Evaluation of Vegetation Characteristics from a Drone-Based Lidar Point Cloud for Estimating Floodplain Depth-Dependent Roughness

Abstract

Roughness is an important variable for estimating flow and creating hydrodynamic models, but it is one of the most uncertain parameters. This parameter is the measure of friction or drag that slows the flow of water, often affected by vegetation, sediments, flow depth, and channel form. Roughness changes with different depths of flow, as increasing water height overcomes the drag forces from the bed; this concept is called depth-dependent, or relative, roughness. Equations calculating vegetative roughness from plant height and certain biomechanical properties have been developed in flume experiments. Further research testing these equations in the field, using remotely sensed data, is necessary. The current study proposes a method to calculate roughness across a site for a range of flow depths and across the channel and floodplain for Stroubles Creek at the Stream Research, Education, and Management (StREAM) Lab. Certain roughness equations were conditionally calculated based on water elevation, vegetation height, ground elevation, and location within a grid of discretized pixels. An equation was developed relating vegetation hanging into the channel onto the surface of water and roughness. Manning's n, the friction parameter, was calculated for each pixel, and a weighted average roughness value for these pixels was calculated using inverse distance weighting for the range of flows analyzed at each research bridge on site. These roughness calculations were then validated with values that were measured from in-situ velocity sensors, located at Bridges 1 and 2 at the site. This process was then applied to the entire site, resulting in a 10-cm resolution roughness raster for a range of water depths. Roughness did not strictly increase or decrease with depth; instead, roughness values had distinct trends that could be related to the spacing and type of vegetation or the location of water across the landscape. Manning's n values ranged from 0.021 to 0.033 at Bridge 1 and 0.021 to 0.034 at Bridge 2. These results were applied to hydrodynamic models for two flow events and compared to previously calibrated single roughness values for the channel and floodplain. These models found relatively low error at an RMSE of 0.061 m for the relative roughness and 0.041 m for the static roughness for one event, and an RMSE of 0.130 m and 0.122 m for the relative and static roughness approaches, respectively. Flow depths were taken from the hydrodynamic model and used to recalculate roughness, and models were run again; this method resulted in an RMSE of 0.055 m and 0.107 m when compared to measured events. Field data from groundwater wells and velocity sensors that the modeled flow depths were compared to were limited. Further work to overcome the limitations of this study would include applying this process to other sites, collecting concurrent datasets, analyzing the effects of canopy changes to roughness, or applying alternative roughness equations. Despite these limitations, this method of determining roughness allows for continual calculations of roughness through different seasons and flow events from drone flights, providing insights into the changes in roughness across these scales. Furthermore, these roughness estimates, which do not require calibration, result in directly calculated values that are comparable to calibrated values in the application of a hydrodynamic model.