Figure 1
The Khor Rori study area, SW Oman, showing the study area perimeter, the location of the 15 sampled locations (S1–S15), and the distribution of control points across the landscape. Basemap: includes material © CNES (2025), Distribution Airbus DS.
Figure 2
General view of the S4 sampling campsite and the enclosed structure.
Figure 3
Map of campsite S4 illustrating the functional areas (corral, domestic) and the location of collected soil samples. Basemap: includes material © CNES (2025), Distribution Airbus DS.
Figure 4
Spatial distribution of control points relative to the buffer boundaries of campsites S6 and S7. Basemap: includes material © CNES (2025), Distribution Airbus DS.
Figure 5
Spearman correlation matrix of the elemental concentrations analysed from S4 soil samples.
Figure 6
PCA biplot displaying sample scores and elemental loadings on the first two principal components (PC1 and PC2).
Figure 7
Back-transformed MAF-cokriging results for campsite S4. (Left) Predicted elemental concentrations (%) for each analysed element; (right) corresponding kriging standard errors (delta-method propagated).
Table 1
Quantitative performance metrics for the XGBoost regression model.
| METRIC | TRAINING SET | TEST SET |
|---|---|---|
| R² Score | 0.95 | 0.62 |
| RMSE | 0.14 | 0.33 |
| MAE | 0.09 | 0.23 |
Table 2
Diagnostic performance of the regression model when evaluated as a binary anomaly detector.
| METRIC | VALUE |
|---|---|
| ROC-AUC | 0.86 |
| Recall (Sensitivity) | 83.33% |
| Precision | 87.91% |
| F1-Score | 0.86 |
| Overall Accuracy | 81% |
| True Positives | 80 |
| False Negatives | 16 |
Figure 8
Visualisation of the model classification performance, highlighting the identification of high-enrichment pixels in the vicinity. Basemap: includes material © CNES (2025), Distribution Airbus DS.
Table 3
Site-level model performance. Measured and predicted enrichment balance (mean ± SD), bias (predicted – measured), RMSE, and within-site Spearman correlation (ρ) for each sampled location. ρ not reported for S12 (n = 5). n = pixels per location; the remaining 103 pixels are off-site samples.
| SITE | N | MEASURED (MEAN ± SD) | PREDICTED (MEAN ± SD) | BIAS | RMSE | ρ |
|---|---|---|---|---|---|---|
| S1 | 135 | 0.24 ± 0.17 | 0.21 ± 0.15 | –0.03 | 0.13 | 0.67 |
| S2 | 20 | –0.01 ± 0.12 | 0.02 ± 0.11 | 0.03 | 0.10 | 0.62 |
| S3 | 75 | 0.04 ± 0.19 | 0.08 ± 0.15 | 0.04 | 0.14 | 0.75 |
| S4 | 31 | 0.30 ± 0.35 | 0.28 ± 0.27 | –0.02 | 0.10 | 0.89 |
| S5 | 53 | 0.11 ± 0.16 | 0.14 ± 0.11 | 0.03 | 0.13 | 0.56 |
| S6 | 32 | 0.55 ± 0.67 | 0.50 ± 0.49 | –0.05 | 0.30 | 0.61 |
| S7 | 52 | 0.39 ± 0.47 | 0.44 ± 0.30 | 0.04 | 0.24 | 0.69 |
| S8 | 40 | 0.76 ± 0.60 | 0.70 ± 0.47 | –0.06 | 0.19 | 0.82 |
| S9 | 31 | 1.07 ± 0.80 | 1.06 ± 0.71 | –0.01 | 0.29 | 0.87 |
| S10 | 14 | 0.52 ± 0.51 | 0.52 ± 0.34 | –0.00 | 0.27 | 0.68 |
| S111 | 14 | 0.06 ± 0.21 | 0.10 ± 0.09 | 0.04 | 0.16 | 0.78 |
| S121 | 5 | 0.16 ± 0.17 | 0.13 ± 0.10 | –0.02 | 0.08 | — |
| S13 | 39 | 0.66 ± 0.46 | 0.61 ± 0.30 | –0.05 | 0.31 | 0.76 |
| S14 | 33 | 0.23 ± 0.14 | 0.23 ± 0.14 | –0.01 | 0.12 | 0.69 |
| S15 | 31 | –0.09 ± 0.16 | –0.02 ± 0.12 | 0.06 | 0.16 | 0.54 |
Figure 9
Predictive model of anthropogenic soil enrichment across the study area, highlighting zones of high (red) and low (blue) potential human activity.