Model Configuration

The regional climate model we present here is a limited area model, and thus must receive forcing data, i.e. initial and boundary conditions, from another source model. First we summarize these large-scale simulations and follow with a detailed discussion of the regional model configuration.

A.    Forcing data

Reanalysis simulations with a general circulation model (GCM) provide an excellent forcing dataset for validating the regional model. Reanalysis grids are similar to free-running global models in terms of spatial resolution. However, since observations are assimilated into the simulation, simulated large-scale fields such as temperature, geopotential heights, and humidity are closely constrained to the actual atmospheric state. Thus, the output fields represent an idealize GCM in which the large-scale atmospheric state and its time evolution is well represented on the daily, seasonal, and interannual scale. Regional simulations forced by reanalysis fields help isolate deficiencies in the regional model without the complexity of biases inherited from the forcing model. Here, we use the NCEP/NCAR Reanalysis Project (NNRP) data {Kalnay 1996} for the 10-year period 1990-1999. Data are simulated at 6-hourly intervals, 2.5 degree x 2.5 degree horizontal resolution (approx. 275 km x 200 km), and 17 pressure levels.

We also present here simulations forced by two free-running global climate models, NCAR/DOE Parallel Climate Model (PCM) and ECHAM5. The ECHAM5 results are also used for climate change simulations in a companion paper {Salathé, 2007} and will be the main focus of the present paper. PCM is a global model derived from the National Center for Atmospheric Research (NCAR) Climate System Model with a parallel ocean model developed at the Department of Energy (DOE); the simulation used here was run at T42 resolution (approx 220x300km grid spacing). ECHAM5 is based on the fifth-generation atmospheric general circulation model developed at the Max Planck Institute for Meteorology (ECHAM5). This model is the most recent version in a series of ECHAM models evolving from the spectral weather prediction model of the European Centre for Medium Range Weather Forecasts (Roeckner et al., 2003). ECHAM5 was coupled to the Max Planck Institute ocean model (MPI-OM) and was forced for the current study with observed radiative parameters. ECHAM5 was run at T63 spectral resolution, which corresponds to a horizontal grid spacing of approximately 140x210 km grid spacing at mid-latitudes.

Output from the forcing model was used to provide initial and lateral boundary conditions every six hours for the regional climate model. The global model output was also used to nudge the outermost regional model domain, which is discussed in detail below. Regional simulations will be referred to as NNRP-MM5, PCM-MM5, and ECHAM5-MM5 to indicate the forcing model.

B.    Regional Model

The Pennsylvania State University (PSU)-National Center for Atmospheric Research (NCAR) mesoscale model (MM5) Release 3.6 was used as the regional climate model. Although MM5 was developed for mesoscale weather forecasting and has been operating as such in real-time at the University of Washington (UW) and many other places for over a decade, there is precedent for its use as a tool for regional climate modeling {Leung, 2004 #583}. MM5 is a limited-area, non-hydrostatic, terrain-following sigma-coordinate model designed to simulate or predict mesoscale atmospheric circulation {Grell, 1993 #558}. Parameterizations include Kain-Fritsch convective parameterization {Kain, 1993 #557}, Medium Range Forecast model (MRF) planetary boundary layer (PBL) scheme {Hong, 1996 #570}, CCM2 radiation scheme {Hack, 1993 #559}, and Simple Ice cloud microphysics {Dudhia, 1989 #572}.

High regional model resolution is achieved by using multiple MM5 nests at 135 km, 45 km, and 15 km horizontal grid spacing. Figure 4 shows the MM5 nests used in this study. It is worth noting that “one-way” nesting is utilized; that is, the global climate model is run independently first, without updates from the regional model solution; furthermore, information only passes from the outer to inner nests in the mesoscale simulation.

In order to capture the large-scale processes important for Pacific Northwest climate outermost MM5 domain encompasses nearly the entire North American continent and much of the eastern Pacific Ocean. The use of such a large outer domain keeps the outer mesoscale boundaries far from region of study and weather systems approaching the Pacific Northwest well represented by the time they reach the region. The second nest covers the western United States and portions of Canada and Mexico, capturing storm systems and Southwest Monsoon circulations that influence the Pacific Northwest. The innermost domain covers the states of Washington, Oregon, and Idaho and the entire Columbia River Basin.

