Hierarchical Bayesian Modeling for Flexible Experiments in Decentralized Participatory Plant Breeding

Abstract

ABSTRACTDecentralized participatory plant breeding (PPB) is a method for creating new varieties adapted to agroecology based systems. Selection is decentralized in the target environments and relies on collaboration between farmers, Non Governmental Organizations (NGOs), and researchers. In the wheat PPB project studied, the decentralization was extensive as each farmer participating in the project performed the selection experiment on his farm. This led to unbalanced designs with few residual degrees of freedom available for the within‐farm comparison of means. We investigated a hierarchical Bayesian modeling of trials’ residual variances using information from a large network of environments (48 farm × year). We found that the hierarchical approach allowed for robust results leading to reliable mean comparisons on farms. For each variable studied, differences among breeding populations within farms were large enough to carry out selection. This was possible because the farms shared a common experimental design and the number of environments in the network was large. Our approach is useful for farmers who cannot set up replicated trials on their farm but wish to participate in the program. The only constraint being the replication of at least one control on each farm, farmers are free to choose the populations they wish to test. This allows for the evaluation and selection of a wide range of diversity over the network of farms. This approach could also be interesting in other types of decentralized unreplicated trials such as those encountered in genetic resources screening or in multiple environments breeding.