Multimodal Sensor Fusion Dataset for Early Water Stress Detection in Banana: Integrated Soil, Morphological, and UAV Canopy Features Under Four Irrigation Regimes, Magdalena, Colombia (2024)

Description

Longitudinal operational dataset derived from the banana irrigation deficit experiment (Musa AAA, cv. Williams), structured for training and validation of machine learning models for early water stress detection. The dataset contains 3,380 weekly plant-level observations (130 plants × 26 weeks) across 85 variables integrating four sensing layers: Soil variables: calibrated VWC (%), soil water depletion fraction, lagged variables (lag1, lag2), rolling means (roll3), and stress-threshold indices at multiple depletion levels (p35, p45, p55) Morphological variables: pseudostem height and functional leaf count, linearly interpolated between measurement dates UAV canopy variables: canopy metric (CM) with inter-flight imputation Meteorological and irrigation variables: weekly ET₀, VPD, and treatment-specific applied irrigation depth Target variable: water stress level label by treatment (T1–T4) Designed for Leave-Treatment-Out Cross-Validation (LTOCV) schemes and multi-class classification models.

Institutions

• University of Magdalena

  Magdalena Department, Santa Marta
• Pontificia Universidad Javeriana

  Bogota D.C., Bogotá

Categories

Licence