Authors: Min Xu*, Department of Geography, The University of Alabama, Tuscaloosa, AL 35487, USA, Hongxing Liu, Department of Geography, The University of Alabama, Tuscaloosa, AL 35487, USA, David Mitchell, Department of Field Science, The National Ecological Observatory Network, Tuscaloosa, AL 35405, USA, Richard Beck, Department of Geography and GIScience, University of Cincinnati, Cincinnati, OH 45221, USA
Topics: Water Resources and Hydrology, Remote Sensing
Keywords: turbidity, water quality, remote sensing, ensemble model
Session Type: Paper
Start / End Time: 9:35 AM / 10:50 AM
Room: Virtual Track 1
Presentation File: No File Uploaded
In previous studies, a single empirical model was normally used to map turbidity for the entire waterbody under study, whereas its performance is often limited due to the complex optical properties of inland waters. Moreover, traditional empirical models are often one-time applications that cannot be extended to other waterbodies for regional water quality mapping. To overcome the limitations, this study adopts a multi-predictor ensemble model, which synergistically exploits a set of empirical models to obtain optimal estimation of turbidity in different water conditions. In this study, we processed four spatially consecutive Landsat 8 multispectral image scenes acquired on May 2nd, 2019, which were combined to cover the rivers of the Tombigbee River Basin, the Mobile River and the Mobile Bay. During the Landsat 8 overpass, in situ water-truth data at 20 sites were collected with YSI water quality sonde. The results show that the multi-predictor ensemble model has remarkably improved the turbidity prediction accuracy by 33% compared with the best traditional empirical model. Our analysis shows that Tombigbee River have significantly higher turbidity level than the Black Warrior River near Demopolis where they converge. Within the Tombigbee River Basin, the lower Tombigbee is more turbid than the upper Tombigbee. We also observed that the Mobile River and Mobile Bay has very high turbidity levels. This study demonstrates that our multi-predictor ensemble model possesses an improved prediction ability and stronger spatial extensibility than traditional empirical models and hence is critical for basin and regional scale river water quality monitoring and assessment.