Characterizing the large-scale sedimentary make-up of heterogeneous riverbeds (Nelson et al., 2014), which consist of a patchwork of sediment types over small scales (less than one to several tens of meters) (Dietrich and Smith, 1984) requires high resolution measurements of sediment grain size. Capturing such variability with conventional physical (e.g. grabs, cores, and dredges) or underwater photographic sampling (Rubin et al., 2007; Buscombe et al., 2014a) would be prohibitively costly and time-consuming. However, characterizing bed sediments using high-frequency (several hundred kilohertz) acoustic backscatter from swath-mapping systems has the potential to provide near complete coverage of the bed (Brown and Blondel, 2009; Brown et al., 2011; Snellen et al., 2013), at resolutions down to a few centimeters, which photographic sampling could not practically achieve within the same time and with the same positional accuracy.

In shallow water, the physics of high frequency scattering of sound are relatively poorly understood, therefore acoustic sediment classification are almost always statistical (Snellen et al., 2013). Many such methods proposed to date are designed for characterizing large areas of seabed (Brown and Blondel, 2009; Brown et al., 2011) at relatively poor resolution (tens of meters to several hundred meters) and therefore rely on aggregation of data over scales much larger than the typical scales of sediment patchiness on heterogeneous riverbeds. In response to this need, Buscombe et al. (2014b, 2014c) developed a new statistical method for acoustic sediment classification based on spectral analysis of backscatter. This method is both continuous in coverage and of sufficient resolution (order meter or less) to characterize sediment variability on patchy riverbeds. Here, we apply these methods to multibeam echosounder (MBES) data collected from the bed of the Colorado River in Marble and Grand Canyons.

Sediment dynamics on the Colorado River in Grand Canyon National Park have been studied for several decades (e.g. Howard and Dolan, 1981; Rubin et al., 2002). Particular focus has been given to sandbars in large eddies downstream of tributary debris fans (Schmidt, 1990) because they are considered valuable resources by stakeholders and managers. Due to the severe limitations in sand supply imposed by Glen Canyon Dam (Howard and Dolan, 1981; Topping et al., 2000; Hazel et al., 2006), understanding the effectiveness of sandbar management practices, such as controlled floods (Rubin et al. 2002; Topping et al., 2006; Hazel et al., 2010), and the long-term fate of sand in Grand Canyon over decadal timescales, requires construction of accurate sand budgets, which involves detailed monitoring of influx, efflux and changes in sand storage (Topping et al., 2000; Topping et al., 2010; Grams et al., 2013) and assessments of uncertainties in sand-budget calculations (Grams et al., 2013).

In order to estimate the sand budget, it is necessary to estimate what component of observed morphological changes is sand and what component is coarser. Grams et al. (2013) classified sand and coarse substrates using topographic roughness derived from digital elevation models, but the classification skill was estimated to be only 60-70%. In addition, sand bedforms had to be delineated manually, and validation was based on grain-size observations with positional uncertainties up to tens of meters. Because the morphology of the Colorado riverbed in Grand Canyon is mapped - to a large extent - using MBES (Kaplinski et al., 2009), the primary motivation for the present study is to examine how uncertainties in sand budgets can be constrained by producing maps of surface sediment types using the completely automated methods of Buscombe et al (2014b, 2014c) based on statistical analysis of MBES acoustic backscatter.