Variable endmember framework improves crop FAPAR estimation from UAV imagery
image: The field images and typical object spectra of (A) sorghum and (B) rice. The colored bars are the spectral variation ranges. Note: The in situ spectra were measured from the crop fields using an ASD Field Spec 4 spectrometer (Analytical Spectral Devices Inc., Boulder, CO, United States).
Credit: Plant Phenomics
Estimating the fraction of absorbed photosynthetically active radiation (FAPAR) is essential for tracking crop photosynthesis, growth and yield potential. Unmanned aerial vehicle (UAV) multispectral imagery offers an efficient way to map FAPAR, but vegetation index (VI)-based methods often struggle with mixed pixels, multiple scattering, spectral variability and dynamic canopy–background interactions. This study introduces VE-MLM, a variable endmember-based multilinear mixing framework for FAPAR estimation. The framework builds adaptive foreground and background spectral libraries, iteratively unmixes each pixel with a multilinear mixing model (MLM), and selects the best endmember combination using root mean square error (RMSE). Tests on sorghum and rice show that the derived canopy abundance is strongly correlated with field-measured FAPAR, offering a robust and generalizable approach for precision agriculture.
Traditional fraction of absorbed photosynthetically active radiation (FAPAR) retrieval uses either complex physical models or simple empirical relationships with vegetation index (VIs). Physical models require many parameters, while empirical VIs are easily affected by background signals, canopy structure and saturation. Spectral mixture analysis (SMA) can separate canopy and background contributions, but fixed endmembers and linear mixing models (LMMs) rarely capture real crop fields. Spectral variability, insufficient pure pixels and multiple scattering among leaves, stems, panicles, soil and water further reduce accuracy. Although multiple endmember SMA and nonlinear bilinear models have been developed, they still face challenges in representing dynamic, multilayered crop scenes. Given these challenges, in-depth research is needed to develop adaptive unmixing methods that account for endmember variability and complex light propagation in crop canopies.
Researchers from Wuhan University and collaborators in China and Hong Kong SAR developed the variable endmember-based multilinear mixing framework (VE-MLM) framework. The study was published (DOI: 10.1016/j.plapbe.2026.100202) in Plant Phenomics (2026). Using unmanned aerial vehicle (UAV) multispectral images and field FAPAR measurements from sorghum and rice across multiple growth stages, the team tested whether adaptive spectral unmixing could estimate crop FAPAR more accurately than conventional VIs and machine learning models. It aimed to improve FAPAR estimation under spectral variability and complex canopy–background mixing.
The VE-MLM framework has three modules. First, variable endmember extraction builds a spectral library from close-range UAV images by identifying pure foreground and background pixels in red–near-infrared (R–NIR) feature space, then reducing redundancy with k-means and an iterative endmember selection (IES) algorithm. Second, iterative unmixing pairs every foreground endmember with every background endmember and applies the MLM pixel by pixel, accounting for multiple scattering and transmission. Third, optimal selection chooses the combination with the lowest RMSE and outputs foreground abundance (Af). The adaptive library captured spectral variability across growth stages, while the MLM estimated a recollision probability for each pixel, helping describe light interactions within multilayered canopies. Compared with fixed endmembers and linear or bilinear models, VE-MLM achieved the lowest relative error (RE) and spectral angle distance (SAD), with stable performance across sorghum and rice. The resulting Af explained FAPAR with coefficient of determination (R²) = 0.900 and relative RMSE (rRMSE) = 7.753% for sorghum, and R² = 0.807 and rRMSE = 2.200% for rice, outperforming the best VIs and random forest (RF) models. It also remained consistent when red or yellow panicles appeared at maturity.
The authors said the advance lies in letting the crop scene define its own spectral endmembers instead of forcing one fixed set onto every image. By combining that adaptive library with a multilinear mixing model, they said, VE-MLM better follows how light repeatedly interacts with leaves, stems, panicles, soil and water. They added that the framework reduced saturation and background interference that commonly limit VI-based FAPAR estimates, and that the canopy abundance it produces has a clear physical link to the probability of light interception by crop canopies.
The approach could support UAV-based crop phenotyping, variety selection, nitrogen management and growth monitoring, where accurate FAPAR estimates help breeders and farmers assess photosynthetic capacity. Because VE-MLM uses canopy abundance rather than a fixed VI, it may adapt better to different crops, growth stages and field backgrounds. It may also improve yield prediction and climate-resilience assessments by providing a more physical FAPAR input. The authors caution that the current workflow relies on high-resolution UAV imagery for pure-pixel endmember extraction and involves pixel-by-pixel computation. Future work will optimize endmember selection, test satellite–UAV synergy, and validate the framework across more crops and environments.
###
References
DOI
Original Source URL
https://doi.org/10.1016/j.plaphe.2026.100202
Funding information
This work was supported by Kweichow Moutai Co., Ltd. (university-enterprise cooperation project: MTGF2023047, MTGF2023048); the Department of Natural Resources of Hubei Province (provincial-ministry collaborative pilot project: 2023ZRBSHZ025); the State Administration of Science, Technology and Industry for National Defense (advanced research project: Remote sensing inversion of key land surface elements and development of quantitative products); and the Basic Surveying and Mapping Information Center of Anhui Province (provincial-level natural resources science and technology project: 2023-K-2).
About Plant Phenomics
Plant Phenomics publishes breakthrough research that integrates genomics, genetics, physiology, molecular biology, bioinformatics, statistics, mathematics and computer science to advance plant phenotyping—from the cellular level to entire populations. The journal tackles key scientific challenges in phenomics by showcasing cutting‑edge technologies for data acquisition, management, interpretation and modelling. Its scope spans practical applications across all plant sciences, aiming to translate novel phenotyping tools into real‑world solutions for breeding, crop management and fundamental plant biology. By fostering interdisciplinary innovation, Plant Phenomics serves as a hub for researchers developing high‑throughput, accurate and scalable methods to decipher complex plant traits, ultimately accelerating progress in agriculture and plant science.