2.3.2. Sediment

Data on sediment removed from the sediment traps at the LLCW outlet (Figure 1) were used for calibration. Both upper and lower traps were excavated in the Spring and Fall of 2005, Winter 2006, and each Fall from 2007–2012 (*N* = 7). The sediment trap efficiency, or the proportion of the total sediment trapped in the sediment basin, was calculated based on Morris and Fan [46] and Urbonas and Stahre [47]. See Biggs et al. [33] for a detailed description.

Preliminary simulations suggested that the channel erosion module of AnnAGNPS resulted in excessive sedimentation in the channels, which was not observed in the field. We, therefore, calculated the simulated sediment load as the total amount of sediment by source (sheet and rill, and gully erosion) that makes it to the stream channel network (Figures 1 and 4), plus channel erosion estimated in previous work [38]. The load was compared to the total sediment load with the total amount of sediment being excavated from the sediment traps for specific dates. The stream channel network was defined as permanent channels within the watershed based on field observations and visual examination of these channels using high resolution imagery. Channel erosion estimates (t/yr) were taken from Taniguchi et al. [38], who calculated channel erosion from the difference between the cross sections observed in 2014 with the cross section under reference (pre-urban) conditions, which was divided by the time since urbanization. Taniguchi et al. [38] estimated that channel erosion accounted for 25% to 40% of total sediment yield to the estuary over 2002–2017. In this scenario, we estimate channel erosion by multiplying the hill-slope erosion estimated by AnnAGNPS by 0.33 and 0.67 to get channel contributions of 25% and 40% of total load, and adding that load from channel erosion to the hill-slope load to get the total load.

The data from the sediment trap, corrected for trap efficiency, were compared with AnnAGNPS simulation results of total load, including both hill-slope and channel erosion. Critical shear stress τ*<sup>c</sup>* and sediment delivery ratio (SDR) were then calibrated to match the observed sediment yield at the LLCW outlet. An initial value of <sup>τ</sup>*<sup>c</sup>* was set to 1.6 N·m−<sup>2</sup> for sandy soils based on the average value from nine samples collected on the Lf soil type [14]. Initial values of τ*<sup>c</sup>* for conglomerate soils were taken from USGS [37] dataset for fine cobbles (64 N·m−2) and were modified during calibration to <sup>τ</sup>*<sup>c</sup>* = 32 N·m<sup>−</sup>2, which corresponds to very coarse gravel. The parameters used to calibrate sediment yield are presented in Table 3. The SDR for coarse soil formations (CfB and CbB) were calculated internally by the model. The SDR for the Lf type was set to 1 and based on field observations of extensive rill and gully formation, which results in the delivery of most sediment from sheet and rill erosion to the channel network.

**Table 3.** AnnAGNPS soil erosion parameters used to calibrate the watershed scale model for sediments.


The percent bias (PBIAS) was used as a measure of the average tendency of the simulated results relative to the observed data, which indicates over (positive PBIAS) or underestimation (negative PBIAS), respectively [48,49]. The PBIAS was calculated using the equation below.

$$\text{PBIAS} = \frac{\sum\_{i=1}^{N} (observed - simulated) \times 100}{\sum\_{i=1}^{N} observed} \tag{2}$$

where *i* is the index of the storm events and '*N*' is the number of events (14).

The measurements of sediment accumulation at the outlet provides an aggregate measure of sediment load for the watershed, but does not validate the spatial pattern of sediment load from different soil and land use units in the watershed. Grab samples of water were collected for suspended sediment analysis at 10 sites in the watershed during a large storm (81 mm, total depth) on 27 February 2017. All samples were collected over a 0.5 hour period, which corresponded to a period of maximum runoff. The observed SSC of the storm-water samples were then compared with the simulated AnnAGNPS SSC (SSC = storm event sediment mass/storm event runoff volume) to explore the influence of soil properties and land use on sediment production in the watershed. While SSC at a given location changes during an event, the samples were collected during similar hydrological conditions, and provide a snapshot of the spatial variability of SSC during an event. Table 4 summarizes the data type and parameters set for model calibration and evaluation.

**Table 4.** Field data collection and time periods for model calibration.


We used the entire dataset of observations at the outlet, including annual sediment accumulation in the traps (N = 6) and event runoff (N = 14), for model calibration due to the small number of observations. Use of an entire dataset for calibration is consistent with other AnnAGNPS applications in Mediterranean environments such as References [50,51].
