  |
|
Upland/Lowland Stratification
The upland/lowland stratification was developed using the Minnesota land-use/land-cover map. This stratification improved the cover type accuracies for both strata. |
 | |
Thermal Bands
 |
The thermal bands improved results, more for the forest cover type classification (shown here) than for volume estimation.
| |
Spectral Band Weighting
The weighting of spectral bands improved the cover type classification accuracy from 70% to 82% for k = 1, when all subplots were included in the analysis. However, optimization was subject to local optima. |
 | |
Confusion Matrix Examination
 |
A typical example (on the left) of a confusion matrix for volume estimation reveals a typical problem in estimating the low and high end values. This problem increases as the number of nearest neighbors increases. | |
FIA 4-Subplot Design
The four-subplot cluster has some unique aspects when used with kNN methods. Since the subplots are very close to each other, the pixel values tend to be closest within the cluster. Further, subplots in the same condition class on a plot are classified collectively, in cover type by FIA.
|
| |
){kind=spacer}
){kind=spacer}
){kind=spacer}
){kind=spacer}
){kind=spacer}