H111-0005
Toward Improving Parameter Regionalization for the National Water Model Using the CAMELS Dataset

Friday, 11 December 2020
Poster
Yuqiong Liu1,2, Arezoo Rafieeinasab3, Xia Feng4,5, Wanru Wu1, Brian A. Cosgrove6, Aubrey L Dugger7 and David Gochis8, (1)NOAA/NWS/OWP, Silver Spring, MD, United States, (2)LEN Technologies, Oak Hill, VA, United States, (3)National Center for Atmospheric Research, Boulder, CO, United States, (4)NOAA/NWS/NWC, Tuscaloosa, AL, United States, (5)University Corporation for Atmospheric Research, Boulder, CO, United States, (6)NOAA/NWS Office of Water Prediction, Silver Spring, MD, United States, (7)University of California Santa Barbara, Santa Barbara, CA, United States, (8)NCAR, Boulder, CO, United States
Abstract:
Parameter estimation represents a critical and challenging task in large-scale distributed hydrologic
modeling. For the National Water Model (NWM), which runs operationally over the contiguous United
States (CONUS) at a 1-km resolution, parameter estimation adopts a two-step strategy: 1) the model
parameters are calibrated for selected basins with minimum human interference and reliable
streamflow observations, and 2) the calibrated parameters are applied to uncalibrated areas of the
CONUS domain via regionalization, based on hydrologic similarity characterized by various physical
attributes. It is therefore crucial that the chosen attributes adequately describe the dominant hydrologic
processes in the region.


We present an intercomparison study where the characterization of hydrologic similarity is based on
three different scenarios (with the first two used in the previous and current versions of the NWM): 1)
US Environmental Protection Agency Ecoregions that denote areas of similarity in major ecosystem
components; 2) hydrological landscape regions (HLRs) that define similarity with attributes of land
surface form, geologic texture, and climate characteristics, and 3) a new scenario with similarity defined
by a wide range of catchment attributes and streamflow signatures from the Catchment Attributes and
Meteorology for Large-Sample Studies (CAMELS). The analysis is conducted over 370 CAMELS basins
with a take-one-out approach: for each basin, a donor basin that is “closest” to the given basin is
identified from the remaining basins either based on ecoregions (Scenario 1) or the Gower’s distance
measure computed from the chosen attributes (Scenarios 2 & 3). The calibrated model parameters from
the chosen donor basins are then directly transferred to the receiver basins.


Analysis of streamflow simulations for the period of 2008-2016 indicates that, while all three
regionalization scenarios expectedly outperform the uncalibrated model and underperform the
calibrated model, regionalization based on CAMELS is shown to improve considerably upon the
approaches based on ecoregions or HLRs. This is likely due to the inclusion of a wider range of basin
attributes and streamflow signatures in the CAMELS dataset, which are being considered for
implementation in the future versions of the NWM.

<< Previous Abstract | Next Abstract >>