Spatio-temporal hierarchical modeling for Improved National Water Model Predictions
Abstract
The NOAA's National Water Model (NWM) is a process-based hydrologic model operating in real-time and providing river forecasts at over 3000 locations over the entire continental United States (CONUS). However, conditional and unconditional biases exist in the NWM simulated records, particularly for high flows. To improve daily mean riverflow predictions across CONUS, we have developed a post-processing scheme that develops three spatio-temporal hierarchical models (HMs)over a fixed 10-day window between the NWM and observed daily flows considering geographic, hydroclimatic, anthropogenic, and basin characteristics as predictors. These models were calibrated (1993-2008) and validated (2009-2018) using all flows for different basin types (i.e., controlled, natural, and coastal sites). The validation results for modeled high-flows (>67% normality) against the NWM estimations suggest that the HMs improved all sites median Nash-Sutcliffe efficiency (NSE) from 0.37 to 0.42 and 66.3% of all sites mean NSE increased over time. A 20-fold cross-validation was also applied to demonstrate model performance over space, by leaving 5% of sites out as ungauged locations. Analysis of HMs performance in predicting high flows show HMs improved all sites median NSE to 0.57, and 68.1% of sites mean NSE. The geographic and hydroclimatic analysis showed that adjusted flows in eastern and colder regions sites improved more than sites in the western and warmer regions. Among basins with significant anthropogenic influence (e.g., downstream of a reservoir) , natural sites median NSE improved more than 0.1. However, coastal sites appeared to have only a moderate improvement over space and time. Seasonal analysis indicated that the most influenced predictors in controlled and natural sites are the previous 3-day mean riverflow and the aridity index, respectively. Thus, the proposed space-time hierarchal modeling framework has a great potential in improving real-time streamflow forecasts to be issued by the NWM.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFM.H22A..01F