By dynamically adjusting sampling frequency based on uncertainty and prior information, this method cuts data acquisition by up to 80% without compromising essential biological insights. Among five tested Bayesian techniques, the Markov Chain Monte Carlo (MCMC) and Gaussian Process (GP) approaches demonstrated the best balance between compression, precision, and computational cost, marking a major step toward efficient, real-time phenotyping.
Advances in plant imaging and computer vision have transformed agriculture and biology by enabling continuous and objective trait quantification. However, monitoring large plant populations or long-term processes—such as germination and growth—creates vast data streams that are costly to produce, process, and store. Fixed-rate sampling, though simple, often results in redundant data because plant growth is nonlinear and varies across species and environments. Adaptive temporal sampling offers a solution by dynamically adjusting measurement timing according to the biological process being observed. Based on these challenges, the research team developed a Bayesian adaptive sampling framework to optimize data collection for non-linear plant growth processes.
A study (DOI: 10.1016/j.plaphe.2025.100067) published in Plant Phenomics on 21 June 2025 by David Rousseau’s team, Université d’Angers, presents a practical solution for high-throughput plant imaging and monitoring tasks, enabling researchers to reduce costs associated with data production, storage, and analysis.
The study employed five Bayesian adaptive sampling techniques—Importance Sampling (IS), Markov Chain Monte Carlo (MCMC), Gaussian Process (GP), Extended Kalman Filter (EKF), and Sequential Importance Resampling Particle Filter (SIR-PF)—to evaluate their efficiency in monitoring seed germination kinetics through time-lapse imaging. Each method dynamically adjusted sampling frequency based on prior data and model uncertainty, allowing the researchers to assess performance across three key dimensions: compression-distortion trade-off, robustness to variations in germination speed, and computational cost. The compression-distortion analysis demonstrated that adaptive sampling could drastically reduce data volume by 80%, achieving a compression ratio of 0.2 while preserving accuracy. Among the tested methods, IS, MCMC, and GP exhibited the lowest distortion for both simulated and real datasets. In robustness tests across fast, normal, and slow germination scenarios, MCMC consistently produced the lowest mean square error (MSE) and global bias, showing strong adaptability to variable biological conditions. GP also performed reliably, offering unbiased parameter estimation even when germination speeds differed from prior expectations. In terms of computation time, EKF was the fastest at 0.02 milliseconds per estimation, while MCMC, though the slowest at 2.6 seconds, maintained computational feasibility for real-time biological monitoring. Analysis of the adaptive threshold σT revealed that while GP and EKF allowed precise control over the number of samples, distortion could not be perfectly constrained, indicating a need for further refinement. Overall, the Bayesian adaptive sampling framework provides an operational, cost-efficient solution for continuous plant monitoring and can be extended to other dynamic biological processes, such as circadian leaf cycles or pathogen spread, with future improvements potentially achieved through advanced non-linear filtering techniques like the Unscented Kalman Filter.
Beyond seed germination, the methodology can be extended to a wide range of time-dependent biological phenomena such as circadian leaf movement, seedling emergence, and disease progression. Its compatibility with non-linear models and low computational requirements make it well-suited for integration with Internet of Things (IoT) systems and automated phenotyping platforms. By optimizing when and how data are collected, this approach supports sustainable, efficient, and scalable agricultural monitoring strategies for the digital farming era.
###
References
DOI
10.1016/j.plaphe.2025.100067
Original URL
https://doi.org/10.1016/j.plaphe.2025.100067
Funding information
This research was funded by La Région des Pays de la Loire under the TANDEM program.
About Plant Phenomics
Plant Phenomics is dedicated to publishing novel research that will advance all aspects of plant phenotyping from the cell to the plant population levels using innovative combinations of sensor systems and data analytics. Plant Phenomics aims also to connect phenomics to other science domains, such as genomics, genetics, physiology, molecular biology, bioinformatics, statistics, mathematics, and computer sciences. Plant Phenomics should thus contribute to advance plant sciences and agriculture/forestry/horticulture by addressing key scientific challenges in the area of plant phenomics.