Abstract
The process of abscission - the shedding of parts of an organism - is closely linked to a series of physiological events whose optimal execution is key to the survival of the species. Environmental variations impact species development, and in particular abscission processes at multiple developmental stages. Identifying environmental factors and the times at which they modulate the abscission process is crucial, particularly in the context of climate change. Considering environmental variables as time series, i.e. groups of variables correlated over time, poses statistical problems for selecting the relevant groups, the environmental variables, and the variables correlated within them, the time periods. In this paper, we address these objectives by introducing and discussing a set of Bayesian fused and fusion priors via a general parameterization. This paper highlights a trade-off between the priors used on differences and coefficients. In particular, we show the effectiveness of horseshoe-type priors on differences as well as on coefficients with appropriate parameterizations in terms of selection, estimation and algorithmic stability whatever the number of groups and their size. The study was motivated by fruit abscission in oil palms, which impacts the bunch harvesting timing. Disturbance of abscission can therefore affect oil yield and quality, and consequently have an impact on economic income. This application, which is based on an experimental design in Benin Republic, illustrates the performance of the proposed prior in selecting both environmental variables and the developmental stages involved in the timing of bunch harvesting.