1. INTRODUCTION
An extensive research effort sponsored by
the DOD has been conducted over the past three years to determine the
feasibility of using a deterministic limited-area high-resolution
dynamical-numerical model to generate local climate statistics at any desired
location around the globe. The principle motivation for this development is the
limitations encountered in the use of long-term observational datasets and
probabilistic methods based on statistical relationships. The limitations of using observational data
and probabilistic methods include the nonavailability of long-term point
observations for large areas of the world, the problem of representativeness of
point observations in complex terrain, the relocation of observing sites, and
the change in land use around a site over time. The quality of observations
also changes in time as new instrumentation is used. Additionally, there is the
problem of determining the statistical relationships between the climate
observed at a point and the climate at another point at some distance away
based upon information such as elevation, slope and aspect.
In this research a dynamical-numerical
model is used to determine the local climate from a set of long period
mesoscale simulations. The objective is to simulate the actual climate
statistics for a particular period of time over a specified region. This is
achieved by executing the model for a long period of time in a data
assimilation mode and allowing the model to fuse the available data into a 3-D
climatological dataset.
Initial results suggested that this
method would provide useful climate information (Zack et al. 1996). This paper examines various modeling issues
such as the impact of model resolution, observed data availability, data
assimilation, convective parameterization scheme, and planetary boundary layer
____________________________________________
*Corresponding author address: Glenn E. VanKnowe,
MESO, Inc., 185 Jordan Rd., Troy, NY 12180,
e-mail: glenn@meso.com
formulation associated with the method used. The paper presented by Doggett et al. (1999) concentrated on the results of this technique and compares the modeled climatologies with the traditional techniques.
2.
METHODOLOGY
The first step of the basic methodology is the initialization of a dynamical-numerical model with large-scale gridded data fields. The model then uses the large-scale gridded data fields and lateral boundary conditions and periodically ingests the available observed data through one of several possible data assimilation techniques (Newtonian relaxation, periodic reanalysis, etc.). Next the model dynamically fuses the available observations with its knowledge of the surface characteristics of the earth and the basic principles of physics to generate estimates of local climate statistics at locations for which no observational data is available. This technique has been given the name CLImate statistics by a dynamical MODel (CLIMOD). The research has shown that the quality of the simulated climate statistics is significantly impacted by several modeling factors including the model resolution, method of data assimilation, data availability, specific model configuration and subgrid parameterization schemes.
3. MODEL CONFIGURATION
The simulation model used in CLIMOD is a version of a dynamical-numerical atmospheric mesoscale model called the Mesoscale Atmospheric Simulation System (MASS) (Kaplan et al. 1982 MESO, 1998; Manobianco et al. 1996). MASS version 5.9 was used for the initial study over the eastern Great Lakes and MASS version 5.11 was used in the more recent study. Hydrostatic formulations of MASS were used in both studies although subsequent investigations may use a non-hydrostatic version of the model to create very high resolution climatologies. The specific features of the MASS model used in CLIMOD are described in the MASS Reference Manual (MESO 1998). Unless otherwise noted, all the experiments described in this paper used a baseline configuration of the Blackadar PBL scheme, a modified Kuo cumulus parameterization and 24-hour discontinuous data assimilation. All the simulated "surface" variables used for comparison with standard observational data were interpolated to the 2-meter level through the use of boundary layer similarity functions.
Figure 1. Depiction of the 24-hour discontinuous data assimilation scheme employed in the baseline run of the CLIMOD simulations over the eastern Great Lakes.
4. EXPERIMENTAL DESIGN
4.1. General Design
A set of experiments was formulated to assess the basic quality of the climate statistics produced by the CLIMOD technique and to determine the sensitivity of the statistics to a variety of factors. The experiments were designed to evaluate the impact of resolution, various data assimilation strategies, data availability, boundary layer schemes, and cumulus parameterization schemes. The objective was to generate information that could be used to construct an optimal configuration of the modeling system for the purpose of generating high-resolution climate statistics within the limits of the current state of moderate cost computational power.
Figure 1 shows the discontinuous data assimilation strategy employed in most of the experiments. For all experiments, a coarse mesh simulation (40 km grid) was initialized at 00 UTC on 1 January and 1 July of each simulated year over the outer area shown in Fig. 2. This simulation was
initialized from an optimum interpolation (OI) analysis which utilized the Global Optimum Interpolation (GOI) analysis data archived at NCAR as a first guess and then incorporated rawinsonde and surface data. The NCEP-NCAR Reanalysis data has been used in subsequent experiments. The coarse mesh model was then executed for 24 hours using interpolated GOI data at 12-hour intervals to specify the lateral boundary conditions. The data from the 24th hour of this simulation was used as the first guess for the initialization analysis of the next simulation. Once again, rawinsonde and surface data was used to update the first guess. All of the fields describing the state of the earth’s surface (soil moisture, snow depth, etc.) were carried through each subsequent reinitialization process. Thus, the subsequent simulations began with the values existing at the end of the previous simulation.
