3.2.1 Verification against stream gauges
Gauged water heights from five USGS stream gauges within the model
domain are retrieved during model simulations at the 15-min interval.
Surface water heights from two simulation schemes are extracted at
collocated stream gauge locations. It is worth mentioning that the
terrain elevation imposes great uncertainties when comparing model
simulations to observations, as the sub-grid variation cannot be
resolved in the current settings. Despite the resolution mismatch, these
gauge readings are still the most widely used source to verify the model
performance. Table 4 shows the respective performance for with and
without re-infiltration scenarios with respect to observations. The
re-infiltration scheme greatly improves the NSE scores (+139.9%) and CC
(+7.24%) while reducing RMSE (-18.2%). Especially for the gauge
08075900, there is more than a 400% increase in NSE score, jumping from
0.12 to 0.69. By breaking down the RMSE into systematic error
RMSES and random error RMSER, we see the
reduced error is largely attributed to minimized systematic error
(-31.2%), relative to the random error (-13.1%). The systematic bias
is much alleviated by considering the re-infiltration scheme.
The reason for such a leap comes from better characterization of its
flow recession limbs, as shown in Figure 6. Both schemes are capable of
simulating the peak water height values without delays, but water in the
scenario without re-infiltration falls mildly in the recession stage,
resulting in much higher water height than the observations. On the
other hand, flow for the re-infiltration scenario follows the gauge
readings closely, especially after the first peak (from 2017-08-26 to
2017-08-27). Apart from this best-performing gauge, the re-infiltration
scheme improves capturing falling water across all the gauge stations,
thereby leading to significant performance gains. Consistently, previous
studies also highlighted that the re-infiltration markedly reduces
recession limbs in the hydrograph (Nahar et al., 2004).
[INSERT TABLE 4 HERE]
[INSERT FIGURE 6 HERE]