Satellite monitoring of surface water variability in the drought prone Western Cape, South Africa

Abstract Surface water plays a fundamental role in supporting life and the functioning of various ecosystems. In this regard, its availability and variability influence the functioning of various ecosystems and most importantly human livelihoods. This becomes critical in drought prone and water-scarce areas of sub-Saharan Africa. The use of satellite remote sensing is increasingly becoming the primary data source for water resources monitoring. However, with emerging and advances in satellite remote sensing data, their ability in monitoring water resources remains uncertain. In addition, the remote sensing of surface water mapping has been limited to specific periods, which undermines the influence of seasonality on their spatial variability. This study therefore assessed the potential of Landsat-8 Operational Land Imager and Sentinel-2 Multi-Spectral Instrument derived Normalised Difference Water Index (NDWI), Modified Normalised Difference Water Index Land Surface Water Index (LSWI+5) and Modified Normalised Difference Index (MNDVI), for monitoring seasonal surface waterbodies in the Western Cape, South Africa. The derived remote sensing estimates were assessed using the Producers', User's, Overall accuracies as well the Kappa coefficient. The spatial variations were also linked to monthly evapotranspiration, rainfall and temperature. These climatic factors all contribute to seasonal surface water availability. Further, the satellite-derived estimates were compared using the analysis of variance (ANOVA), to test whether there were any significant differences (α = 0.05) in their and mapping detection abilities. The results indicated that both sensors managed to detect and map surface water variations and it was observed that surface waterbodies varied spatially over time. During the wet season, more surface waterbodies were detected than during the dry season. The majority of waterbodies were detected by both sensors. Most of these waterbodies were found in the south western part of the study area and all the indices showed similar sizes and presence of seasonal surface waterbodies. Normalised Difference Water Index (NDWI) performed the best and yielded the highest accuracy, however, the Land Surface Water Index VI (LSWI+5) results overestimated the size and occurrence of surface waterbodies, when compared to other indices, especially during the wet season. The derived waterbodies followed a similar pattern with the observed climate data and evapotranspiration rates during the study period. Landsat 8 derived Normalised Difference Water Index (NDWI) performed slightly better when compared to other indices, with 92.33% and a Kappa coefficient of 84.67% in detecting and mapping surface waterbodies. On the other hand, Sentinel-2 produced lower classification estimates. Statistical analysis results demonstrated significant differences (α = 0.05) in the detection and mapping ability of different satellite-derived indices. Overall, this work demonstrates the potential of using Landsat 8 and Sentinel 2 data in understanding spatial variations in surface water.