Water Balance of the Mississippi River Basin from a Macroscale Hydrologic Model and NCEP/NCAR Reanalysis
Edwin Maurer1, Dennis Lettenmaier1, John Roads2
1Dept. of Civil & Environmental Engineering, University of Washington; 2Scripps Institution of Oceanography
[POSTER PRESENTATION AT AGU FALL 1999 MEETING]



ABSTRACT

Understanding the annual and seasonal water cycle of continental river basins requires the ability to reproduce the complex dynamics of land-atmosphere interactions. At least two pathways can be followed in such endeavors. The first is to use coupled land-atmosphere models, which assure by construct that the water and energy balances of the atmosphere and land surface are closed. This is accomplished, however, at the expense of being unable to constrain some critical surface processes, such as precipitation, to match observations. A second approach is to use macroscale hydrologic models, forced at the surface with observed hydrometeorological variables, such as precipitation – thus avoiding biases in (downward) surface fluxes. A variation of the coupled modeling approach makes use of reanalysis, in which coupled land-atmosphere models incorporate assimilation of (mostly atmospheric) observations using a "frozen" version of a numerical weather forecast model. This approach produces many of the same variables either predicted, or used to force, macroscale hydrological models.

Comparison of reanalysis and macroscale hydrologic model predictions of surface variables such as turbulent heat fluxes, net radiation, and streamflow, offer an opportunity to diagnose both the coupled land-atmosphere and uncoupled hydrologic models. This study utilizes the NCEP/NCAR reanalysis, and an off-line simulation of the Variable Infiltration Capacity (VIC) hydrologic model for the Mississippi River basin for the period 1988-97. The VIC model runs uses a spatial resolution of 1/8 degree, to which the NCEP/NCAR reanalyses were spatially interpolated. The VIC and NCEP/NCAR temporal resolutions are 3 and 6 hours, respectively. Streamflow corresponding to runoff generated by both the VIC and reanalysis is predicted through application of a routing model to the outlet of five major sub-basins of the Mississippi. Evapotranspiration predicted directly by the VIC model via its surface budget is compared with an atmospheric budget calculation using reanalysis atmospheric divergence, in combination with precipitation from the VIC forcings, for the five subbasins, and the entire Mississippi basin.

Variable Infiltration Capacity (VIC) Macroscale Hydrologic Model
Study Area

Features of the VIC model include:
Infiltration and soil moisture accounting
    -Infiltration capacity determined by Xinanjiang nonlinear model
    -Gravity drainage between layers regulated by Brooks-Corey relationship
    -Baseflow generated by Arno formulation
Vegetation
    -Each cell allocated an arbitrary number of land cover classes
    -Covers specified by their Leaf Area Index, resistances and root distribution
    -ET, infiltration, soil moisture fluxes, runoff and baseflow computed for each cover type
Snow
    -Snowpack accumulation and ablation simulated by a 2-layer energy balance model with canopy effects
Ground Heat Flux
    -Explicit two-layer parameterization
Other Model Components
    -Routing
    -Spatially distributed precipitation
    -Frozen soils capability

Contrasting NCEP/NCAR Reanalysis and VIC

The NCEP/NCAR Reanalysis project (Kalnay, et al., 1996) addressed the problem of using numerical weather prediction model output for climate studies, namely that periodic changes in the model create discontinuities in the variables that mask climatic changes and trends. Using a wide array of historic observations, a "frozen" state-of-the-art model was used to generate continuous, consistent data sets from 1948-97. The land surface parameterization used in the NCEP/NCAR Reanalysis is designed to model the essential characteristics of the land interactions with the atmosphere primarily for partitioning of net radiation into latent and sensible heat (Mahrt and Pan, 1984). The need for computational simplicity results in many values being fixed globally, such as soil characteristics (wilting point, field capacity) and vegetation coverage. Precipitation input to the land surface is derived from the atmospheric model. By contrast, the VIC model computes a full solution to the energy and water budgets at each time step using spatially distributed input parameters (with sub-grid variability) (Liang, et al., 1994). For this study, input data sources for the VIC simulation include:


The principal available data against which the VIC model can be calibrated and tested is runoff, which is routed to key outlet points in the sub-basins. With its characterization of the land surface using observationally-based databases above, and driving the model with observed precipitation and temperature, the VIC output gives us a surrogate time series of land surface data that are spatially distributed. This is valuable as a benchmark for diagnosing the performance of other models, such as NCEP/NCAR Reanalysis, especially where the modeled parameters lack a good database of observations for comparison, such as for soil moisture. (Maurer, et al., 2000)

Validation of the VIC Model

The VIC model is verified using two measures. One of the only long-term databases of soil moisture in the Mississippi River basin is the Illinois data set of Hollinger and Isard (1994). The average monthly values from the Illinois study (from 17 point measurements taken every 2-6 weeks from 1981-92) are compared with VIC model results for the state of Illinois. Second, the runoff from each 1/8 degree grid cell is routed to points with measured, or estimated naturalized flow values, where the hydrographs are compared.

