Assessing the effects of sampling time on the accuracy of extracting water quality indicators through geospatial techniques: A case study of Rietvlei Dam

Abstract

Freshwater is scarce, and the Earth’s ability to supply it is equally limited. Anthropogenic activities, urbanisation, and the effects of climate change degrade freshwater quality at a rapid rate. To design an effective water quality monitoring strategy that can respond to the rapid changes in surface water, regular sampling and analyses are essential. In developing countries, traditional methods of water sampling and laboratory analyses are adopted to generate water quality information. There is, however, a growing global adoption of optical remote satellite sensing for monitoring water quality, in conjunction with field surveys. One of the main challenges of optical remote sensing is sensitivity to cloud cover, which makes it difficult to synchronise with field surveys. This research addressed the question: “What is the effect of sampling time on the accuracy of extracting water quality indicators through remote sensing and geospatial techniques?”, using a case study of Rietvlei Dam, South Africa. The research established the correlation between laboratory-analysed water quality and the Sentinel-2 (S2) data extracted from the months before, during and after the given field sampling date. This covered four sampling periods i.e., mid-winter, late winter, early spring, and mid-spring. Through multiple regression and accuracy analyses, the study determined the efficacy of the S2 remote sensing models for the chlorophyll-a, turbidity, coloured dissolved organic matter (CDOM) and total suspended solids (TSS) water quality indicators from the different field and satellite sampling periods. The S2 remote sensing models predicted the spatial distribution of the water quality indicators and high concentration area coverage of the dam in hectares (ha). S2 models were found to be best for early spring. Chlorophyll-a was extracted five days after field sampling at R2= 0.79; for late winter CDOM was derived 44 days before the field sampling period at R2 = 0.64; for late winter turbidity was optimally extracted 36 days after field sampling season at R2= 0.54; and mid-winter TSS was extracted seven days before field sampling season at R2=0.62. The turbidity and chlorophyll-a S2 models’ spatial prediction of highly concentrated areas covered 36.14 and 28.92ha, respectively. In conclusion, the accuracies of mapping water quality indicators with remote sensing data and field observations differ based on an indicator measured and the time sampling difference between them. In general, the mapping accuracies decrease outside a seven-day sampling difference, whether before or after satellite image acquisition. This study contributes to the scientific understanding of remote sensing applications to the management of water quality in a given area, especially inland water quality.