Classification and assessment of land cover and land use change in southern Ghana using dense stacks of Landsat 7 ETM+ imagery

Abstract Ghana is the focus of extensive economic development interest, and is undergoing a substantial increase in population from 5 million in 1950 to an estimated 50 million in 2050. Population growth is impacting the natural environment, mostly through land cover and land use change (LCLUC), and particularly associated with agriculture expansion and urbanization. Monitoring LCLUC is necessary in order to understand the overall dynamics of population, LCLU and quality of life. However, extensive cloud cover in the region complicates satellite-based monitoring of LCLUC. Our objectives are to evaluate an innovative “dense stack” approach to image classification with extremely cloudy, multi-temporal Landsat 7 ETM+ imagery, map and quantify LCLU within southern Ghana for circa 2000 and circa 2010, examine LCLU changes, and assess the utility of the approach for monitoring human-induced change. Maximum value composite images (derived from the dense stacks) provide unique information for classifying the LCLU classes of interest, and accuracy assessment results indicate effective overall classification of the six LCLU classes mapped using semi-automated methods. A product we developed and refer to as the spectral variability vegetation index (SVVI) plays a major role in discriminating three natural vegetation classes and agriculture. Derived circa 2000 and circa 2010 LCLU maps indicate that approximately 26% of the study area exhibited LCLU change during the study period. Sixty-two percent (62%) of the changes are associated with conversion to Agriculture, with 33% from Secondary Forest, 26% from Savanna, and 3% from Forest. During the same period, 18% of circa 2000 land classified as Agriculture was fallow or abandoned by circa 2010. Change to Built represented 6% of the LCLU change, which includes 5% from Agriculture and 1% from Secondary Forest circa 2000. Freely available Landsat 7 ETM+ imagery and the time-series classification methods developed here may be used to further monitor LCLU change in the region and throughout the world, particularly in cloud-prone equatorial areas.