Investigating UAV multispectral imagery for TSS and turbidity monitoring in small streams

Authors: PRIOR, ELIZABETH - Auburn University; O'DONNELL, FRANCES - Auburn University; BRODBECK, CHRISTIAN - Auburn University; DONALD, WESLEY - Auburn University; Runion, George

Submitted to: American Geophysical Union
Publication Type: Abstract Only
Publication Acceptance Date: 10/2/2019
Publication Date: 12/9/2019
Citation: Prior, E.M., O'Donnell, F., Brodbeck, C., Donald, W.N., Runion, G.B. 2019. Investigating UAV multispectral imagery for TSS and turbidity monitoring in small streams [abstract]. American Geophysical Union.

Interpretive Summary:

Technical Abstract: Unmanned aerial vehicles (UAV) can be used for field data collection and remote sensing. Ease of use, ability to carry sensors, low cost, and precise navigation makes them a versatile tool. This study investigated if UAV multispectral imagery could be used to assess turbidity and total suspended solids (TSS) of creeks. UAV TSS prediction has been demonstrated for rivers, but not creeks. UAV multispectral imagery (four bands: green, red, red edge and near infrared) and water samples (surface, middle and bottom of the water column) were collected before and after rain events from a restored reach of Moores Creek, Lanett, AL. Turbidity and TSS of water samples were measured. Linear regression models relating multispectral imagery to turbidity and TSS were developed. To test developed models, water samples and UAV imagery were also collected at Moores Mill Creek, Chewacla State Park, AL. To assess how high sediment levels influence UAV imagery, a large-scale sediment basin simulation was also conducted at the Auburn University Erosion and Sediment Control Testing Facility (AU-ESCTF). The basin was filled using controlled sediment introduction and flow rates; automatic samplers collected water (three depths and two basin locations) every 15 minutes for 6 hours with corresponding UAV imagery. For Moores Creek, TSS and turbidity regression models for no rain event had R2 values of 0.77 and 0.78, respectively. When sampling after rain, different single bands and band ratios were required for sufficient R2 values, suggesting separate models may be needed for high and low flow events. When Chewacla State Park band values were inputted into Moores Creek models, predicted TSS and turbidity were not comparable to measured values indicating location-specific models may be required. For the AU-ESCTF test, R2 values for TSS and turbidity models were 0.96 and 0.93, respectively. Overall, red band values increased the most indicating that this band could be used for threshold modeling.