sekika/tffit: v1.1.0

tffit is a Python software package for fitting and evaluating soil-to-wheat radiocesium (137-Cs) transfer factor (TF) models. It is designed to support transparent and reproducible analysis of TF relationships between soil properties and crop uptake, as presented in the following published journal article:

• Seki, K., Yamaguchi, N., Eguchi, T., Igura, M., 2026. Improvement of spatiotemporal generalization in radiocesium transfer models for wheat using symbolic regression. Journal of Environmental Radioactivity 298, 108077. https://doi.org/10.1016/j.jenvrad.2026.108077

The primary purpose of this software is to enable full reproducibility of the model calibration, validation, and supplementary uncertainty analyses reported in the study, using the dataset provided in the Supplementary Material of the paper. In addition, the codebase is structured to facilitate reuse in related studies, allowing users to apply, compare, or extend TF models for different datasets or experimental conditions.

Key features include:

• Implementation of multiple published and derived TF models based on soil chemical properties, including exchangeable K, RIP, CEC, pH, and radiocesium concentration
• Model fitting on log10-transformed transfer factors
• Automated reproduction of fitted model coefficients and in-sample statistics reported in the associated study
• Rigorous external cross-validation frameworks:
  • Leave-One-Site-Out (LOSO) cross-validation for spatial generalization
  • Leave-One-Year-Out (LOYO) cross-validation for temporal generalization, including per-year error metrics
• Supplementary uncertainty and comparison analyses:
  • Bootstrap confidence intervals for model coefficients
  • Paired comparisons of out-of-fold prediction errors between models
  • Cluster-level loss difference summaries for LOSO and LOYO validations
• Automated generation of results corresponding to the main and supplementary tables of the accompanying study
• Robust input validation to ensure consistency of log-scale transformations
• Support for text, CSV, and Markdown table output
• Generation of observed vs. predicted scatter plots for model performance visualization
• Modular and extensible design for implementing new TF models

This repository contains code only. The dataset used in the study is not included and must be obtained from the Supplementary Material of the accompanying publication.

For detailed usage instructions, input data requirements, command-line options, and examples, please refer to the GitHub repository:

https://github.com/sekika/tffit