Tree-level multi-source sensing and average fruit weight data for Hass avocado

Description

This dataset contains tree-level multi-source sensing data collected from 30 Hass avocado trees during a single production cycle in a commercial orchard in Antioquia, Colombia. The data include soil, microclimatic, physiological, sap, phenological, and harvest-related measurements aggregated into seven biologically meaningful phenological phases. The response variable used for modelling is average fruit weight at the main December harvest, expressed in g fruit\(^{-1}\). The dataset supports the analyses reported in the associated manuscript on phenology-structured prediction of average fruit weight using supervised feature filtering, ElasticNet regression, and nested cross-validation under small-sample conditions.

Steps to reproduce

The dataset can be used to reproduce the analyses reported in the associated manuscript together with the publicly available source code at https://zoco.ugr.es:3000/aparra/pasantia_ugr. Raw and processed measurements were harmonized at the individual-tree level and aggregated into seven phenological phases. The modelling workflow includes data preprocessing, phenology-based feature engineering, supervised feature filtering, ElasticNet modelling, strictly nested cross-validation, predictor-stability analysis, sensitivity analysis across random seeds, and phenological ablation. The scripts should be executed following the workflow described in the repository README. All supervised feature-selection procedures are fitted exclusively within the corresponding training partitions of the nested cross-validation procedure.

Institutions

• Universidad de Antioquia

  Antioquia, Medellín

Categories

Funders

• Colombian General System of Royalties (Sistema General de Regalías, SGR)

  Grant ID: BPIN 2022000100012

Licence