Google Earth Engine for large-scale land use and land cover mapping: an object-based classification approach using spectral, textural and topographical factors. Issue 6 (18th August 2021)

• Record Type: Journal Article
• Authors: Shafizadeh-Moghadam, Hossein; Khazaei, Morteza; Alavipanah, Seyed Kazem; Weng, Qihao
• Is Part Of: GIScience & remote sensing. Volume 58:Issue 6(2021)
• Page Start: 914
• Page End: 928
• Publication Date: 2021-08-18
• Subjects: Land use and land cover -- simple non-iterative clustering -- multi-temporal NDVI -- topographic data -- arid and semi-arid region mapping -- climate zones; Geodesy -- Periodicals; Cartography -- Periodicals; Aerial photogrammetry -- Periodicals; Remote sensing -- Periodicals; 526.05
• DOI: 10.1080/15481603.2021.1947623
• Languages: English
• ISSNs: 1548-1603

Abstract

ABSTRACT: Mapping the distribution and type of land use and land cover (LULC) is essential for watershed management. The Tigris-Euphrates basin is a transboundary region in the Middle East shared between six countries, but a recent fine-scale LULC map of the area is lacking. Using Landsat-8 time series, a 30-m resolution LULC map was produced for the Tigris-Euphrates basin. In total, 1184 Landsat scenes were processed within the Google Earth Engine (GEE). For the collection of ground truth data, differential manifestations of green cover were considered by dividing the study area into five climatic regions and the training samples were taken from each sub-region. To account for the temporal variation of LULC types, six two-month interval composite layers, including the spectral and thermal bands of Landsat-8, texture and spectral indices, as well as topographic factors were created for the target year 2019. Image segmentation and classification were performed using the simple non-iterative clustering (SNIC) and Random Forest (RF) algorithms, respectively. A computationally effective parallel processing approach was developed, which created a number of tiles and sub-tiles and a bulk command was converted into smaller parallel commands. The generated LULC map showed a satisfactory overall accuracy of 91.7%, with the highest User's accuracy in water and wetland, and the lowest in rainfed crop and rangeland and the highest Producer's accuracy in water and barren areas, and the lowest in garden and rangeland. This study highlights the necessity of using multi-temporal data for LULC mapping, in particular, multi-temporal NDVI, for the separation of different green cover types in arid and semi-arid environment.