Remote sensing agricultural crop type for sustainable development in South Africa

Abstract

Effective water resource management improves food production and consents scare water resources, especially in developing countries such as South Africa. The goal of this project was to assist the Republic of South Africa Department of Water Affairs & Forestry to identify a cost-effective remote sensing methodology to accurately measure and monitor agricultural land use for National Water Act water management purposes. The most accurate results were obtained using object-oriented image segmentation techniques applied to SPOT multispectral and panchromatic data (overall accuracy = 89.41%). Accurate classifications were also obtained using just the SPOT multispectral data (85.33%). When applied to Landsat ETM+ and TM data in 2003 and 2005, the image segmentation approach out-performed all other algorithms yielding classification accuracies of 85.71% and 84. 30%, respectively. This is important because Landsat TM type data cover more geography (e.g., 185 & 185 km) than SPOT data (e.g., 60 × 60 km) and are more economical per km2. These findings are of value to the Republic of South Africa as well as other countries throughout Africa that are trying to implement National Water Acts that rely on remote sensing to provide some of the critical land cover information.