OneForest Analytics

“ShinyForest Analytics: A Containerized Web-Based System for the Automation of Near-Infrared Model Development in Wood Quality Traits of Forest Trees”

Near-infrared (NIR) data modeling involves complex multi-stage procedures throughout model calibration development. While NIR spectroscopy is becoming the standard in forestry, as it provides a cost-efficient assessment of wood quality traits, more sophisticated modeling algorithms now require extensive workflows to effectively implement model maintenance, update, and reproducibility. Here, we present ShinyForest: a web-based system for the automation of NIR model calibration that integrates data preparation, modeling, and prediction pipelines using parallelization and process isolation of Linux container technology. Three core procedures were implemented: i) data preparation - outlier detection and preprocessing methods, ii) model calibration - including wavelength selection in Partial Least Squares (PLS), Support Vector Machines (SVM), and Random Forest (RF) regression algorithms, and iii) prediction of new data - implementing model inference space and prediction uncertainty.