Dataset for phenology-aware UAV multispectral inversion of rice physiological traits using group-aware machine learning

This dataset supports the manuscript “Phenology-aware UAV multispectral inversion of rice physiological traits using group-aware machine learning.” Dataset overview The dataset contains 1,290 paired UAV–ground observations collected from a 51.2-ha irrigated lowland rice landscape in the Plaosan agricultural area, Tlogoadi Village, Mlati District, Sleman Regency, Special Region of Yogyakarta, Indonesia. Data collection was conducted during three UAV and field-measurement campaigns on 14 February, 24 February, and 4 March 2026. Multispectral imagery was acquired using a fixed-wing vertical take-off and landing Trinity Pro UAV equipped with a MicaSense RedEdge-P sensor. The sensor acquired blue, green, red, red-edge, and near-infrared reflectance. Ground physiological measurements were collected using SPAD-502 Plus and LI-600 instruments. Dataset contents The curated tabular dataset includes: 1. Observation, flight, sampling-unit, and validation-group identifiers.2. UAV acquisition dates and days after planting.3. Rice phase and estimated rice growth-stage labels aligned with the BBCH rice phenological concept.4. Statistical descriptors of blue, green, red, red-edge, and near-infrared reflectance.5. Statistical descriptors of multispectral vegetation indices, including NDVI, GNDVI, NDRE, EVI2, SAVI, OSAVI, CCCI, MTCI, ARVI, and TCARI.6. Ground-measured physiological traits comprising SPAD, stomatal conductance (gsw), electron transport rate (ETR), effective quantum yield of photosystem II (PhiPS2), leaf temperature (Tleaf), and leaf vapor pressure deficit (VPDleaf).7. Variables required for group-aware regression, phenological classification, residual analysis, and predictor-importance assessment. Phenological coverage The observations represent vegetative, reproductive, and ripening rice phases. The estimated growth stages comprise tillering, stem elongation/panicle initiation, booting, heading/flowering, grain filling, and maturity. The maturity stage contains a relatively small number of observations and should therefore be interpreted cautiously in stage-specific analyses. Quality control Target-specific quality control was applied. Observations with missing or invalid values for one physiological trait were excluded only from the corresponding trait analysis and remained available for other valid targets. The numbers of valid observations were 1,289 for SPAD, 1,280 for gsw, 1,288 for ETR, 1,288 for PhiPS2, and 1,290 for both Tleaf and VPDleaf. Recommended analytical use The dataset was developed for group-aware machine-learning analysis. Sampling-unit or field-group identifiers should be kept entirely within either the training or testing subset to reduce information leakage from spatially related observations. Random row-wise splitting may produce overly optimistic performance estimates. Data scope and limitations This repository contains the curated tabular UAV–ground dataset. It does not contain the original raw UAV images or complete multispectral orthomosaics. The dataset represents one rice-growing landscape, one seasonal window, and the sensor and field conditions described above. External validation is recommended before transferring models to other regions, seasons, cultivars, management practices, or stress conditions. License The dataset is released under the Creative Commons Attribution 4.0 International license. Users may share and adapt the data provided that appropriate credit is given to the creators and the Zenodo dataset DOI is cited.