|
 |
 |
 |
 |
 |
 |
 |
 |
| MORE INFORMATION ABOUT WP2 - Sub-pixel mapping of LC gradient info (time series) |
Use of regression analysis and multi-layer perceptrons for spectral unmixing requires reference proportions from an independent source. For this purpose, we make use of a land-cover map obtained by classification of a high resolution satellite image. A temporal filtering technique is applied to detect land cover changes between this reference map and the images of the time series. All the applied unmixing techniques have the problem of confusing some urban surface types (e.g. bright roofs) with bare soil and fallow land,
|
which is a prominent land cover type in our study area. This problem cannot be directly solved in the unmixing process itself. We therefore develop an urban mask, i.e. a raster map indicating which pixels of the medium resolution image belong to the urban fabric and which do not. Using this mask, we subject only urban pixels to the unmixing analysis, assuming that no bare soil occurs within the area covered by the mask.
|
|
Last modification date =
12-03-2009
|