Remote sensing of land cover changes in the Jeffara Plain, North-West Libya

• Main Author: El-Aswed, Tarek
• Other Authors: Duck, Robert ; Cutler, Mark
• Published: University of Dundee 2009
• Subjects: 630
• Online Access: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.505649

In Libya groundwater is the key source of freshwater, providing an essential supplement to surface water sources. Libya is mostly arid and semiarid and sparsely populated large North African country with annual average precipitation rates of 200 nun. More than 950/0 of the country receives less than 100 mm, and as consequence, recharge of groundwater is extremely limited. Groundwater availability and quality are also vulnerable both to climate change and over-abstraction, and in regions where the water table has lowered there has been a consequent impact on agricultural activities. This research examines the impact of water table change on land cover (particularly agricultural activities) in part of the Jeffara Plain NW Libya, during the period 1988 to 2000 using remotely sensed data. Landsat Thematic Mapper 5 images from 1988, 1992, 1996 and 2000 have been used in addition to various thematic maps of the study area and bore-hole data to assess the nature and extent of change. A supervised Maximum Likelihood approach (ML) was used to classify each image into land cover classes that were likely to have been directly affected by groundwater changes, with resulting accuracies between 670/0 and 76%, obtained. The change in the extent of land cover classes in all images was clearly visible and occurred as either an increase or a decrease between successive dates. From the questionnaire survey, and interviewing local farmers, it is clear that groundwater changes (quantity and quality) have had a significant impact upon the vegetation cover and agricultura activities of the area. To verify the changes and assess new tools for image classification, a second approach was tested with the application of Artificial Neural Networks (ANN) as alternative image classification method, and gave results with high accuracy (over 900/0), greater than those from the ML.