Standards dramatically advance streamflow and flood forecasting in US and elsewhere – Part 2 of 5

Contributed by: 
David K. Arctur, Research Scientist and Fellow, University of Texas at Austin; David Gochis from NCAR, and Ahmad Tavakoly from USACE ERDC


In my previous segment, I introduced the U.S. Open Water Data Initiative (OWDI) and the National Flood Interoperability Experiment (NFIE). This part of my water data series is about the NFIE project in 2014-2015. Following the dedication of the new National Water Center (NWC) in May 2014 by the National Weather Service (NWS) of the National Oceanic and Atmospheric Administration (NOAA), Dr. David Maidment recruited initial funding support  to start working on NFIE from the National Science Foundation (NSF) through a research grant under the EarthCube initiative, called the EarthCube Building Block for Integration of Discrete and Continuous Data (DisConBB). This project was begun several months earlier to develop methods and tools for relating time series observations of water resources at point locations (discrete data) with gridded data of land surface characteristics (continuous data).  In terms of relevant OGC standards, a key task was to convert time series data between WaterML 2.0 and netCDF-CF content models and formats. NFIE became the “use case” focus of DisConBB, providing a complex and critically-needed application of the DisConBB research objectives. DisConBB then organized a workshop in Fall 2014 that served as a kickoff for NFIE, drawing in nearly twenty organizations including Federal agencies, academia, and commercial companies. This meeting outlined the work plan for the next year, which Dr. Maidment filled out and published in February 2015 [1]. At this workshop, the NWS agreed to help support this development. A second DisConBB-hosted workshop was held in Spring 2015 at the NWC in Tuscaloosa, Alabama. By this time, the prototype workflow was proving capable of sustaining an operational schedule (more on this below). So this second workshop organized a seven-week Summer Institute to engage a broader community of graduate students and faculty to evaluate and refine the NFIE workflow and generate useful products. The last section below summarizes the outcomes of this Summer Institute.  

With that context, we’ll now look at the technical challenges and capabilities of this remarkable project. 

Caveat: The emerging operational National Water Modeling framework for streamflow forecasting within the U.S. National Weather Service builds on the NFIE model, but diverges in some ways, mainly due to requirements and procedures for operational forecasting and data management. So I want to emphasize that this story is about the NFIE design and capabilities, and should not be taken as a description of the NWS National Water Model. 

U.S. National Weather Service Today

The NWS has 13 River Forecast Centers covering about 6600 basins, as shown in the map below. National river conditions are published through the NWS Advanced Hydrologic Prediction Service (AHPS).  

Figure 1. NWS River Forecast Centers 

These River Forecast Centers (RFC): 

  •  Prepare river and flood forecasts using models;
  •  Provide forecast guidance to Weather Forecast Offices (WFOs);
  •  Issue daily stage and streamflow forecasts, rainfall and drought data and information, and flash flood guidance;
  •  Work with water managers and other Federal agencies. 

A majority of these basins have a Forecasting Point (FP), which is often associated with a stream gage that is part of the USGS National Water Information System (NWIS) or other streamflow telemetered network. 

This represents a tremendous amount of work, yet these basins do not include many U.S. coastal areas, due to complexities of modeling the interfaces between surface water and seawater.  This means urban areas that are home to some 100 million people in the U.S. are not covered by the NWS river forecast systems. 

With the new National Water Center in place, the NWS will compliment its existing river forecasting activities with a centralized, national water forecasting capability on a continental scale, including coastal areas. Moreover, it will provide a service to evaluate RFC and NWC models and forecasts, and support a modeling testbed for new NWC capabilities. How will all this happen? 

Stream Network and Land Surface Data

The following graphic shows the primary geospatial framework needed and used in NFIE. NHDPlus version 2 is a suite of geospatial products integrating the National Hydrography Dataset (NHD) stream network (1:100,000-scale), the Watershed Boundary Dataset (WBD) hydrologic units (12-digit), and the National Elevation Dataset (NED) topography at 30m resolution or better. At this scale, NHDPlus has about 2.7 million “stream reaches” (stream segments as defined in NHD) and catchments, averaging 2 km in stream length and 3 km2 in catchment area, though these numbers vary quite a bit by region. 

