Differences in growth features between species are driving cereal-legume intercrop yield: a statistical learning approach based on aggregated dataset

Abstract

Increasing crop diversification is crucial for developing more sustainable agricultural systems, and cereal-legume intercropping is a promising strategy. This study investigates the factors influencing the yield of cereal-legume intercrops using data from six field experiments in southwestern France, where durum wheat was intercropped with either faba bean or pea. We assessed how differences in plant traits between the associated species (e.g., height or biomass growth rates) are related to the intercrop productivity. Additionally, we developed a novel modeling approach, combining machine learning and mixed-effects models, to identify the key traits driving intercrop performance based on variable importance.

Our results show that interspecific differences in plant traits, particularly in biomass accumu-lation rate, maximum leaf area index, and elongation rate, were the most important factors explaining intercrop yield. These traits and their differences mainly suggest that compet-itive processes shape the outcome of a mixture and highlight the importance of dynamic measurements in agronomic experiments. The relationship between species yield and trait differences was symmetric for both intercropped species. Furthermore, these relationships were scale-dependent, with trends observed at the aggregate level not always consistent at the level of individual experiments.

Our study highlights the importance of considering trade-offs when designing intercropping systems for practical applications and demonstrates the value of combining machine learning with ecological knowledge to gain insights into complex agricultural systems from aggregated datasets.