The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Publications Copernicus
Articles | Volume XL-7/W3
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XL-7/W3, 455–460, 2015
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XL-7/W3, 455–460, 2015

  29 Apr 2015

29 Apr 2015

Seasonal variation of land cover classification accuracy of Landsat 8 images in Burkina Faso

J. Liu1, J. Heiskanen1, E. Aynekulu2, and P. K. E. Pellikka1 J. Liu et al.
  • 1University of Helsinki, Department of Geosciences and Geography, P.O. Box 64, FI-00014, Helsinki, Finland
  • 2World Agroforestry Centre, United Nations Avenue, P.O. Box 30677, Nairobi, 00100, Kenya

Keywords: Phenology, Classification, Random Forest

Abstract. In the seasonal tropics, vegetation shows large reflectance variation because of phenology, which complicates land cover change monitoring. Ideally, multi-temporal images for change monitoring should be from the same season, but availability of cloud-free images is limited in wet season in comparison to dry season. Our aim was to investigate how land cover classification accuracy depends on the season in southern Burkina Faso by analyzing 14 Landsat 8 OLI images from April 2013 to April 2014. Because all the images were acquired within one year, we assumed that most of the observed variation between the images was due to phenology. All the images were cloud masked and atmospherically corrected. Field data was collected from 160 field plots located within a 10 km × 10 km study area between December 2013 and February 2014. The plots were classified to closed forest, open forest and cropland, and used as training and validation data. Random forest classifier was employed for classifications. According to the results, there is a tendency for higher classification accuracy towards the dry season. The highest classification accuracy was provided by an image from December, which corresponds to the dry season and minimum NDVI period. In contrast, an image from October, which corresponds to the wet season and maximum NDVI period provided the lowest accuracy. Furthermore, the multi-temporal classification based on dry and wet season images had higher accuracy than single image classifications, but the improvement was small because seasonal changes affect similarly to the different land cover classes.