Jump to navigation
Determining The Type And Starting Time Of Land Cover And Land Use Change In Ghana Based On Discrete Analysis Of Dense Landsat Image Time Series
Stow, DouglasWeeks, JohnCaves, Roger
x, 38 pages : illustrations (some colored).
Accra, Ghana and environs have experienced extensive land cover and land use change, which warrants more frequent monitoring. In the study I develop and test approaches for semi-automatically identifying the type and date of land cover and land use change from multi-temporal series of Landsat ETM+ imagery from 2000 to 2014. Clouds, cloud shadows, and scan line corrector-off creates missing or null data in the ETM+ images. Forty-one dates of ETM+ images that partially contain missing data were used in this study. The general approach is to conduct a per-pixel supervised classification on each date of image after masking null data based on stable training sites. Spatial, temporal, and logical filters are applied to correct for misclassification and missing data. Each image is classified into three general classes: Built, Natural Vegetation, and Agricultural, with expansion of Built being our main focus. Reference data for Change-to-Built were independently selected from all available high spatial resolution satellite images, and the beginning time of change was recorded. The change product was used to characterize the urban expansion around Accra. The result shows that the temporal-filtered product identified both the location and the start of Change-to-Built more precisely and accurately. Based on reference data derived from visual imagery analysis, 40% of the Change-to-Built samples were correctly identified without filtering, whereas 80% were correctly identified when applying temporal filter with low amounts of false positive Change-to-Built pixels. The temporal-filtered products have the highest precision and accuracy in identifying the start of Change-to-Built: the identified Change-to-Built samples are on average 2.1 years difference with the reference data. Under the limitation of frequent cloud effect and limited historical archives of high resolution images, a discrete classification approach to LCLUC mapping is shown to be successful by this study, but continuous index approaches needs to be evaluated in future research
Includes bibliographical references (pages 36-38).
Arts and Letters
Master of Science (M.S.) San Diego State University, 2015
© 2015 SDSU Library & Information Access. All Rights Reserved.