Classification of ground lichen using Sentinel-2 and airborne laser data
The northern part of Sweden has two overlapping land-use interests: forestry and reindeer husbandry. Forestry affects reindeer husbandry in several ways; most important of these is its impact on ground lichens. Lichens are the primary winter grazing resource for reindeer, therefore, mapping of liche...
| Autor principal: | |
|---|---|
| Formato: | H3 |
| Lenguaje: | Inglés |
| Publicado: |
SLU/Dept. of Forest Resource Management
2018
|
| Materias: |
| _version_ | 1855572238117371904 |
|---|---|
| author | Larsson, Helene |
| author_browse | Larsson, Helene |
| author_facet | Larsson, Helene |
| author_sort | Larsson, Helene |
| collection | Epsilon Archive for Student Projects |
| description | The northern part of Sweden has two overlapping land-use interests: forestry and reindeer husbandry. Forestry affects reindeer husbandry in several ways; most important of these is its impact on ground lichens. Lichens are the primary winter grazing resource for reindeer, therefore, mapping of lichens is of interest.The objective of this study is to evaluate the use of remote sensing data from the new Sentinel-2 satellite for the classifcation of ground lichen and to assess whether adding information derived from airborne laser scanning (ALS) willimprove the result. The study area is situated in the reindeer husbandry area inland of Umeå, in the north of Sweden, and consists of two Sentinel-2 granules. Two Sentinel-2 images were used, one from 2015-08-19 and one from 2016-10-02. ALS-derived metrics was also used in the form of DEM, wetness index, a canopy density metric and forest height. Classifcation of lichen coverage was carried out with the Random Forest algorithm, and 90 field plots were used as training data. Due to the small field dataset, the evaluation method for this study was internal cross-validation. Fourteen different classifcation schemes were tried with the Random Forest algorithm. Classifcation scheme 6 (0-33 %, 34-66 % and 67-100 % lichen coverage) was the most interesting of the classification schemes with three classes, since it has the lowest out-of-bag error at 29 %. Classifcation scheme 4 (0-25 %, 26-50 % and 51-100 % lichen coverage), which is based on the Swedish National Forest Inventory’s lichen class defnition, also proved to be fairly accurate, with an out-of-bag error of 37 %. Overall, the analysis showedthat bands 4 (red) and 8 (NIR) of the Sentinel-2 2015-08-19 image, along withALS-derived canopy density were the most important variables. Wetness indexwas the least important variable. For the Sentinel-2 2016-10-02 image, bands4 (red) and 5 (red-edge) were the most important. This study showed that aSentinel-2 image from a one date during the summer season worked well forthe classification of lichen into three classes, and that adding an ALS-derivedcanopy density metric could improve the results. The use of both Sentinel-2images together did not give better classification results. |
| format | H3 |
| id | RepoSLU13401 |
| institution | Swedish University of Agricultural Sciences |
| language | Inglés |
| publishDate | 2018 |
| publishDateSort | 2018 |
| publisher | SLU/Dept. of Forest Resource Management |
| publisherStr | SLU/Dept. of Forest Resource Management |
| record_format | eprints |
| spelling | RepoSLU134012019-02-27T11:51:33Z Classification of ground lichen using Sentinel-2 and airborne laser data Larsson, Helene remote sensing random forest reindeer The northern part of Sweden has two overlapping land-use interests: forestry and reindeer husbandry. Forestry affects reindeer husbandry in several ways; most important of these is its impact on ground lichens. Lichens are the primary winter grazing resource for reindeer, therefore, mapping of lichens is of interest.The objective of this study is to evaluate the use of remote sensing data from the new Sentinel-2 satellite for the classifcation of ground lichen and to assess whether adding information derived from airborne laser scanning (ALS) willimprove the result. The study area is situated in the reindeer husbandry area inland of Umeå, in the north of Sweden, and consists of two Sentinel-2 granules. Two Sentinel-2 images were used, one from 2015-08-19 and one from 2016-10-02. ALS-derived metrics was also used in the form of DEM, wetness index, a canopy density metric and forest height. Classifcation of lichen coverage was carried out with the Random Forest algorithm, and 90 field plots were used as training data. Due to the small field dataset, the evaluation method for this study was internal cross-validation. Fourteen different classifcation schemes were tried with the Random Forest algorithm. Classifcation scheme 6 (0-33 %, 34-66 % and 67-100 % lichen coverage) was the most interesting of the classification schemes with three classes, since it has the lowest out-of-bag error at 29 %. Classifcation scheme 4 (0-25 %, 26-50 % and 51-100 % lichen coverage), which is based on the Swedish National Forest Inventory’s lichen class defnition, also proved to be fairly accurate, with an out-of-bag error of 37 %. Overall, the analysis showedthat bands 4 (red) and 8 (NIR) of the Sentinel-2 2015-08-19 image, along withALS-derived canopy density were the most important variables. Wetness indexwas the least important variable. For the Sentinel-2 2016-10-02 image, bands4 (red) and 5 (red-edge) were the most important. This study showed that aSentinel-2 image from a one date during the summer season worked well forthe classification of lichen into three classes, and that adding an ALS-derivedcanopy density metric could improve the results. The use of both Sentinel-2images together did not give better classification results. SLU/Dept. of Forest Resource Management 2018 H3 eng https://stud.epsilon.slu.se/13401/ |
| spellingShingle | remote sensing random forest reindeer Larsson, Helene Classification of ground lichen using Sentinel-2 and airborne laser data |
| title | Classification of ground lichen
using Sentinel-2 and airborne laser
data |
| title_full | Classification of ground lichen
using Sentinel-2 and airborne laser
data |
| title_fullStr | Classification of ground lichen
using Sentinel-2 and airborne laser
data |
| title_full_unstemmed | Classification of ground lichen
using Sentinel-2 and airborne laser
data |
| title_short | Classification of ground lichen
using Sentinel-2 and airborne laser
data |
| title_sort | classification of ground lichen
using sentinel-2 and airborne laser
data |
| topic | remote sensing random forest reindeer |