A fine mesh (10 km grid) simulation was initialized at 00 UTC on 1 January and 1 July of each year by interpolating the initialization dataset of the coarse mesh grid. The fine mesh model was then also executed for 24 hours. However, the lateral boundary condition data was taken from the hourly output of the 40 km simulation. The data from the 24th hour of the 10 km simulation was used to directly initialize the next day’s fine mesh simulation. Rawinsonde and surface data were used to update this field.
The goal of the experimental design was to simulate the climate statistics using several different simulation and sampling strategies for a particular period of time over a region of high data density and complex terrain and land/water distributions. The high density of observations allowed for withholding of observed data and an objective comparison between the simulated and observed climatologies. The area chosen for the initial experiments was the eastern Great Lakes region (Fig. 2) and the time period was 10 months of both Januarys and Julys extending from 1985 through 1994. More recently, the method has been extended to produce data for all 12 months of a 10-year period (1973-1982) over a region of the Middle East and in the Korean area of the Far East. Model output was created hourly on a 3-D grid of twenty vertical layers at both 40 km and 10 km resolutions. The data from the set of 10-year 3-D grids was then interpolated and evaluated for any level or point of interest. The interpolation from the model levels to the 2 meter standard surface observation height was done through the use of similarity theory functions for the atmospheric boundary layer.
Figure 2. The 40 km (outer box) and 10 km (inner box) model domains used for the eastern Great Lakes experiments.
4.2. Data Availability
Experiments
Several sets of experiments were performed to examine the impact of observed data availability on the quality of simulated climate statistics. These experiments addressed three types of availability issues: (1) temporal frequency; (2) spatial density and (3) data type. Only the first set will be discussed in this paper. Three simulations were run to assess the impact of temporal frequency of the data ingestion. The first simulation sequence assimilated rawisonde and surface data at 24 hour intervals through the using of an OI analysis procedure using the 24th hour of the previous simulation as a first guess field. The second simulation used the same procedure but the frequency was increased to every 12 hours. The third simulation sequence did not assimilate any raw observational data and the only new information coming into the model domain was from the lateral boundary condition data which was interpolated at 12-hour intervals from the GOI grid point data set.
4.3. Data Assimilation
Experiments
A major issue in the development of the CLIMOD method was the selection of a data assimilation scheme to assimilate the raw observational data into the simulation sequence. In order to assess the impact of data assimilation method on the resulting climate statistics a set of four simulation sequences
were executed. Each sequence used a different data assimilation approach: (1) no data assimilation, (2) discontinuous (i.e. periodic OI analysis using the end of the previous simulation segment as a first guess); (3) Newtonian Relaxation ("nudging") (Hoke and Anthes 1976); and (4) incremental update analysis (IAU) (Bloom et al. 1996).
4.4. Model Configuration Experiments
Although the basic equation set in most mesoscale models is quite similar, there can be a large degree of variance in how sub-grid processes such as boundary and surface layer physics, moist convective (cumulus) processes and moist microphysics are parameterized. Indeed, many models have multiple choices for each of these types of parameterized physics. Thus, an experiment was designed to test the impact of the choice of parameterization scheme on the climate statistics. The focus was on the impact of the boundary layer schemes and cumulus parameterization schemes. The boundary layer experiment compared the results from the Blackadar (1979) boundary layer scheme with that based on a prognostic turbulent kinetic energy (TKE) budget (Therry and Lacarrere, 1983.
4.5. Model Resolution Experiments
The final aspect that was investigated was the impact of grid resolution on the climate statistics. This was examined by comparing the results of the climate statistics produced on the 40 km and 10 km grids.
5. RESULTS
5.1. Observational Data
Availability Experiments
As noted in section 4.2, three simulations were executed in this experiment: (1) 12-hour discontinuous data assimilation, (2) 24-hour discontinuous, and (3) no data assimilation. These simulations were executed for both the months of January and July. Only the July results are shown here, however, the results from the January simulations showed similar features. Typical results from the experiments are shown for Albany, NY in Table 1 and Fig. 3. It is important to note that data from the Albany, NY surface observing site along with a set of several other "evaluation sites" were not used in any of the simulations. Thus, the impact of the data assimilation is from data reported from other sites but not that from the evaluation sites such as Albany.
|
Table 1. Observed vs. Simulated Monthly Mean Value Comparison July - Albany,NY |
|||||
|
Variable |
Obs |
12hData |
24hData |
No Data |
|
|
Temp |
73.5 |
74.4 |
75.1 |
79.9 |
|
|
Dew Pt |
62.3 |
59.5 |
60.5 |
64.4 |
|
|
W Speed |
5.8 |
6.0 |
5.7 |
6.6 |
|
|
Precip(%) |
8.4% |
6.4% |
8.9% |
12.4% |
|
Figure 3. Observed and CLIMOD-simulated frequency of hourly precipitation at Albany, NY for July.
