identifies the potential of each accession as a promising forage resource. The importance of this method lies in its ability to identify nutritional attributes and anti-methanogenic properties directly in the field with reliable validations. This is particularly advantageous in remote regions lacking access to quality animal nutrition laboratories. Such advancements are instrumental in supporting the selection of promising forages, fostering the development of sustainable production systems in tropical regions, and enhancing climate resilient livestock farming under challenging conditions while mitigating greenhouse gas emissions.

Gonzalez Guzman, J.J.; Penafiel, J.; Cardoso, J.A.; Lopez, D.C.; Jones, C.; Jauregui, R.; Sanchez, M.; Marin, A.; Wenzl, P.; Arango, J.