Predicting grapevine harvest variables using proximal ancillary data sources: application of a multivariate multi-block modelling approach
Résumé
Increased access to multi-temporal site-specific spatial agricultural datasets provides a large amount of heterogeneous information that can be structured into different information blocks. Time-series of spectral canopy data acquisitions from a multispectral canopy proximal sensor, apparent soil electrical conductivity and measurements characterising four crop attributes (yield, berry weight, pruning mass and total soluble berry solids) were obtained at 321 sample locations across a 2.6 ha block of Concord vineyards in the USA in 2020 and 2021. A multi-block classification approach adapted from the chemometrics literature is first presented for precision agriculture use and then was used to predict each harvest attribute from the multi-temporal and multivariate canopy and soil datasets. The multi-block approach achieved good predictive quality despite the high intra-block production variability observed.