Figure 2. NHDPlus geospatial data layers used in NFIE

NHDPlus was jointly developed by the U.S. Geological Survey (USGS) Geospatial Program and the U.S. Environmental Protection Agency (EPA) Office of Water. The key feature of NHDPlus that enables the modeling workflow NFIE uses for streamflow forecasting, is that each stream reach is uniquely related to a single catchment area. The catchment area is also integrated with the National Land Cover Dataset (NLCD) which is maintained by the Multi-Resolution Land Characteristics Consortium (MRLC). The NLCD provides part of the information needed for the land-surface modeling. 

NFIE used the above information, as well as these geospatial data layers: NHD flowlines, waterbodies (lakes, etc.), sinks (locations where a stream leaves the ground surface), catchments and subwatersheds; USGS gages; AHPS forecast points and basins; and state and county boundaries. These can be overlaid on a user-defined basemap, e.g., to include major roads and other topographic or cultural features. 

Predicting Rainfall, Runoff and Streamflow 

The next figure shows a simplified prediction workflow:  1) feeding rainfall forecasts to the land-surface modeling framework; 2) converting the land-surface runoff from regular grid boundaries to NHDPlus catchments, and 3) generating streamflow forecasts for the 2.7 million stream reaches across the continental U.S. 

Figure 3. Forecasting Rainfall to Streamflow

It’s beyond the scope of this article to describe these models in detail, but the following outline provides the basic ideas and links for more details.

  1.  One of the NOAA/NWS weather model forecasts is for near-term use, 0-15 hours, with 3km resolution, called the High-Resolution Rapid Refresh (HRRR). This was run once every three hours. (By summer of 2016, NWC is planning to support additional time frames for streamflow forecasting ranging from a few hours to 30 days.)
  2.  The precipitation output from HRRR was fed into a land-surface model called Noah-MP. Noah-MP predicts runoff, soil moisture, and many other land-surface parameters. 
  3.  The 3km gridded estimates of runoff from Noah-MP were converted to the NHDPlus catchment boundaries using an area-weighted computation. 
  4.  The runoff data and NHDPlus stream network were then fed into the streamflow routing model, RAPID (River Application for Parallel ComputatIon of Discharge), which estimated the streamflow in cubic feet or meters per second, for each of the 2.7 million NHDPlus stream reaches. 

Noah-MP and RAPID were integrated through the WRF-Hydro (Weather Research and Forecast – Hydrological) framework. This took a combination of efforts and accomplishments including extensive cloud usage, high-performance computing, consultation with XSEDE experts, and a lot of attention to performance. As a result, depending on the number of processors used, the workflow from ingest of HRRR precipitation forecasts to output from the combined WRF-Hydro/RAPID system could be run in 10-12 minutes, for the 2.7 million stream reaches in the continental U.S., for each 15 hour forecast

The overall workflow was led by the National Center for Atmospheric Research (NCAR), with contributions from a number of partners and institutions.  One data source used for evaluation of the model results was the USGS NWIS network of streamflow observations [2]. Additional weather models were experimental configurations from the NOAA National Centers for Environmental Prediction (NCEP), along with the NOAA Severe Storms Laboratory national radar product Multi-Radar/Multi-Sensor (MRMS).  Those meteorological forcing data were transferred to NCAR, where they were combined and reformatted for input into the WRF-Hydro system.  Once complete, these forcing data were transmitted to the University of Texas Advanced Computer Center (TACC) where the WRF-Hydro/NoahMP and RAPID models were executed.  Distributed processing of some of the data was also enabled through the Microsoft Azure Cloud.  Throughout the workflow, data inputs and exchanges among the computational models was handled with either the OGC WaterML Encoding Standard or the OGC netCDF-CF Standard, in some cases converting between the two formats, thus realizing in a big way the research objectives of the EarthCube DisConBB project. In total the NFIE 2015 effort demonstrated the viability of a truly distributed and heterogeneous computing activity.

Predicting and Mapping Flood Inundation

An important target community for this streamflow forecasting capability would be local emergency management, in particular for planning, mitigation, and response to major storms and flooding. Emergency responders need maps showing areas and buildings likely to be inundated, not just runoff and streamflow estimates. 

