Estimating Thresholds of Natural Vegetation for the Protection of Water Quality in South African Catchments

Many of South Africa’s current water quality problems have been attributed to diffuse pollution derived from poorly regulated land use/land cover (LULC) transformations. To mitigate these impacts, the preservation of an adequate amount of natural vegetation within catchment areas is an important management strategy. However, it is not clear how much natural vegetation cover is required to provide adequate levels of protection, nor at which scale(s) this strategy would be most effective. To investigate the possibility of estimating minimum thresholds of natural vegetation required to protect water resources, regression analysis was used to model relationships between water quality (measured using Nemerow’s Pollution Index) and metrics of natural vegetation at multiple scales across a sample of sub-catchments located along the western, southern, and south-eastern coast of South Africa. With conspicuous outliers removed, the models were able to explain up to 82% of the variability in the relationship between land use and water quality. Moreover, a statistically significant, nonlinear, and inverse relationship was found between proportions of natural vegetation cover and pollution levels. This relationship was strongest when measured (1) across the whole catchment and (2) within a 200 m riparian buffer zone. The models further indicated that approximately 80 to 90% natural vegetation cover was necessary at these scales to maintain water quality at ecologically acceptable levels. Additional nonlinear thresholds estimated using breakpoint analysis suggested that if proportions of natural vegetation fall below 45% (across the whole catchment) and 60% (within a 200 m riparian buffer zone) a dramatic increase in pollution levels can be expected. The estimated thresholds are recommended as guidelines that can be used to inform integrated land and water resources management strategies aimed at protecting water quality in the study area. Likewise, the methods described are recommended for the estimation of similar thresholds in other regions.