Evaluating opportunities for broad-scale remote sensing of total suspended solids on small rivers

Abstract

Although broad-scale remote sensing applications are increasingly being integrated into water quality assessments, the feasibility of such applications on smaller rivers (< 150 m. width) is not well understood, even though smaller rivers comprise most of the river network and can constitute significant sources of sediment and nutrients. We evaluated the application of a widely used Total Suspended Solid (TSS) algorithm within ten small (16 to 117 m wide) tributaries of the Lake Champlain Basin in the Northeastern US. Sentinel-2 images captured within five hours of depth-integrated TSS samples were identified and corrected for atmospheric conditions to generate TSS estimates. Linkages between sample attributes and errors in the relationship between measured and predicted TSS values were explored to identify potential limits on the application of existing TSS algorithms to small rivers. From a dataset of 64 unique measured-predicted values on rivers whose channel width was greater than one pixel (i.e., 10 m), we found that there are settings along smaller rivers, and discrete flow and seasonal conditions, under which broad-scale remote sensing applications may be applicable. The remote sensing algorithm had the greatest success predicting TSS for rivers with high background levels of sediment (i.e., large concentrations during smaller discharges). Errors in estimated TSS were three times greater on rivers with low background levels (average error of 174%) of sediment than high levels (average error of 57%), likely due to differences in the proportion of fine-grained particle-sizes in transport. We also found that estimated TSS error was less outside the growing season, when biological activity was limited and plants were dormant, and when higher flows occurred such that backscatter from the channel bed is less likely to interfere with the reflective signal. This effect was most pronounced on rivers with low background sediment loads; errors in estimated TSS were more than six times greater during the growing season on rivers with low background loads (average error of 216%) than when plants were dormant (average error of 48%). We hypothesized that algorithm performance on smaller rivers, as measured by channel width, would be prone to larger errors due to the greater influence of contrasting reflectance signals from soils and vegetation along riverbanks. The degree of limitation from this adjacency effect was not evident in our dataset perhaps because of a correlation between river size and sediment characteristics unique to our study area. An alternative parameterization of a single-band empirical TSS model for small rivers in the region, based on inclusion of readily identifiable geologic and temporal characteristics, suggests improved predictive power for small rivers. Results highlight the feasibility of extending remote sensing applications throughout the river network in temperate regions of the Northeastern U.S. Certain geologic settings, specifically those that transport finer-grained loads, are more likely to have accurate predictions, especially during the colder months, when in situ monitoring is particularly challenging, yet highly important because a significant portion of runoff events in the Northeastern U.S. occur in this time.