With the continental-scale prediction of streamflow now being made operational, inundation mapping will be in the next push. A number of approaches are being considered for implementation this year. One approach developed at the U.S. Army Corps of Engineers (USACE) Engineer Research and Development Center (ERDC) was to couple the RAPID model with a regional-scale flood delineation model called AutoRoute [3] to estimate flood extents. The coupled tool, called AutoRAPID [4], can quickly and efficiently simulate flood extents using a high resolution dataset (~10m resolution) at the regional scale (> 100,000 km2). In the AutoRAPID modeling framework, the RAPID output peak flows in each river reach are the primary inputs to the AutoRoute model, which calculates the normal flow depth at a number of stream cross-sections. AutoRoute works with the NHDPlus river network, along with elevation and land cover data. AutoRAPID was recently tested for a 230,000 km2 area (8% of the Mississippi River Basin, including parts of Indiana, Wisconsin, Illinois, Iowa, Missouri, and Kentucky). Discretizing the 230,000 km2 area into 1° by 1° tiles, a flood inundation raster was simulated in under 12 minutes for each tile. The model results are generally in agreement with observed flow and flood inundation levels, and suggest that the AutoRAPID model can be considered for several potential applications requiring streamflow and flood inundation forecasts. This can give important lead time for emergency response teams to protect or evacuate large areas, especially in the absence of stream gages.

Other approaches are being tried as well, but these are not complete enough to report yet. 

NFIE Summer Institute 

From June 1 to July 18, 2015, the NFIE Summer Institute was conducted at the NWC, hosted by the NWS and organized by the Consortium for the Advancement of Hydrologic Science (CUAHSI). Numerous students and faculty from across the U.S. and even some foreign countries attended to work on key themes related to the technical issues addressed by NFIE. The results of these investigations were presented during the 2015 NFIE Summer Institute capstone event at the 3rd CUAHSI Conference on Hydroinformatics. A group of papers from this have been submitted for review for a special theme issue of the Journal of the American Water Resources Association (JAWRA). 

NFIE will continue in 2016 with the formation of a new Summer Institute for hydrologic and hydraulics research in support of the NWS emerging public National Water Model. This is just now in planning stages, check for updates. 

The next segment in this water data series will focus on outreach and capacity building efforts in 2015 to bring this capability to other countries, in particular developing countries. 


Deep gratitude goes to those who were involved in development of this workflow, and review of this report: David R. Maidment, Edward Clark, David Gochis, Cedric David, Jim Nelson, Nawajish Noman, Peirong Lin, Kevin Smith, Fernando R. Salas, Marcelo Somos-Valenzuela, Ahmad A. Tavakoly, Zong-Liang Yang, Cassandra Fagan, Xing Zheng, Justin Yu, Gonzalo Espinoza, Cyndi Castro, and Tim Whiteaker. Many others contributed through CUAHSI, the University of Alabama, and numerous other institutions mentioned in this article.   


[1] David R. Maidment, A Conceptual Framework for the National Flood Interoperability Experiment, February 9, 2015. URL:

[2] Peirong Lin, Mohammad Adnan Rajib, Marcelo A. Somos-Valenzuela, Zong-Liang Yang, Venkatesh Merwade, David R. Maidment, Yan Wang, Li Chen. Spatio-Temporal Evaluation of Evapotranspiration and Streamflow in the NFIE Modeling Framework. Submitted to JAWRA featured collection, Open Water Data Initiative & National Flood Interoperability Experiment. Back

[3] Michael L. Follum. AutoRoute rapid flood inundation model. Vicksburg, Miss: U.S. Army Engineer Research and Development Center. Report ERDC/CHL CHETN-IV-88, March 2013. URL: 

[4] Michael L. Follum, Ahmad A. Tavakoly, Jeffrey D. Niemann, and Alan D. Snow. AutoRAPID: A Model for Prompt Estimation of Streamflow and Flood Inundation Mapping Over Regional to Continental Extents. Submitted to JAWRA featured collection, Open Water Data Initiative & National Flood Interoperability Experiment. Back