Satellite image based vegetation classification of a large area using limited ground reference data: a case study in the Usa Basin, north-east European Russia
Predicted global changes can be studied effectively by combining spatially explicit data sets on vegetation and other landscape properties with process models. However, detailed knowledge of the vegetation distribution of remote Arctic areas is relatively scarce. This paper shows how a mesoscale veg...
| Published in: | Polar Research |
|---|---|
| Main Authors: | , , |
| Format: | Article in Journal/Newspaper |
| Language: | English |
| Published: |
Norwegian Polar Institute
2004
|
| Subjects: | |
| Online Access: | https://polarresearch.net/index.php/polar/article/view/2073 https://doi.org/10.3402/polar.v23i1.6266 |
| Summary: | Predicted global changes can be studied effectively by combining spatially explicit data sets on vegetation and other landscape properties with process models. However, detailed knowledge of the vegetation distribution of remote Arctic areas is relatively scarce. This paper shows how a mesoscale vegetation and land cover classification of a large, remote Arctic area can be conducted at a fine spatial resolution (30 m cell size) using a limited ground reference data set. The study area is the catchment of the River Usa (93 500 km2) in north-eastern European Russia. Vegetation zones in the Usa Basin range from taiga in the south to forest-tundra and tundra in the north, and to alpine in the Ural mountains in the east. Classification was done using a mosaic of spectrally adjusted Landsat TM5 images from five different dates and a semi-supervised method. Ground reference data were collected during the summers of 1998, 1999 and 2000. Accuracy of the 21-class vegetation type/land cover classification produced was tested against test points interpreted from oblique aerial photographs taken from a helicopter (logistic limitations prohibited the collection of representative ground reference data). The main vegetation types (forests, willow dominated stands and meadows, peatlands, tundra heaths, mainly unvegetated areas, and water bodies) were distinguished with relatively high accuracy: 84% of the test points were classified correctly. Spatially detailed land cover data sets like the one described here allow detailed landscape-level analysis and process modelling on many different subjects. |
|---|