Speaker
Description
Crop diversification is increasingly promoted as a mean to improve the sustainability of agriculture while maintaining a sufficient level of food production [1]. Intercropping is a farming practice that combines at least two crop species in the same field for most of their growing periods [2], and where the two components are harvested and eventually sorted. Since plants vary in their ecophysiological functioning, a mixture of different crop species could improve the resource use efficiency, relative to its component species grown separately in sole crop. Mixtures between cereal and legume species are a prime example, on one hand because of their complementarity in nitrogen use [3], on the other hand because of its technical operationality [4].
Reports from meta-analysis [5] overall indicate that intercropping is a candidate practice for sustainability, based on its increased productivity per unit area. However, reports based on the outcomes of field experiments [6] indicate that these benefits are largely context-dependent [7]. We argue that shifting from a broad logic linking diversity to productivity to an understanding of how species features impact crop productivity is key to developing intercropping practices. More precisely, we propose to assess how the difference between associated species features are linked to the mixture productivity.
For that, we gathered a set of 37 field experiments in Europe [8] containing species-dependent measurements along with mixture and sole crop productivity. We first characterized each species by summarizing time series for plant growth (leaf area, height, biomass) into key features such as maximum value, rate or delay in growth. We then quantified plant-plant interactions with indicators based on differences between these features (relative trait distances in community ecology). We also characterized the cropping environment using climate and soil variables, including an indicator related to the plant nitrogen status (nitrogen nutrition index NNI, [9]).
We modeled the mixture performance as a function of environment and plant-plant interactions using a mixed-effect random forest (MERF, [10]), which is a method combining a machine learning and a mixed effect model. We used two steps to interpret our model and identify which features were important for the mixture performance. We first reduced the set of features to focus on using a stringent variable selection process during model fitting [11] and then ranked the selected variables by estimating their contribution to the variance of the mixture performance (variable importance).
We showed that features related to inter-specific plant interactions (differences between species within the mixture) were more selected and important than the ones related to species response to mixture (differences between managements). The selected features thus mainly indicated that competitive processes shape the outcome of a mixture. Among them, the difference in biomass accumulation rate, representing the strength of the competition, was consistently more important than other features. Overall, the yield of a species in mixture was positively correlated to its dominance, here captured by its stature (height, leaf area, biomass), at the expense of the associated species. Features related to the climate and to cultivars were not selected.
While we miss features related to below-ground interactions, our study contributes to understanding how management options, such as species choice or nitrogen management, have an impact on the mixture performance through their effect on the competitive strength of one given associated species.
References
[1] Beillouin D et al., 2019, https://doi.org/10.1088/1748-9326/ab4449.
[2] Willey RW, 1980, https://doi.org/10.1017/s0014479700010802.
[3] Landschoot S et al., 2024, https://doi.org/10.3389/fpls.2023.1228850.
[4] Verret V et al., 2020, https://doi.org/10.1016/j.eja.2020.126018.
[5] Martin-Guay M-O et al., 2018, https://doi.org/10.1016/j.scitotenv.2017.10.024.
[6] Jones SK et al., 2023, https://doi.org/10.1016/j.baae.2022.12.005.
[7] Duru M et. al, 2015, https://doi.org/10.1007/s13593-015-0306-1.
[8] Mahmoud R et al. 2024, https://doi.org/10.24072/pcjournal.389.
[9] Louarn G et al., 2021, https://doi.org/10.1016/j.eja.2021.126229.
[10] Hajjem A et al., 2012, https://doi.org/10.1080/00949655.2012.741599.
[11] Kursa MB et al., 2010, https://doi.org/10.18637/jss.v036.i11.
| Keywords | agronomy; ecology; machine learning; intercrop |
|---|
