By Martin von Fintel
Water is a scarce resource in South Africa, and 62% of the water used in South Africa is used for irrigation. This figure is likely to rise with increasing pressure on the agricultural sector to produce more food. Water for irrigation is stored in many small dams, scattered across the country. Due to the increasing gap between need and availability of water, which is exacerbated by climate change, there is mounting pressure to build more dams. These small dams can have negative effects on catchment areas and on the availability of water if they are not managed correctly. Therefore, there is a need for a new monitoring and management system to be developed, as the current system is ineffectual.
This study addressed this problem by determining what minimum surface area is required for a waterbody to be detected using Sentinel-1 Synthetic Aperture Radar imagery. A Random Forest classifier was used to detect waterbodies on a Sentinel-1 image calculated from a time series of imagery taken over three months. It was found that steep incidence angles outperformed shallow incidence angles, with the classification having an overall accuracy of 80%. This could possibly be explained by the effect incidence angle has on vegetation penetration, with steeper incidence angles penetrating vegetation more effectively. This would cause a greater contrast between water and vegetation for steeper incidence angles and would enable the classifier to separate waterbodies from the surrounding land cover classes better. Detection rates were almost 90% for waterbodies of 1 hectare and greater, with no false positives, and a 10% false-negative rate. These findings provide the foundation for a detection and monitoring system to be developed, which will allow for better management of water in South Africa.
Figure 1: Waterbodies can be seen as dark areas in the Sentinel-1 imagery (left). The classification of the same area can be seen on the right, with waterbodies shown in blue.
The above is the result of post-graduate research carried out by Martin von Fintel at Stellenbosch University, who graduated with a BSc Honours in Geoinformatics this past December. Martin has since accepted a position at the Department of Transport and Public Works for the Western Cape Government doing road network management.