High spatial resolution is critical to success in capturing the essential features of the Pacific Northwest climate, and the 15 km resolution chosen here should be sufficient. Mass et al. (2002) demonstrated, using MM5, that increasing horizontal grid spacing – from 36 km to 12 km grid spacing – improved mesoscale weather forecasts for precipitation, 10m wind, and 2m air temperature, and sea level pressure in the Pacific Northwest. Similarly, {Leung, 2003 #90} showed an improvement in reproducing precipitation patterns over topography when increasing regional model resolution from 40 km grid spacing to 13 km grid spacing.

1.    Nudging

As with the MM5-based real-time numerical weather forecasting system used at the University of Washington {cite}, nudging is applied to the outermost regional model domain from the forcing fields. Nudging relaxes the regional model solution for wind, temperature, and moisture towards the driving global climate model solution. The relaxation takes place throughout the interior of the domain and at all vertical levels above the planetary boundary layer. Particularly over large domains, the regional model solution can drift over time from that of the driving global climate model. If we assume that the global climate model reasonably captures synoptic-scale structures and that the goal of the dynamic downscaling system is simply to obtain fine-scale detail for a given large-scale pattern, then the regional model should not modify the large-scale patterns, and such a drift is undesirable.

Other methods to limit this drift rely on using smaller regional model domains (Jones et al., 1995) or periodic (e.g. every 10 days) reinitializations of the simulation (Pan et al. 1999). Since some spin up time is required after each reinitializations, this approach is computationally inefficient. But more importantly, reinitializing the model looses the slow varying parameters in the model, such as soil parameters and snow cover, that are the essence of climate system modeling. Von Storch et al. (2000) showed that nudging was able to keep simulated states close to the driving state at large scales while still generating small-scale features. The development of nudging has yielded an option that allows for a larger regional model domains and makes continuous model runs possible but still limits model drift.

The inner two domains are not nudged, allowing the mesoscale model to freely develop atmospheric structures at finer spatial scale. This approach attempts to preserve the large-scale state provided by the global model while generating regional meteorological details on the inner nests.

2.    Soil Parameterization

Accurate representation of land-atmosphere interactions in climate models is critical to the realistic simulation of global and regional energy and water cycles (Wang et al., 2004). Realistic modeling of soil moisture and temperature dynamics, in particular, is crucial in capturing moisture and heat fluxes at the surface. These physical processes, in turn, directly influence air temperature, air moisture, and snow dynamics. Snow pack in particular is critical for understanding climate impacts in regions such as the Pacific Northwest where snow melt plays a central role in regional hydrology. In order to capture these dynamics in the climate system over climate change scales, the soil column must freely interact with the atmosphere. Most climate models, however, prescribe the lower-boundary soil temperature to some climatological value, which restricts the response of the soil column to climate forcing. Furthermore, if the prescribed value is not realistic for the simulated climate, a spurious heat source or sink is introduced to the land surface.

The upper few meters of soil act as a heat reservoir, storing heat in the spring and summer and releasing it in autumn and winter (De Vries, 1975). Heat is transferred through the soil column primarily by conduction, penetrating a few centimeters to perhaps half a meter on daily timescales and to depths as large as 10 meters on annual timescales (De Vries, 1975). Observations (Baxter, 1997) and models (Jury, Gardner, and Gardner, 1991) of soil temperature evolution at different soil depths show that soil temperature should be both time-lagged and amplitude-damped with depth with respect to the annual surface soil cycle. In these studies, at a depth of three meters (lower soil boundary in the NOAH LSM), the annual soil temperature cycle is typically time-lagged by 70 days and amplitude-damped to about one-third the amplitude of the surface temperature cycle. Thus, we expect the lower boundary of the soil model would interact with the climate – both on the annual scale and under climate change scenarios. A prescribed lower boundary temperature at 3 meters depth may not be problematic for weather forecasting, typically over time scales of days to weeks, since it takes on the order of months for thermal information at 3m depth to reach the surface. For climate simulations over many years, however, this deficiency on the surface energy budget would accumulate large errors in the surface parameters.

