| Title | Evaluation of Landsat based fractional land cover mapping in the Alberta Oil Sands Region |
| Download | Downloads |
| |
| Licence | Please note the adoption of the Open Government Licence - Canada
supersedes any previous licences. |
| Author | Pouliot, D; Parkinson, W; Latifovic, R |
| Source | Geomatics Canada, Open File 17, 2015, 17 pages, https://doi.org/10.4095/296802  Open Access |
| Year | 2015 |
| Publisher | Natural Resources Canada |
| Document | open file |
| Lang. | English |
| Media | on-line; digital |
| File format | pdf |
| Province | Alberta |
| NTS | 74E |
| Area | Alberta Oil Sands Region |
| Lat/Long WENS | -112.0000 -111.0000 58.0000 57.0000 |
| Subjects | geophysics; fossil fuels; hydrocarbons; hydrocarbon potential; oil; satellite imagery; remote sensing; petroleum resources; resource management; Landsat |
| Illustrations | location maps; tables; profiles; plots |
| Program | Remote Sensing Science |
| Released | 2015 07 21 |
| Abstract | In the Alberta Oil Sand Region (AOSR) high spatial resolution (<5 m) remotely sensed multispectral time series are needed to capture the varying size and rates of change that occur. However, limited
spatial-temporal coverage and cost of current high resolution sensors make such an approach impractical for retrieval of historical information and long term monitoring. Using moderate spatial resolution (~30m) time series such as that available with
the Landsat series of sensors provides an alternative. In this research the potential to derived sub-pixel information on land cover types was evaluated for the AOSR using Landsat time series. Sub-pixel land cover fractions were trained for Landsat
using high resolution (2 m) Geoeye data classified into basic land cover types. Cover types evaluated included conifer forest, broadleaf forest, shrub, low vegetation cover, bare, and water. The point spread function of Landsat was modeled to ensure
that the reflectance properties measured were coincident with the training footprint in the higher spatial resolution Geoeye scenes. Decision tree classifiers were used for the fractional modeling. Results showed that land cover fractions could be
estimated over the region with an average absolute error ranging from 7-17%. Sampling exerted a significant effect where validation using a holdout Geoeye scene preformed inferior to sampling from all available scenes as expected. Water and bare
covers had limited sampling for fractions between 25-75% and therefore the results for these covers are uncertain in this range. Better controlling for spectral variability, fractional training and Landsat data quality in site specific analysis
suggests significant improvement in accuracy compared to the regional analysis. The improvement for the site specific analysis ranged from 5-10%. Examination of forest fraction sensitivity to change revealed good agreement with forest harvesting and
fire, but did not capture insect related damage well. These findings suggest there is potential for fractional land cover retrieval, but error is likely to remain moderate if training and remote sensing data are not carefully controlled for large
regional applications. |
| Summary | (Plain Language Summary, not published)
In remote sensing, sub-pixel or fractional land cover represents the area covered by a given land cover type within a defined footprint on the land
surface. Accurate estimation of fractional land cover is of interest to provide more detailed information on land cover and its change over time. In the Alberta Oil Sands this is of particular importance due to the small size and rapid rate of the
change observed. Results of the analysis undertaken in this study show that fractional estimation of conifer, deciduous, shrub, low vegetation, bare, and water is possible, but significant improvements can be made with enhanced data
processing. |
| GEOSCAN ID | 296802 |
|
|