FitEval-Based Evaluation of Sediment Particle Size (d50) Prediction Models and QnD Savanna Performance Across Four Ecozones

Abstract

Pseudomonas fluorescens is one type of Pseudomonas bacteria that is good for plants. This study presents a comprehensive evaluation of two environmental modeling tasks using the FitEval software: the QnD: Savanna ecological model and sediment particle size (d50) prediction models. The QnD: Savanna model was assessed across four ecozones (Le, Ph, Sa, Sk) under scenarios with and without observational error. Results showed strong performance in Le, moderate in Sa, and unsatisfactory outcomes in Sk and Ph, with statistical significance often lacking. For sediment modeling, four approaches (XGBoost, Random Forest, a new statistical model, and the Foster model) were evaluated. XGBoost demonstrated superior predictive accuracy and robustness, while the statistical model showed potential for exploratory use. Random Forest and Foster models were found inadequate. FitEval’s bootstrapping-based framework enabled uncertainty quantification and significance testing, revealing that model reliability depends not only on performance metrics but also on structural soundness and intended application. The findings emphasize the importance of integrating statistical rigor, uncertainty analysis, and diagnostic tools in environmental model evaluation.