Power v. threshold: Near-channel morphology controls sediment rating curve shape in coastal redwood watersheds

We explored discharge-suspended sediment relationships in 71 watersheds within nine larger basins in the North Coast of California. These watersheds are highly influenced by their tectonic setting and have a long history of timber harvest activity. We summarized these relationships by generated sediment rating curves (SRC) taking the form of a power function in log-log space: SSC = aQ^b. Where SSC represents suspended sediment concentration and Q represents normalized discharge. The rating parameters a and b define the vertical offset and slope of the relationship, respectively. We investigated how these relationships varied spatially and temporally among our study watersheds. 

We quantified watershed and near-channel characteristics and land management metrics to understand controls on SRC relationships using Random Forest modeling. Random Forest models are a class of machine learning methods that can model complex, nonlinear interactions between response variables and a large number of predictor variables. 

We found two distinct SRC shapes within our study watersheds: simple power functions and threshold functions. We found a longitudinal trend in the SRC offset and slope, with the most extreme parameters located in basins closest to the Mendocino Triple Junction, an area of extremely high tectonic activity. Additionally, we found that SRC offsets increase and slopes decrease following timber harvest, with suspended sediment concentrations at low flows increasing faster than concentrations at high flows. Lastly, our Random Forest models explain about 40% of the variance in the SRC parameters. We found that timber harvest activity and near-channel local relief influence SRC offset while uplift rates and precipitation patterns influence SRC slope. Furthermore, our model correctly classified 96% of the SRC shapes using only nearchannel morphological metrics, specifically, near-channel precipitation-sensitive deepseated landslide susceptibility and near-channel soil erodibility.