Comparison of VIC and Illinois Data
Comparison of Modeled and Measured Hydrographs

Land Surface Water Balance Components:

The VIC model includes a surface runoff and a groundwater baseflow component, where Reanalysis only simulates surface runoff. The VIC model closes its water budget at each time step, while Reanalysis includes a nudging term, U, which is used to maintain an assumed monthly climatological average soil moisture. Modeling of ET provides the mechanism for establishing the partitioning of net radiation at the surface into latent and sensible heat, which feed back into the atmospheric dynamics. Soil moisture, which responds slowly to imposed forcing, provides a means for persistence in the modeled land-atmosphere system, i.e. extended wet or dry periods.

Scematic of Water Balance Components

Water Balance Study Results

Lower Mississippi River Basin
Ohio River Basin
Missouri River Basin
Upper Mississippi River Basin

Spatial Summary Plots

Precipitation
Evapotranspiration
Average Soil Moisture

Time Series Summary Plot

Entire Mississippi River Basin

Atmospheric Residual Evaporation

Using the atmospheric water balance components from Reanalysis (vertically integrated horizontal moisture convergence, MC, and precipitable water, PW) with observed P, the residual ET can be computed as:

Though this method reduces the ET bias, the one-month lag autocorrelation of evaporation anomalies decreases from 0.15 (Reanalysis) to 0.06 (Residual), compared with 0.30 for the VIC model results, showing the importance of a land-surface parameterization for modeling climate persistence.

Summary of Monthly Atmospheric Residual ET by Basin

Water Budget Changes with Reanalysis-2

NCEP/DOE AMIP-II Reanalysis (Reanalysis-2) is a follow-up to NCEP/NCAR Reanalysis, currently underway at Lawrence Livermore National Laboratory (Ebisuzaki, et al, 1998). The first phase includes 1979-97, for which preliminary data are now available through 1995 [http://wesley.wwb.noaa.gov/reanalysis2/]. Reanalysis-2 corrects errors in the NCEP/NCAR Reanalysis, and makes certain improvements to the model. The most significant change directly affecting the land surface water budget is the assimilation of precipitation observations, using a scheme to update the soil water in accordance with differences in modeled and observed precipitation. The nudging toward an imposed climatology is replaced by this adjustment toward observed precipitation (which is still referred to as U below). The changes in Reanalysis-2 to some of the more pronounced biases identified in this study are shown.

Precipitation and ET Comparisons
Soil Moisture and Snow Water Equivalent Comparisons
Missouri River Time Series Comparison
Monthly Average Water Balance Components for Entire Mississippi River Basin

SUMMARY

- The VIC hydrologic model can function as a benchmark against which to compare water balance components from other simulations
- NCEP/NCAR Reanalysis shows certain regional and temporal biases in precipitation, predominantly in the summer in the southeastern Mississippi Basin
- NCEP/NCAR Reanalysis ET is consistently overestimated
- Snow extent and duration are underestimated in Reanalysis
- Intra-annual soil moisture fluctuations are overpredicted in Reanalysis
- Inter-annual variation and persistence are low as compared with the VIC model simulations
- Nudging of the soil water toward the assumed climatology may be inappropriate, especially in the Upper Mississippi and Missouri Basins.
- Overestimation of ET and latent heat in Reanalysis results in an underestimation of sensible heat
- Estimation of ET from the atmospheric water budget, using observed P, results in an improved ET estimate while losing the ability to capture the persistence in the system.
- Reanalysis-2, by assimilating observed precipitation and including other improvements, reduces some of the biases in NCEP/NCAR Reanalysis fields for water balance components, especially soil moisture. Though they are lower in some regions, biases remain in P and ET, and runoff, N, may be adversely affected by the changes.

REFERENCES