Improving the land-surface components of Climate Forecast System Reanalysis (CFSR)


NOAA Award: NA13OAR4310107

University of Washington

PI: Bart Nijssen

Reporting Period: 07/01/2014 - 06/30/2015

Report date: 06/30/2015

This is a joint project between the NOAA National Centers for Environmental Prediction (NCEP) Environmental Modeling Center (EMC), NOAA’s Climate Prediction Center (CPC), Princeton University, and the University of Washington. The overall intent of the project is to utilize past land data assimilation system activities (notably NLDAS and GLDAS) to improve upcoming reanalyses. Specific objectives are to improve (1) land characterization data sets (e.g. vegetation type and soil texture class, and the characterization of urban areas, etc.), (2) atmospheric forcing data sets (e.g. precipitation, downward solar and longwave radiation), (3) assimilation of near-real time land states (e.g. surface skin temperature, albedo, soil moisture, snow extent, vegetation greenness and density), (4) land-model spin-up procedures, and (5) downscaling techniques for forcing data and land states. The University of Washington has a modest advisory role in the project. As defined in the proposal, the University of Washington is to provide guidance and support … on the implementation of the river routing model into CFSR as well as calibration and validation of the model at a global scale. This work will directly build on current efforts by the UW to implement the same routing model as an option in CESM as part of their collaborative DOE proposal: “Improving Decadal Prediction of Arctic Climate Variability and Change Using a Regional Arctic System Model (RASM)”. CFSR routing results will be compared with offline global simulations from the VIC land model, which uses the same routing model, but different land surface forcings. In addition, snow and soil moisture data fields from the UW’s multi-model surface water monitor will be used to evaluate terrestrial water cycle changes in the CFSR simulations.

During Year 2 we have focused on further development of the routing model, hereafter referred to as RVIC, and integration of RVIC in a coupled regional climate model (RASM) to provide freshwater inputs to the Arctic Ocean. In addition, we have created an off-line version of the model in python, which is also used to provide some of the inputs that are used in the coupled version of RVIC. The offline version of the model is shared on a public code repository (github) and we have improved the model documentation. For details, see below.


The RVIC streamflow routing model is an expanded version of the routing model originally introduced by Lohmann et al. (1996). RVIC is a semi-distributed, source-to-sink, routing model that solves a linearized version of the Saint-Venant equations. The model develops impulse response functions (IRFs) to define the time evolution of flow between pairs of source and sink points. Following the development of the IRFs, a simple convolution is performed combining the IRF for each source to sink pair and the runoff flux from each source point (Fig 1. bottom). RVIC has been configured to run as a stand-alone routing model (e.g. as a post processor of semi-distributed hydrologic models) or coupled within earth system models (e.g. NCAR’s Community Earth System Model).

Compared to the original Lohmann et al. (1996) model, RVIC offers the following improvements:

  1. Development of the IRFs (Fig. 1 top) is done as a separate preprocess. In the original model, the development of the IRFs was done each time the model was executed. For short simulations, or simulations that required frequent restarts or checkpointing, the computational cost of the developing the IRFs limited the model’s applicability in coupled models. In the case of RVIC, the IRFs are developed prior to the beginning of the coupled model simulation and are saved in a netCDF format for use in subsequent simulations.

  2. Impulse response functions may be developed at high-resolution (e.g. 1/16\(^{\circ}\)) before being upscaled to the land surface model grid (e.g. CFSR’s T382 or RASM’s 50-km near equal-area). A major limitation of the original model was the requirement that a flow direction raster needed to exist on any potential land surface model grids. RVIC adds the ability to develop the IRFs using existing, high-resolution, flow direction rasters (e.g. Wu et al., 2011) before using a conservative remapping technique (Jones 1999) to upscale the IRFs to the native grid of the land surface model. This feature 1) eliminates the need to develop flow direction rasters for many spatial grids, and 2) preserves the high-resolution flow characteristics of the initial flow direction raster.

  3. Both the stand-alone and coupled versions of RVIC have bit-for-bit checkpoint and restart capability. This feature allows RVIC to be used in forecasting and climate model simulations that require frequent restarts.

Schematic of the RVIC streamflow routing model. The three major steps of the RVIC model are shown as 1) Development of the impulse response functions (IRFs; top row), 2) Upscaling the IRFs (middle row), and 3) flow convolution using runoff fluxes from the land surface model.

RVIC in the Regional Arctic System Model

The Regional Arctic System Model (RASM) is a high-resolution, fully-coupled, earth system model, applied over the pan-Arctic domain (Fig 2.). The RASM project, which has been supported by the United States Department of Energy (DOE) and the Department of Defense (DOD), aim to improve the representation of high latitude processes in the Arctic region and to improve multi-decadal simulations of Arctic climate. RASM uses the coupling infrastructure from the National Center for Atmospheric Research’s (NCAR) Community Earth System Model (RASM) to combine individual atmosphere (WRF), ocean (POP), sea ice (CICE), land (VIC), and streamflow (RVIC) model components. RVIC has been coupled within RASM to deliver the freshwater flux (streamflow) from the land to the ocean.

A) Coupling schematic for the Regional Arctic System Model. B) Model domain for the Regional Arctic System Model.

In the past year, the initial configuration of RVIC within RASM has been finalized and fully-coupled RASM simulations now use RVIC as the default streamflow routing model. As applied in RASM, we develop RVIC IRFs on a 1/16\(^{\circ}\) flow direction raster from Wu et al. (2011) which are upscaled to the 50-km near equal area land-atmosphere grid. RVIC is coupled to the CESM flux coupler ever 20 minutes and routes streaflow from all land grid cells draining to the inner ocean domain of the regional model (see Fig. 2b). Figure 3 shows the combined monthly mean hydrograph for all coastal grid cells in the RASM domain north of 60 degrees N.

Total RASM simulated coastal freshwater flux north of 60 N\(^{\circ}\) compared to stand-alone VIC and CORE2 simulations.

RVIC development

Recent development of the RVIC model has focused on the following items:

  1. Introduced support for Python 3: The development of the IRFs and the stand-alone convolution routines are mostly written in Python. In the past year, we have expanded the capability of RVIC to run using Python versions 2.6 and greater, including the current version of Python 3.4. These software infrastructure updates will allow RVIC to continue to use many of the widely supported Python packages.

  2. Development of tools for basin delineation: In conjunction with collaborators at Arizona State University, a new set of Digital Elevation Model and Basin Delineation tools have been developed and added to the RVIC software repository. This collaboration came about as a result of the RVIC model being made publicly available on Github:

  3. Added ability to search for channel: An issue frequently encountered in basin delineation when using flow direction rasters is the incorrect assignment of the pour point to a single grid cell. A simple routine for searching pour points nearest neighbors during the development of the IRFs has been added to RVIC this year.

  4. Improved documentation: The documentation for RVIC is now publicly available at:


  1. Dag Lohmann, Ralph Nolte-Holube, Ehrhard Raschke. A large-scale horizontal routing model to be coupled to land surface parametrization schemes. Tellus A Co-Action Publishing, 1996. Link

  2. Huan Wu, John S. Kimball, Nate Mantua, Jack Stanford. Automated upscaling of river networks for macroscale hydrological modeling. Water Resour. Res. 47, n/a–n/a Wiley-Blackwell, 2011. Link

  3. Philip W. Jones. First- and Second-Order Conservative Remapping Schemes for Grids in Spherical Coordinates. Mon. Wea. Rev. 127, 2204–2210 American Meteorological Society, 1999. Link

[Someone else is editing this]

You are editing this file