As one might expect, the results show that the assimilation of data tends to produce better results than the no-data assimilation experiment. In the case of no data assimilation, the model biases are free to emerge over the 31-day period of each simulation. In the case of July at Albany a substantial warm bias in temperature and moist bias in dewpoint is present when no data (other than lateral boundary condition data is used during the course of the simulation). There is also a noticeable high bias in the wind speed. In addition, the no-data-assimilation simulation produced a substantial high bias in the precipitation frequency. This is consistent with the fact that there was too much heat and moisture in the low levels which led to more frequent triggering of the model's convective parameterization scheme (a modified Kuo scheme in this case). It also may be affected by the "trigger function" in the model's convective parameterization scheme and possibly other aspects of the convective scheme.
The incorporation of data at 24-hour intervals served to significantly reduce the biases. For example, the temperature bias is reduced from over 6°F to under 2°F when the 24-hour discontinuous data assimilation is employed. The use of a 12-hour data assimilation frequency rather than a 24-hour frequency did not result in a substantial improvement for most variables. However, the discontinuous assimilation of data at 12- or 24-hour intervals also introduces some anomalies into the hourly statistics. This can be seen in the hourly precipitation frequencies shown in Fig. 3. There is a notable afternoon maximum in precipitation frequency in all of the simulations and the observations. However, the two data assimilation simulations exhibit a pronounced reduction in precipitation frequency immediately after the time of the reanalysis. In the 24-hour assimilation run this occurs just after the 00 UTC analysis time (00-03 UTC) and in the 12-hour assimilation case it occurs both after the 00 UTC (00-03UTC) and 12 UTC (12-15 UTC) reanalysis times.
The reduced precipation is primarily caused by the slight low moisture bias in rawinsonde data. Even when the air is actually saturated, rawinsonde data usually observes relative humidity (RH) readings of only 98-99%. This lower reading of RH of 1-2 % is insignificant from an observational point of view, but crucial to the model. The model requires the atmosphere to be saturated in order to produce clouds. The inclusion of nonsaturated conditions at even a few points that were really saturated, even if the data showed nearly saturated conditions, forces a 1-2 hour period to bring the conditions back to saturation. During this period of adjustment, the model clouds and precipitation are artificially reduced. The reduction of cloud cover artificially also increases radiational cooling during nighttime hours and heating during the daylight hours and this can have an impact on the diurnal cycle other variables as well.
A series of experiments are now in progress to further evaluate the question of data availability using NCEP-NCAR Reanalysis gridded data as the initial field and lateral boundary conditions with observational data assimilation occurring every 12 hours in each experiment. The additional experiments are being run (1) with all available surface and upper air rawinsonde observational data, (2) with rawinsonde and surface observations with a uniform 50% spatial reduction, (3) with a data-void area created by removing all of rawinsonde and surface observations in a location but still reduce observational data by a total of 50%, and (4) with all available surface data only.
Figure 4. Observed and
simulated frequency of precipitation for all hours for Albany, NY for January.
5.2.
Data Assimilation
Experiments
As indicated in section 4 there are many methods of assimilating data into the model. The impact of the methods outlined in section 4 on the climate statistics was evaluated for both January and July.
The most significant difference among the schemes was in the frequency and the amount of precipitation. Figure 4 illustrates the frequency of precipitation for all hours of the day for January in Albany, NY. The no-data assimilation simulation indicates that the model when left to itself tends to underestimate the frequency of precipitation in January This is in contrast to the July results in Fig. 3 which indicated a significant overestimate in the frequency of precipitation in the case of no-data assimilation. Once again, the 24-hour discontinuous data assimilation was effective in reducing this bias. However, the use of a 12-hour nudging scheme actually worsened the bias. This was largely attributable to the previously mentioned dry bias in the rawinsonde data. The nudging scheme forces the simulation towards a less moist state for an extended period of time (the nudging period) and thus makes it more difficult for the model to overcome the rawinsonde dry bias.
