FFRK: Focal-Feature Regression Kriging

This repository contains the implementation of FFRK (Focal-Feature Regression Kriging), baseline/comparison methods, experiment data, paper figures, and reproducible results.

1. Repository Structure

.
├── Code/
│   ├── FFRK/                         # Core FFRK R package
│   ├── Fig6_Evaluation_14models/     # 14-model benchmark (Fig. 6)
│   ├── FFRK_Ablation/                # FFRK ablation experiments
│   ├── FFRKvsFFR/                    # FFRK vs FFR comparison
│   ├── Fig7&8_Sensitivity_KQ/        # Sensitivity analysis of k and q (Fig. 7-8)
│   ├── Tab2_Evaluiation_HIER/        # Table 2 (HIER) evaluation
│   ├── Tab2_Evaluiation_STK/         # Table 2 (STK) evaluation
│   ├── FigS1-S3_Scatter/             # Supplementary scatter plots (S1-S3)
│   ├── Figx_Variograms/              # Variogram analysis
│   ├── Figx_Variance/                # Kriging variance analysis and visualization
│   ├── Fig5_Global/                  # Global prediction visualization (Fig. 5)
│   └── VIF_P-value_Feature Importance/
├── Data/                             # Input datasets
├── Results/                          # Generated outputs (csv/jpg/txt, etc.)
├── Figures/                          # Final paper figures
└── readme.md

2. Requirements

• R >= 4.1
• Recommended OS: macOS / Linux / Windows

Main R packages used in this project:

devtools, this.path, caret, sp, gstat, automap, geosimilarity, doParallel, progress, rpart, randomForest, dplyr, ggplot2, tidyr, readxl, patchwork, ggpubr, gridExtra, car

You can install most dependencies with:

install.packages(c(
  "devtools","this.path","caret","sp","gstat","automap","geosimilarity",
  "doParallel","progress","rpart","randomForest","dplyr","ggplot2","tidyr",
  "readxl","patchwork","ggpubr","gridExtra","car","readr","MASS","scales"
))

3. Quick Start

From the repository root:

Rscript Code/Fig6_Evaluation_14models/Fig6_Evaluation_14models.R

This script calls core functions in Code/FFRK/ and writes benchmark outputs to:

Results/Fig6_Evaluation_14models/

4. Main Reproducible Experiment Entrypoints

• Rscript Code/Fig6_Evaluation_14models/Fig6_Evaluation_14models.R
  Output: Results/Fig6_Evaluation_14models/
• Rscript Code/FFRK_Ablation/FFRK_Ablation.R
  Output: Results/FFRK_Ablation/
• Rscript Code/FFRKvsFFR/FFRKvsFFR.R
  Output: Results/FFRKvsFFR/FFRKvsFFR.csv
• Rscript "Code/Fig7&8_Sensitivity_KQ/Fig7&8_Sensitivity_KQ.R"
  Output: Results/Fig7&8_Sensitivity_KQ/
• Rscript Code/Tab2_Evaluiation_HIER/Tab2_Evaluiation_HIER.R
  Output: Results/Tab2_Evaluiation_HIER/
• Rscript Code/Tab2_Evaluiation_STK/Tab2_Evaluiation_STK.R
  Output: Results/Tab2_Evaluiation_STK/
• Rscript Code/FigS1-S3_Scatter/FigS1-S3_Scatter.R
  Output: Results/Scatter/

5. Data Description

• Data/data_Cu.csv, Data/data_Pb.csv, Data/data_Zn.csv: modeling inputs for Cu, Pb, and Zn.
• Main fields include: target variable (e.g., Cu_ppm), coordinates (DLONG, DLAT), and environmental covariates (Dlith, Dfault, Slope, Water, NDVI, MainRd, Road, SOC, pH).
• Data/Global/: inputs related to global mapping and pre-trained model objects.

6. Where FFRK Is Implemented

Core method code is in Code/FFRK/R/:

• feature_extract.R: geo-self feature extraction (IDW, quantile features, similarity feature)
• modeling.R: training/evaluation workflows for OK/UK/ML/RK/FFRK/HIER
• prepare_data.R: data preprocessing (including optional log-transform)
• metrics.R: R2/MAE/RMSE metrics

7. Notes

• Results/ and Figures/ already include most outputs corresponding to the manuscript.
• For reproducibility, keep random seeds fixed (the scripts use 20250718) and use consistent package versions.

8. Citation

If this repository is useful for your research, please cite the corresponding FFRK paper: Luo, P., Wu, Y., Song, Y., Focal-Feature Regression Kriging. Geographical Analysis