3.2.2 Verification against High Water Marks
Because a direct assessment for flood inundation is not feasible, some watermarks or stains in the aftermath of a flood event can be used as proxy data for model evaluation. The USGS team routinely publishes their surveyed High Water Marks (HWMs) after some major flood events, which can subsequently be used for model evaluations (Chen et al., 2021; Li et al., 2021b; Sebastian et al., 2021; Wing et al., 2017). Figure 7 shows the cell-wise maximum flood depth of the two schemes compared to the HWMs. Both schemes present better performance upstream of Halls Bayou, with a difference smaller than 0.5 meters. However, the model over-predicts water depth in Greens Bayou up to 1.5 meters. This is consistent with the over-prediction of in-channel water depth, as shown in Figure 6. The distribution of the differences is shown in Figure 7c, pointing to a generally better performance of the re-infiltration scenario than without it, as the absolute mean depth difference of the re-infiltration (0.51 m) is 17.2% smaller than that of the scenario without re-infiltration (0.60 m). It is worth noting that HWMs themselves come with uncertainties that are due to the data quality, and errors could be up to 0.2 meters (Koeig et al., 2016). For instance, tranquil water represents a smooth trend that has small uncertainties. There are also spurious errors that are related to human mistakes or values being rounded off. This is particularly true for the recorded geographical coordinates which requires more floating points to pin down the location exactly. In Figure 7d and e, the two HWMs are marked with an absolute difference greater than 1 m but only several pixels away (i.e., tens of meters) from their true values. Despite this, the re-infiltration greatly alleviates the over-prediction of the previous model.
[INSERT FIGURE 7 HERE]