Predicting the Distribution of Farm Dams in Rural South Africa Using GIS and Remote Sensing

Abstract

AbstractFarm dams are critical for their role in livestock watering and irrigation in areas that experience seasonal rainfall variability, but although cumulatively they store large volumes of water they are rarely considered as part of the water resources base or the water management system of a river basin. This study mapped farm dams and identified the factors influencing their spatial distribution in a third-order agricultural catchment of KwaZulu-Natal Province, South Africa, using GIS and remote sensing techniques. Using Landsat 8 imagery and verified using Google Earth, 864 farm dams were identified. Six physical properties of the catchment (slope, aspect, elevation, land use, soil type, and geology) were then used in a predictive model to evaluate the extent to which they influence the siting of the farm dams. Results showed that all of these factors except for aspect gave significant p-values. Slope, geology, and elevation have a greater influence on the farm dam location than soils. A multivariate logistic regression model was then created for predicting future farm dam sites. Slope, elevation, land use, and geology were fitted to the model, but aspect and soil type had no significant p-values in the model. The model was validated using 300 farm dam sites, and predicted 60% of sites correctly. This study highlights that remote sensing and GIS methods can usefully inform on strategies for more efficient water management and infrastructure planning through the locating of farm dams, and can therefore help in more efficient water use in locations that are already water-stressed.KeywordsCatchment propertiesFarms damsLandsat imageryRegression modellingWater resources management