Water | Built & Bare | Fields and grass | Herbaceous & reed | Forest | Shrubs |
Water | 18 | 3 | 5 | 7 | 0 | 2 | 51% |
Built & Bare | 0 | 201 | 5 | 6 | 6 | 5 | 90% |
Fields and grass | 0 | 6 | 188 | 9 | 3 | 1 | 91% |
Herbaceous & reed | 0 | 9 | 7 | 177 | 1 | 11 | 86% |
Forest | 0 | 2 | 3 | 6 | 195 | 15 | 88% |
Shrubs | 0 | 9 | 2 | 10 | 20 | 125 | 75% |
100% | 87% | 90% | 82% | 87% | 79% | 86% |
A special case is confusion of water pixels with other land cover
classes. As river water tables vary in time and the reference ground
truth data does not, most water class errors seemed attributable to
this. Figure 5 shows a day-classification result based on a single
image. Clearly around some floodplain water bodies (in red circle) bare
soil can be observed, when comparing these to the fixed legal map
(Figure 6) this base soil comes out as change (red) because is it seen
as water to ‘build and bare’ change.