Abstract
We propose a method to describe the impact of dams on design floods for ungauged areas and validate the method over the conterminous US (CONUS). A Random Forest (RF) model was chosen to capture the relationship between the change in 100‐year return period flow up‐ and downstream of different dams and dam parameters available in the Global Reservoir and Dam (GRanD) database. The results showed that: (1) the RF model showed a greater accuracy in terms of Nash Sutcliffe efficiency coefficient (0.92 in training and 0.88 in testing) than a benchmark Multiple Linear Regression model (0.68 in training and 0.61 in testing); (2) Dam inflow, upstream catchment area, and long‐term average discharge at reservoir location were the three most important factors for dam outflow; (3) flood attenuation effect indices (FAI) for >1400 dams over the CONUS were derived with the proposed method. To further validate the accuracy of the FAI, a new module considering flood attenuation effects was developed for the LISFLOOD‐FP hydrodynamic model and two dams in the CONUS were selected to compare simulated flooded area against Federal Emergency Management Agency flood hazard maps. The result showed that the overestimation in flood hazard maps caused by not taking dams into account can be significantly corrected using the FAI and the enhanced LISFLOOD‐FP model. We conclude that the proposed methodology is a valid approach to describe the impact of dams on design floods, thereby improving the accuracy of flood hazard maps, especially in ungauged areas.
Original language | English |
---|---|
Article number | e2019WR025380 |
Number of pages | 15 |
Journal | Water Resources Research |
Volume | 56 |
Issue number | 3 |
Early online date | 23 Mar 2020 |
DOIs | |
Publication status | Published - 27 Mar 2020 |
Keywords
- design flood
- dams
- continental U.S
- flood hazard
- ungauged areas
- machine learning