As of this writing, the Arctic Regional Climate Model Intercomparison Project (ARCMIP) was gaining momentum. This effort, a coordinated set of limited-area simulations of present-day Arctic climate, focuses on the SHEBA year (October 1997-October 1998). This year was chosen due to the availability of high-quality data from the field camp in the Beaufort and Chukchi seas, aircraft observations, as well as data from the Barrow Arctic Radiation Monitoring site and remote sensing. Use is also being made of observations from the Mackenzie GEWEX study. At least seven different modeling groups are participating. Like other MIPs, the simulations employ common initial and boundary conditions for each model. Initial studies are examining performance biases of the ARCMIP models for geopotential height, 2-m temperature, cloud cover and surface radiation fluxes. Results indicate that the model composite mean biases are on the order of differences between different observations or data analyses. Echoing the GCM comparisons, there is considerable scatter between the simulations for individual models.
Examples of three regional climate models participating in ARCMIP are:
• Arctic Regional Climate System Model (ARCSyM) Development of ARC-SyM is described by Lynch et al. (1995). ARCSyM can be adapted to treat the ocean at various levels of complexity. It includes a dynamic/thermodynamic sea ice model. In various simulations, the required driving fields at the model boundaries have been provided from the NCEP/NCAR reanalysis and ECMWF operational analyses.
• High Resolution Limited Area Model (HIRLAM) HIIRLAM was developed at the Alfred Wegner Institute (Dethloff et al., 1996) and its different versions have been applied variously to the Arctic basin, the Antarctic and mid-latitudes. In most simulations, ECMWF operational analyses have driven the model at the boundaries.
• Pennsylvania State University/National Center for Atmospheric Research Fifth Generation Mesoscale Model (MM5) This atmospheric model (Grell et al., 1995) has been very widely used and has been highly modified for applications to the polar regions (Polar MM5). Modifications are reviewed by Bromwich et al. (2001a) and are ongoing.
A good example of a regional model application is that of Lynch et al. (2001a). This study used ARCSyM to simulate the extremely low sea ice extent and concentration observed over the western Arctic in the summer of 1990. Development of the open-water feature was examined as a case study in Chapter 7 from an observational viewpoint. Lynch et al. (2001a) conducted a control experiment, followed by several sensitivity experiments. The model was initialized on April 15, 1990, and integrated over a six-month period. Sea ice was initialized to ice area defined by SSM/I data, with thickness initialized from output of a ten-year simulation using a different stand-alone sea ice model. ARCSyM was forced at the lateral boundaries with ECMWF operations analyses. The control run simulation reproduced the observed ice anomaly quite well, along with the major aspects of strong cyclone activity, early opening of leads and polynyas along the Eurasian coast and enhanced ice melt, that were important to its development (see Chapter 7). Sensitivity experiments (a) through (e) used the same configuration as for the control run, except that:
(a) The model was initialized with a uniform 2.0-m ice thickness (PRECON)
(b) The ice dynamics subroutine was turned off (NODYN)
(c) The surface albedo for all grid points was fixed to the initial sea ice albedo (NOIALB)
(d) The latent heat flux from leads and open water was set to the latent heat flux from ice (NOLHF)
(e) Ice area was constrained to average conditions (no anomaly) from May through July (NOANOM)
Results from these simulations appear in Plate 7 as ice concentrations averaged for September 1990. They are quite instructive in showing the power of the model sensitivity experiments in isolating the key mechanisms of ice loss.
Ice concentrations from the NOLHF are quite similar to those from the control run. This indicates that removing the extra source of moisture represented by open water has little effect. Given that the ice anomaly pattern from NOLHF is very similar to that of the control run (not shown), we can compare it with anomaly patterns from the other experiments. Results from the PRECON experiment point out the importance of specifying the initial ice thickness. In the control experiment, ice is much thicker than 2 m over most of the domain. Starting with an overly thin ice cover in the PRECON
Plate 7 Average September ice concentration from five sensitivity experiments using the ARCSyM model addressing development of the 1990 sea ice anomaly (from Lynch et al., 2001a, by permission of AGU). See color plates section.
experiment leads to an unrealistically large ice loss. The reason is that less total energy is needed to completely melt the thin ice. When the ice dynamics are turned off (NODYN) we see the role of ice motion in the development of the feature. With no dynamics to move the ice offshore, the spatial pattern of observed ice loss cannot be reproduced.
Removing the ice albedo feedback (NOIALB) gives a September ice field that is more extensive and more compact, due to reduced summer melt. This is in accord with observations, which suggest the importance of unusually early melt and consequent albedo reductions in the development of the ice anomaly. Without the early opening of coastal leads and polynyas seen in the observations (the NOANOM experiment), a smaller ice anomaly is generated, and the ice edge is more compact than in the control experiment.
A very different regional model application is the study of Bromwich et al. (2001a), which addressed mesoscale modeling of the climate and katabatic wind regime over Greenland with Polar MM5. They present verification of two months, April and May 1997, of 48-hour Polar MM5 forecasts for the Greenland region. Verification data consist of global atmospheric analyses as well as records from AWSs and instrumented aircraft. Monthly mean near-surface temperatures and wind speed predicted by Polar MM5 differ from observations by less than 1 K and 1ms-1, respectively. The model captures the observed diurnal cycles in these variables, as well as their large-scale, synoptically forced changes. Aircraft observations also point to the ability of the model to capture profiles of wind speed, direction and potential temperature in the katabatic wind layer (see Chapter 8 for a discussion of katabatic winds).
Figure 9.13 provides an example of results. The figure shows time series of 24-48 hour model forecasts of a series of surface variables in comparison to data from the Summit AWS located over central Greenland. The generally excellent performance of Polar MM5 is obvious. Performance at other locations over Greenland is similar. There is a tendency at some locations for significant error in the near-surface temperature in cases of weak winds and strong static stability, which is attributed to the difficulty in accurately parameterizing the turbulent heat fluxes in these conditions. At least part of the error in mean surface pressure is caused by uncertainty in the elevation of the AWS sites. Polar MM5 has a slight moist bias near the surface, as seen in the comparison between the observed and modeled near-surface water vapor mixing ratio (the ratio of water vapor mass to the mass of dry air) (Figure 9.13, bottom panel).
Was this article helpful?