The potential of hyperspectral remote sensing in determining water turbidity as a water quality indicator.
Globally, water turbidity remains a crucial parameter in determining water quality. South Africa is largely regarded as arid and is often characterised by limited but high intensity rainfall. This characteristic renders most of the country’s water bodies turbid. Consequently, the use of turbidity as a measure of water quality is of great relevance in a South African context. Generally, turbidity alters biological and ecological characteristics of water bodies by inducing changes in among others temperature, oxygen levels and light penetration. These changes may affect aquatic life, ecosystem functioning and available water for industrial and domestic use. Siltation, a direct function of turbidity also impacts on the physical storage of dams and shortens their useful life. To date, determination of water turbidity relies on the tradition laboratory based methods that are often time consuming, expensive and labour intensive. This has increased the need for more cost effective means of determining water turbidity. In the recent past, the use of remote sensing techniques has emerged as a viable option in water quality assessment. Hyperspectral remote sensing characterizes numerous contiguous narrow bands that have great potential in water turbidity measurement. This study explored the applicability of hyperspectral data in water turbidity detection. It explored the visible and near-infrared region to select the optimal bands and indices for turbidity measurement. Using the Analytical Spectral Device (ASD) field spectroradiometer and a 2100Q portable turbidimeter, spectral reflectance and laboratory based turbidity measurements were taken from prepared turbid solutions of predetermined concentrations (i.e. 10g/l to 150g/l), respectively. The Pearson’s coefficient of correlation and R2 values were employed to select optimal spectral bands and indices. The findings showed a positive linear relationship between reflectance, the amount of soil in water and turbidity values. The strongest relationships came from bands 528, 489, 657, 1000 and 983, reporting adjusted R2 values of 0.7062, 0.7004, 0.6864, 0.7120 and 0.6961, respectively. The highest coefficient came from band 1000nm. The strongest indices were 625/440 and (770-1000)/(770+1000), with adjusted R2 values of 0.6822 and 0.6973 respectively. The use of hyperspectral data in turbidity detection is ideal for optimal band interrogation. Although good results were generated from this study, further investigations are needed in the near-infrared region.