To address these issues, we have implemented a deep soil temperature parameterization that yields an annual cycle of deep soil temperature that simulates the observed temperature cycles deep in the soil column. The soil temperature at depth follows the variations in the surface skin temperature, but with a phase lag and attenuation with depth. The phase lag and attenuation depend on the frequency of the surface variation, but we shall base our methodology on the annual cycle. The desired response may be obtained by taking a simple weighted average of the skin temperature over the previous year, where the weighting is adjusted to yield the desired attenuation and phase lag. We choose a weighting function with two parameters as follows. We take the mean skin temperature over the full year () and the n days () prior to the time of interest. The soil temperature is then the weighted mean of these two values,
           
By selecting appropriate values of a an n we can tune this equation to produced the desired attenuation and phase lag for the depth of interest. For 3m depth, we use the published observed values of 30% attenuation and 70 days lag to obtain a=0.6 and n = 140 days.

To test this method, we used surface and 1-m deep soil temperatures observed at Ames Iowa, US. For this case, we use =0.3 and n = 46 days to obtain the observed lag and attenuation. Note that, compared to the values for 3m depth, a smaller n yields a smaller lag and the smaller a yields less attenuation. In figure 10, the blue line shows the observed skin temperature and the black line the observed 1-m soil temperature. The weighted mean of the skin temperature yields the red line, which closely captures the form of the observed soil temperature over four seasonal cycles. Note also how the parameterization effectively removes the high-frequency variations at the surface in accordance with observations.

When applied to the MM5 climate modeling system, the parameterization uses surface skin temperatures generated by MM5 and the NOAH LSM to derive the lower boundary temperature. For the first year of each decade-long simulation, however, MM5 output is not available, and must be derived from a spin up simulation (i.e. a preliminary simulation not used in the analysis of results, but only to equilibrate model parameters). This spin-up is also required to bring the NOAH LSM soil temperature and moisture fields for the entire soil column into equilibrium with the simulated atmospheric state. Studies (Cosgrove et al., 2003, for example) have shown that, especially in drier regions, a spin-up period of at least a year is necessary for soil moisture, as the land surface model slowly adjusts soil parameters away from default values (which can be unrealistic). Thus, the spin-up simulations are initialized at the end of the summer (September 1989), when soils are climatologically most dry in the Pacific Northwest. For simplicity, we use the same forcing data for the spin-up as for the first year of the actual climate simulation. At initialization, the global forcing model (i.e. NNRP OR ECHAM5) is used in Eqn. (1) to set the boundary temperature. The deep soil temperature parameterization was implemented at each internal time step to avoid biases in temperature that would result from a once-daily algorithm implementation. Soil temperatures for all intervening layers (between the surface and three meters) in the NOAH-LSM are linearly interpolated and then allowed to evolve according to the LSM throughout the simulation. Soil moisture is initialized to the climatological values contained in MM5 and then allowed to evolve over the spin-up year according to the LSM. As MM5 surface data became available at each time step, the parameterization was updated at each grid point using the available MM5 output, thereby phasing out the global model data. At the completion of the spin-up year, a complete year of MM5-derived skin temperatures is available for the deep soil parameterization and the soil temperature and moisture profile has spun up to the atmospheric forcing. Figure 11 illustrates the soil temperature for the surface down to three meters for spin up and first two years of the PCM-MM5 simulation interpolated to the SeaTac Airport meteorological station. Due to the gradual phase-in of MM5 data over the spin-up year, a difference between values for the deep (300cm) soil temperature cycle over the first year and subsequent years is noticeable.

This soil temperature parameterization not only yields a deep soil temperature cycle that is more realistic – with amplitude that is significantly damped – but also allows for a change in the deep soil temperature pattern over time. That is, whereas the default deep annual temperature cycle is the same year after year, this system allows the temperature at three meters to evolve with the rest of the climate system. Thus, when a change in atmospheric radiative forcing occurs and climate change results, the entire soil column will respond accordingly instead of being constrained at depth by the same annual cycle year after year.