Description

Abstract: Kenya’s Rift Valley has been undergoing rapid land cover change for the past two decades, which has resulted

in ecological and hydrological changes within the region. An effort is underway to quantify the timing and rate of

these changes in an experimental watershed near the towns of Njoro and Nakuru using remote sensing and

geographic information system (GIS) methods. Three Landsat TM images representing a 17-year period in which

the area underwent significant land cover transition were classified. Baldyga et al. (2004) showed that vegetation

diversity and temporal variability posed many challenges and resulted in large classification errors for four scenes

captured during the dry season. While the distinction between forested and unforested areas was clear, capturing

variability in agriculture informational classes proved difficult. Several enhancements were employed in an effort to

capture this variability. Band separability analysis coupled with ancillary data indicated that 14 informational classes

are distinguishable using various band combinations. Field validation was used to quantify error throughout the

watershed, and uncertainty in land cover classification among vegetation types with similar spectral response was

reduced relative to previous attempts to classify land cover into 10 classes. Resulting classified land cover scenes

will serve as input to GIS-based models as part of a systems approach to understanding watershed dynamics.

Research findings will improve historical land cover transition estimates and aid in interpreting land cover change

impacts on biophysical and human dimensions.