**3. Results**

#### *3.1. Rainfall Data*

Total event rainfall at the rain gauge in LLCW (RG.HM, Figure 1) correlated closely with daily rainfall at nearby stations in the United States (Figure 5). For the events when rainfall data were recorded for the LLCW watershed at RG.HM (2013–2017), the gauge at San Diego Brownfields (SDBF) has the highest correlation coefficient and smallest RMSE out of the stations with good data availability. Rainfall at RG.HM was higher than that at all other stations for larger events (> 60 mm), but matched

the SDBF data well for rainfall between 10 and 50 mm (Figure 4). The SDBF gauge had a higher correlation coefficient and lower error compared to stations closer to LLCW in the Tijuana Estuary (IB3.3). Therefore, SDBF was used to estimate rainfall in LLCW for years when no data was available at RG.HM.

**Figure 5.** Event-total precipitation at the Hormiguitas rain gauge (RG.HM) versus three other nearby stations (see Figure 1). The dashed line is the 1:1 line. Taken from Biggs et al. [33].

#### *3.2. Rainfall-Runo*ff *Relationships*

Event-total rainfall for the 14 events with rainfall (P) and runoff (Q) data ranged from 7 to 83 mm (Table 5). The event-wise runoff coefficients (Q:P) ranged from 0.02 to 0.67. Event-total runoff increased with event-total rainfall and fits a watershed-mean SCS CN of 80–90 (Table 5 and Figure 6). The highest SCS CN occurred for the smallest events and CN generally decreased with the event size (Figure 6). This was consistent with runoff production from surfaces with low infiltration capacity during small events, and from all surfaces, including those with high infiltration capacities, during large events. The largest event (rainfall 81 mm) has a runoff coefficient of 0.51, where most points fell between SCS CN 80 and 90 (Figure 6), which is consistent with literature values for partially urbanized land cover [45]. Thus, no adjustments were needed for the CN as the fit was adequate with the observed storm-wise rainfall-runoff relationships (Figure 6). The 24-hour rainfall distribution used for most of the simulated storms was type II [27] because it is representative of semi-arid regions of South-western USA, and matches the most frequent storm type calculated using rain gauge measurements in the LLCW [33]. Some storms were assigned different storm types based on their rainfall distribution compared to SCS storm types.

The RMSE of the simulated storm-wise runoff was 6.6 mm (89% of mean), and 13 m3·s−<sup>1</sup> (177% of the mean) for peak runoff. The RMSEs were notably influenced by a single large storm of 81 mm total-event precipitation (27 February 2017, Table 5). RMSE without that large storm was 3.6 mm (75% of the mean) for total runoff, and 6.9 m3·s−<sup>1</sup> (105% of the mean) for peak runoff. The AnnAGNPS model was most accurate for medium-sized events (event precipitation between 2 and 20 mm, Figure 7), which are the most frequent events. Therefore, we did not calibrate the model to minimize the error. Peak discharge was generally underestimated for small storms and overestimated for large storms [33].


**Table 5.** Summary of storm events used for model calibration.

**Figure 6.** Rainfall-runoff relationship for all observed storm events, with several SCS CN rainfall-runoff relationships, in non-log (top) and log-log (bottom). Taken from Biggs et al. [33].

**Figure 7.** Relationship between observed and simulated total (a, mm per storm) and peak (b, m3/s) runoff. The solid lines are the 1:1 lines.

#### *3.3. Simulated Sediment Production*

The SSC samples were collected from sub-watersheds with different soil characteristics. Fractional covers of soil types in the sub-watersheds draining to each SSC sample location vary from 20% to 100% erodible, non-cobbly soils (Lf soil type) (Figure 8). The observed SSC of the storm-water samples correlated with the AnnAGNPS-simulated SSC (Figure 8b), and modelled SSC correlated with the fraction of the sub-watersheds covered by Lf (Figure 8a,c), which highlight the influence of soil properties on the modelled sediment production. The modelled SSC values were higher than the observed SSC values (mean model = 210.7 g/L, mean observed = 48.7 g/L, RMSE = 203.5 g/L) because the grab samples were not all taken at the time of the peak discharge. No samples were available for areas drained by cobble and gravel soils in the southern and northern parts of the watershed.

**Figure 8.** (**a**) Geographic location of storm-water samples, suspended sediment concentration (SSC), and soil types along the LLCW. (**b**) Relationship between observed and simulated SSC (the black line is the 1:1 line) and (**c**) relationship between Las flores (Lf) soil fraction and simulated and observed SSC.
