There are many challenges associatedwith the task ofmapping LCLU
in southern Ghana using Landsat 7 imagery. In addition to the frequent
and extensive cloud cover across the region, data gaps associated with
the Landsat 7 scan-line corrector failure (SLC-off) (Markham, Storey,
Williams, & Irons, 2004), the lack of a capability to record Landsat 5
TM imagery in western Africa, and discrimination of small and informal
agricultural fields/plots (that are intermixed with natural vegetation)
using data with significantly reduced temporal coverage (due to the
cloud-related and SLC-off data gaps) are particularly challenging. To
overcome these challenges, we created temporal composite images
from available data, and classified each LCLU class individually using
the temporal composite image that provided the most robust discrimination
for that class. Specific waveband, spectral index and image texture
inputs used for image classification were determined through a
robust exploratory analysis of multiple image metrics and extensive
trial and error evaluation. Five temporal composite products were generated
by 1) masking cloud, cloud shadow, and water (except during
the classification ofWater) pixels fromeach of the images used to create
the composite, 2) stacking the masked input layers into a single multiband
image, and 3) computing the maximum value for each pixel
through the temporal stack.
There are many challenges associatedwith the task ofmapping LCLUin southern Ghana using Landsat 7 imagery. In addition to the frequentand extensive cloud cover across the region, data gaps associated withthe Landsat 7 scan-line corrector failure (SLC-off) (Markham, Storey,Williams, & Irons, 2004), the lack of a capability to record Landsat 5TM imagery in western Africa, and discrimination of small and informalagricultural fields/plots (that are intermixed with natural vegetation)using data with significantly reduced temporal coverage (due to thecloud-related and SLC-off data gaps) are particularly challenging. Toovercome these challenges, we created temporal composite imagesfrom available data, and classified each LCLU class individually usingthe temporal composite image that provided the most robust discriminationfor that class. Specific waveband, spectral index and image textureinputs used for image classification were determined through arobust exploratory analysis of multiple image metrics and extensivetrial and error evaluation. Five temporal composite products were generatedby 1) masking cloud, cloud shadow, and water (except duringthe classification ofWater) pixels fromeach of the images used to createthe composite, 2) stacking the masked input layers into a single multibandimage, and 3) computing the maximum value for each pixelthrough the temporal stack.
การแปล กรุณารอสักครู่..
