A comparison of Landsat 8 and Sentinel 2 in mapping chlorophyll-a and turbidity in the Vaal Dam, and Lakes Shinji and Nakaumi of Japan

Abstract

The Vaal Dam of South Africa and Lakes Shinji and Nakaumi of Japan play an important part in human activities such as agriculture, industry and domestic utilities. Two major problems confronting both the Vaal Dam and Lakes Shinji and Nakaumi are eutrophication and high levels of turbidity. Remote sensing technologies provide an effective means of monitoring the quality of water bodies. Landsat 8, Sentinel 2 and field hyperspectral sensors were investigated using support vector regression to find the bands and band ratios that best predict turbidity and chlorophyll-a in Lakes Shinji and Nakaumi (LSN), and the Vaal Dam. For Vaal Dam chlorophyll-a, the best bands/band ratios were found to be the global environmental monitoring index (GEMI) for Landsat 8, B2/B1 for Sentinel 2 and Rrs687/Rrs675 for the field hyperspectral remote sensing reflectance (Rrs) data. For Lakes Shinji and Nakaumi, the best bands/band ratios for chlorophyll-a were found to be B2/B3 for Landsat 8 B7/B6 for Sentinel 2 and Rrs881/Rrs882 for the field hyperspectral Rrs data. For retrieving turbidity, the best bands/band ratios for Vaal Dam were found to be B4/B1 for Landsat 8, B3/B1 for Sentinel 2 and Rrs819/Rrs794 for the hyperspectral Rrs data. For LSN, the best bands/band ratios for turbidity were found to be B9/B8 for Landsat 8, B11/B1 for Sentinel 2 and Rrs580/Rrs575 for the hyperspectral data. Support vector regression was found to perform well even with relatively small samples for the Vaal Dam achieving R2 values of 0.92 for Hyperspectral Rrs data, 0.91 for Sentinel 2 and 0.79 Landsat 8 in predicting chlorophyll-a, and 0.68 for hyperspectral Rrs data, 0.89 for Sentinel 2 and 0.70 for Landsat 8 in predicting turbidity. Lakes Shinji and Nakaumi did not achieve support vector regression results as good as those of the Vaal Dam due to small sample size. For chlorophyll-a, R2 values of 0.75 for the field hyperspectral Rrs data, 0.70 for Sentinel 2 and 0.13 for Landsat 8 were achieved. Turbidity R2 values were 0.68 for the field hyperspectral data, 0.39 for Sentinel 2 and 0.44 for Landsat 8. This work finds that Landsat 8 and Sentinel 2 can be effectively used to monitor water quality in small water bodies. As a general deduction from the results, Sentinel 2’s performance is better than that of Landsat 8 in predicting both chlorophyll-a and turbidity and is comparable to the field hyperspectral Rrs data. However, it is acknowledged that Landsat 8 has more historical archives useful for time series analyses. The two satellites should be used in a complementary manner for greater probabilities of obtaining cloud-free images