Reducing a Stream Networks Horton-Strahler Stream Order Improves the Skill of Flood Inundation Maps from Height Above Nearest Drainage Method

The National Water Model (NWM) currently requires the post-processing of forecast discharges to produce forecast flood inundation maps (FIM) to support the National Weather Services mission of protecting life and property. Height Above Nearest Drainage (HAND) is a means of detrending digital elevation models (DEM) by normalizing elevations to the nearest, relevant drainage line (creeks, rivers, etc). Its worthy of producing high-resolution FIMs at large spatial scales and frequent time steps using reach-averaged synthetic rating curves. However, HAND based FIMs suffer from a known limitation caused by independent catchments that lack the ability to cross catchment boundaries and ridgelines. To counter this constraint, a version of HAND known as Generalized Mainstems (GMS) is proposed that reduces the Horton-Strahler stream order of the stream network. To demonstrate skill enhancement, we constructed HAND derived at three different stream resolutions including the NWM full resolution (FR), the NWM mainstems (MS) resolution, and the NWM GMS resolution stream networks. The FR stream network contains all NWM forecast locations and the MS resolution stream network contains all river segments at or downstream of NWS river forecast points. GMS contains all segments within the FR stream resolution but instead of deriving HAND by accounting for all river segments at once, it is derived independently at the level path (LP) scale. LPs are unique identifiers propagated upstream from a sub-basins outlet along the direction of maximum flow distance and repeated recursively until all segments are assigned LP identifiers. These serve as processing units for HAND dataset production and FIMs are made at the LP scale. These FIMs are then mosaiced together, effectively turning the stream network into discrete groups of homogenous unit stream order by removing the influence of neighboring tributaries. Improvement in mapping skill on the order of 2% points of Critical Success Index for MS and 2% points more for GMS is demonstrated by comparing to HEC-RAS FIMs. Additionally, both Probability of Detection and False Alarm Ratio improve which can be partly explained by a positive correlation of stream order with river stage at fixed discharge values within the synthetic rating curves produced by HAND.