*2.3. Study Site and Data Availability*

We applied the WRF-Hydro-Sed model on the Goodwin Creek Experimental Watershed (GCEW), Mississippi, USA to assess its performance (Figure 3). GCEW is located at northwest Mississippi, close to Batesville. The watershed has a drainage area of 21.3 km2 with an outlet located at its southwest corner (89◦54 50" W, 34◦13 55" N). The Goodwin Creek is a tributary of the Long Creek, which flows into the Yocona River, one of the main rivers of the Yazoo River Basin [13]. The weather is hot and humid in summer and mild in winter, with an average annual rainfall of 1440 mm and a mean annual runoff of 145 mm during 1982 to 1992 [9]. Within the watershed, the elevation ranges from 68 m to 130 m above sea level. Around 50 percent of the watershed has a slope less than 0.02 and 15 percent has a slope larger than 0.03 [9]. The channels in the watershed extend mainly from northeast to southwest with an average slope of 0.004. Based on the State Soil Geographic Database [25], soils within the watershed are mainly silt loam and sandy loam, with the former one dominating. According to the 24-category Land Use Categories by the U.S. Geological Survey (USGS), the most common land cover types in this watershed are "dryland cropland and pasture", "irrigated cropland and pasture", and "deciduous broadleaf forest".

GCEW was originally established in 1977 and has been operated by the U.S. Department of Agriculture (USDA) National Sedimentation Laboratory (NSL) to study the influence of land use and upland erosion on sedimentary process and channel stability and to test numerical models. It is highly instrumented, with 32 standard recording rain gauges distributed uniformly within the watershed, 14 stream gages and supercritical flow structures located along the channel to collect discharge and sediment concentration data. In addition, periodic surveys are conducted to track land use conditions, channel geometry, and channel migration. In this study, rainfall data from 16 rain gauges and the streamflow and sediment concentration data at the outlet (MSGC1) was collected to calibrate and validate the model (Figure 3). The data interval is ~15 min.

**Figure 3.** Topography of Goodwin Creek Experimental Watershed (GCEW) and gage location (DEM data source: National Hydrography Dataset Plus (Version 2), [26]).

#### *2.4. Model Setup*

WRF-Hydro supports the coupling between hydrological components and atmospheric models. However, as the focus of this study is to introduce the sediment module instead of investigating the interaction between atmosphere and hydrological processes, WRF-Hydro is used in a "one-way coupled" mode.

The Noah land surface model with multi-parameterization options (Noah-MP) [27] is responsible for simulating land surface physics in the model (Figure 1). In this study, Noah-MP has a domain of 14 km × 11 km with a spatial resolution of 1 km. To drive the land surface model, static land surface physiographic data was generated using the WRF Pre-processing System (WPS). The land use information was interpolated from the 24-category USGS land use database derived from the 1-km Advanced Very High Resolution Radiometer (AVHRR) satellite images [28].

The meteorological forcing data is from the North American Land Data Assimilation System Phase 2 Forcing Dataset (NLDAS-2 hereafter, [29]). NLDAS-2 contains incoming longwave and shortwave radiation, near surface wind, specific humidity, air temperature, surface pressure as well as rainfall intensity with hourly temporal resolution. In this study, the 1/8-degree NLDAS-2 was regridded to 1 km to match the Noah-MP grid. Given its coarse resolution, the rainfall intensity of NLDAS-2 might not be able to fully represent the condition of such a small watershed. Therefore, we replaced NLDAS-2 s original rainfall data field with interpolated precipitation from the records at the 16 rain gauges using the inverse distance weighting interpolation method.

With a disaggregation factor of 20, the hydrological physical processes are simulated at the spatial resolution of 50 m. The subsurface routing, overland routing and channel routing of WRF-Hydro are all activated. The time step of the overland routing and channel routing is six seconds. Since the sediment processes are driven by the overland flow and channel flow, the setup of the sediment model highly depends on that of the hydrological model. The sediment model is calculated on the same grid as the hydrological model (50 m), with a time step of six seconds as well. Model tests and simulation were carried out on Louisiana Optical Network Initiative (LONI)'s QB2 and NCAR's Cheyenne super computers. In a serial mode, it took 2.5 h to conduct a 2-year simulation of streamflow with sediment module deactivated, and 30 min to finish a 24 h sediment simulation.
