Table 2.
Contrasting the accuracy between three classifiers and combinations of predicator variables. The highest overall accuracy (OA) per test is identified with bold font.
| Test | Input Layers | Decision-Tree (DT) Classification | Rule-Based (RB) Classification | Random Forest (RF) Classification | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Training Data | OA on Testing Data | Training Data | OA on Testing Data | Training Data | OA on Testing Data | ||||||||||
| # Tree Leaves | Error (%) | Mean (%) | 95% CI | # “If- Then” Rules | Error (%) | Mean (%) | 95% CI | Out-of-Box Error (%) | Mean (%) | 95% CI | |||||
| 1 | 4 traditional bands (B2 + B3 + B5 +B7) | 222 | 6.7 | 66.9 | 65.1 | 68.7 | 136 | 7.3 | 66.5 | 64.7 | 68.3 | 9.9 | 73.1 | 71.4 | 74.7 |
| 2 | 5 traditional bands (B1 + B2 + B3 + B5 +B7) | 252 | 5.4 | 69.2 | 67.4 | 70.1 | 157 | 6.1 | 66.6 | 64.8 | 68.4 | 8.8 | 74.0 | 72.4 | 75.7 |
| 3 | 8 traditional bands (B1-B8) | 270 | 3.5 | 73.1 | 71.4 | 74.7 | 168 | 4.0 | 74.7 | 73.0 | 76.3 | 6.6 | 76.7 | 75.1 | 78.2 |
| 4 | 8 traditional bands + NDVI | 255 | 3.4 | 73.4 | 71.7 | 75.0 | 161 | 4.0 | 73.5 | 71.8 | 75.1 | 6.8 | 75.7 | 74.0 | 77.3 |
| 5 | 8 traditional bands + NDWI | 251 | 3.5 | 72.8 | 71.1 | 74.5 | 152 | 4.1 | 73.6 | 72.0 | 75.3 | 6.6 | 77.0 | 75.4 | 78.6 |
| 6 | 8 traditional bands + NDSI | 219 | 3.5 | 73.1 | 71.4 | 74.7 | 165 | 3.8 | 71.9 | 70.2 | 73.6 | 6.6 | 77.0 | 75.4 | 78.6 |
| 7 | 8 traditional bands + texture | 226 | 2.7 | 78.2 | 76.6 | 79.7 | 156 | 3.1 | 77.7 | 76.1 | 79.2 | 4.9 | 81.1 | 79.6 | 82.6 |
| 8 | 8 traditional bands + elevation dataset | 139 | 2.2 | 61.6 | 59.8 | 63.4 | 162 | 2.5 | 61.9 | 60.0 | 63.7 | 4.2 | 75.3 | 73.6 | 76.9 |
| 9 | 8 traditional bands + 4 indices (NDVI, NDWI, NDSI, texture) | 225 | 2.4 | 77.2 | 75.6 | 78.8 | 140 | 2.9 | 78.7 | 77.2 | 80.2 | 5.1 | 80.6 | 79.0 | 82.0 |
| 10 | 8 traditional bands + 4 spectral indices; with boost (10 trials) | Boost | 0.1 | 80.1 | 78.6 | 81.6 | Boost | 0.0 | 80.0 | 78.5 | 81.5 | ||||
| 11 | 8 traditional bands + 4 spectral indices + 3 hydrogeomorphology variables | 49 | 0.8 | 55.3 | 53.4 | 57.1 | 48 | 0.8 | 54.8 | 52.9 | 56.6 | 1.6 | 74.7 | 73.1 | 76.3 |
| 12 | 8 traditional bands + 4 spectral indices + 3 hydrogeomorphology variables; with boost (10 trials) | Boost | 0.0 | 60.2 | 58.3 | 62.0 | Boost | 0.0 | 58.9 | 57.1 | 60.8 | ||||
| 13 | 8 traditional bands + 4 spectral indices + 3 hydrogeomorphology variables + elevation dataset | 153 | 0.7 | 58.0 | 56.1 | 59.8 | 100 | 0.8 | 58.3 | 56.4 | 60.1 | 1.3 | 73.0 | 71.2 | 74.5 |
| 14 | 8 traditional bands + 4 spectral indices + 3 hydro attributes + elevation dataset; with boost (10 trials) | Boost | 0.0 | 63.1 | 61.3 | 64.9 | Boost | 0.0 | 59.9 | 58.0 | 61.7 | ||||
| 15 | Uncorrelated and parsimonious (B1 + B3 + B5 + B7 + texture) | 221 | 3.4 | 75.7 | 74.0 | 77.3 | 154 | 3.9 | 74.0 | 72.4 | 75.7 | 6.2 | 81.2 | 79.7 | 82.6 |
| 16 | Uncorrelated and parsimonious (B1 + B3 + B5 + B7 + texture) with boost | Boost | 0.7 | 80.7 | 79.2 | 82.1 | Boost | 0.7 | 77.8 | 76.2 | 79.3 | ||||