A Network-Based Method to Detect Patterns of Local Crop Biodiversity

Abstract

In this paper we develop new indicators and statistical tests to characterize patterns of crop diversity at local scales. Households growing a large number of species or landraces are known to contribute an important share of local available diversity of both rare and common plants but the role of households with low diversity remain little understood: do they grow only common varieties-following a nestedness pattern typical of mutualistic networks in ecology-or do 'diversity poor' households also grow rare varieties? This question is pivotal in ongoing efforts to assess the contribution of small farmers to global agrobiodiversity at local scales. We develop new network-based approaches to characterize the distribution of local crop diversity at the village level (species and infra-species) and validate these approaches using meta-data sets from 10 countries. Our results highlight the sources of heterogeneity in the local crop diversity. We often identify two or more groups of households based on their different levels of diversity. In some datasets, 'diversity poor' households significantly contribute to the local crop diversity. Generally, we find that the distribution of crop diversity is more heterogeneous at the species than at the infra-species level. This analysis reveals the absence of a general pattern of crop diversity distribution independent of agro-ecological and socio-cultural context.

Urbanization drives an early spring for plants but not for pollinators

“…To analyse the overall homogeneity of flowering plant species composition among sites, we used latent block models (LBMs) based on presence–absence data and Bernoulli distribution (R package blockmodels ; INRA and Leger 2015). LBMs are probability‐based models that simultaneously cluster sites (rows) and species (columns) based on latent blocks (Govaert and Nadif 2008, Keribin et al 2015, Thomas et al 2015). An incidence matrix was generated to highlight homogeneous blocks of sites and species (Supplementary material Appendix 3 Fig.…”

Block models for multipartite networks.Applications in ecology and ethnobiology

“…• Example 2 is issued from Thomas and Caillon [2016] and Thomas et al [2015]. In this dataset, we observe on the one hand, seed circulation between farmers -resulting in a non-symmetric adjacency matrix -and on the other hand the crop species grown by the farmers, resulting in an incidence matrix.…”

“…These inventory data can be seen as a bipartite network. The seed circulation data were analyzed in Thomas and Caillon [2016] and the inventory data were analyzed in a meta-analysis in Thomas et al [2015]. On the basis on the joint modeling we propose in this paper, we aim to provide a clustering on farmers and crop species on the basis of the seed circulation network and the inventory bipartite network.…”

Reconstructing undirected graphs from eigenspaces

“…Networks have become a natural and popular way to model interactions in applications such as information technology (Rossi and Latouche, 2013), social life (Jiang et al, 2013;Matias et al, 2015), genetics (Giraud et al, 2012), ecology (Thomas et al, 2015;Miele and Matias, 2017). In this paper, we investigate the reconstruction of an undirected weighted graph of size N from incomplete information on its set of edges (for instance, one knows that the target graph has no self-loops) and an estimation of the eigenspaces of its adjacency matrix W. This situation depicts any model where one knows in advance a linear operator K that commutes with W.…”