Figure 4 : Leave-one-out cross validation to assess A, C) Na+ and B, D) NO3-model performance. A-B) Model predictions are plotted against observed values for both MLR (open triangles) and SSN (black circles) models. C-D) Prediction standard error is plotted against relative watershed area, with headwater positions on the left side of the plot and lower watershed positions on the right side.

3.5 Longitudinal patterns across two watersheds with inverse burn patterns

The two paired watersheds with inverse burn patterns exhibited distinct patterns in stream NO3- concentration. 72% of Brush watershed was burned and most of the burn occurred in the upper half of the watershed (Figure 5A). Mean NDMI was generally low throughout Brush, but was inversely related to burn extent (Figure 5C). Stream NO3- concentrations spanned a 4.6 mg/L range throughout Brush Creek. The minimum concentration (0.4 mg/L) occurred at the upper most sampling location and the highest observed concentration (5.0 mg/L) occurred nearby within the upper watershed (Figure 5E). Nitrate generally declined in the lower half of the watershed and reached 0.9 mg/L at the lowest sampling location. Conversely, the majority of the burned area in Pine Creek occurred in the lower watershed (Figure 5B). Burn extent was again inversely related to mean catchment NDMI, but NDMI remained relatively high throughout (Figure 5D). Pine stream NO3-concentration increased gradually downstream from below detection levels in the headwaters to 0.3 mg/L at the outlet (Figure 5F). Maximum and mean stream NO3 concentrations was 14- and 17-times higher in Brush than Pine. Our NO3- SSN model predictions agreed with measured stream NO3- concentrations during our 2019 sampling (Supplemental Figure 4).