The introduction of the dry bias also has an impact on the temperature climatology. The less moist atmosphere results in fewer clouds and is more favorable for radiative cooling at night and heating during the day. This can be seen in the diurnal temperature climatology shown in Fig. 5. The nudging simulation sequence produces a significant cold bias in the nighttime temperatures but is similar to the 24-hour data assimilation run in the daytime temperature. The shallow nighttime inversions are more sensitive to this effect than the deeper well-mixed daytime boundary layer.
Figure 5. Observed and simulated hourly mean temperature for
Albany, NY for each hour of the day (UTC) January
5.3. Model Configuration Related Experiments
A series of model configuration experiments were executed. Only the results of the boundary layer parameterization experiments will be shown here. The observed and simulated diurnal temperature cycle at Albany, NY in July is shown in Fig. 6. Both schemes reproduced the diurnal temperature cycle reasonably well. The most significant departure from the observed cycle is that the simulated cycles have a slight warm bias in the early morning hours. The TKE scheme produced slightly cooler temperatures during these hours and is thus in better agreement with the observed mean temperature. Other than this, the schemes are quite similar throughout the remainder of the diurnal cycle.
Figure 6. Observed vs simulated hourly mean temperature for Albany,
NY for July which illustrates the difference between the Blackadar and TKE
simulations.
5.4
Grid Resolution
Experiments
The impact of model resolution was found to be most important in two types of regions, (1) coastal and (2) complex terrain. In general, these regions are more greatly influenced by local surface effects that must be resolved in fine detail to properly capture the climate.
Locations close to large bodies of water showed a significant tendency toward loss of diurnal variability and biases that tended to skew the climate statistics in the direction of the over-water climate as a function of the season and time of day. For example, the wind speeds were simulated higher at all times, the winter time temperatures were generally higher, summertime temperatures were lower during the day, etc. There was significant improvement in the climate statistics in most coastal cases when improving the model resolution from 40 km to 10 km.
The other type of locations significantly impacted by model resolution were regions consisting of mountains and valleys with steep terrain gradients. An example of how increasing model resolution significantly enhances the climate results can be found in looking at the wind direction frequency for Utica, NY (Fig. 7).
Figure 7. Observed and simulated distribution of wind directions at Utica, NY for the month of January.
Utica lies in an east to west oriented valley that causes a sharp bimodal preference for either a near easterly (110 degrees) or near westerly (290 degrees) wind direction to be observed. The westerly peak, which is the stronger peak, is primarily caused by the prevailing synoptic wind that is from the WNW and is captured reasonably well at both the 40 km and 10 km resolution. In this case, the easterly wind is a mesoscale phenomena resulting from a shallow cold air drainage circulation causing the drainage wind to be funneled westward through the valley. This is very poorly resolved at 40 km and much better resolved at 10 km.
7. CONCLUSIONS
The primary conclusion of this research is that the CLIMOD method shows considerable skill in its ability to accurately represent the climatology of a region. However, the skill at which the climatology is represented is significantly influenced by such factors as the model resolution, observed data availability, data assimilation, convective parameterization scheme, and the planetary boundary layer formulation.
The sensitivity study conducted as part of the CLIMOD development process indicated that there was a considerable variation in the response of different variables to changes in the frequency of data availability, method of data assimilation, model physics configuration, and grid resolution. Clouds and moisture were most sensitive to data assimilation and the frequency of data availability. There were secondary impacts on surface temperature, dewpoint and winds. In many cases the impacts were different in January and July. This was especially true for the impact of the configuration of model physics. This suggests that it might be useful to employ different schemes at different times of the year to achieve the best results. This is also likely to be true between different regions. Further research is needed to clarify the dependence of the sensitivity on region and time of year.
It has been well documented that non-hydrostatic events have considerable influence on atmospheric processes that operate on a spatial scale of 20 km or less. Such processes are associated with a variety of phenomena ranging from strong localized convection and to drainage winds associate with downslope conditions. However, there are still significant questions as to the role nonhydrostatic dynamical processes have on the local climate. Additional research is planned that will investigate the use of the CLIMOD method with model resolutions on the order of 1 km. As part of this additional research the impact of using a nonhydrostatic version of the MASS model will be investigated.
8. ACKNOWLEDGMENTS
The authors wish to thank the staff from the Defense Modeling and Simulation Organization, the Air Force Research Laboratory and the Air Force Combat Climatology Center who made this research possible. The authors also wish to thank Sandra F. Van Knowe for her invaluable assistance in proofreading this article. This work has been supported under grant number F49620-95-1-0523 from the Air Force Office of Scientific Research and Air Force Research Lab contract number F19628-97-C-0025.
REFERENCES
Blackadar, A. K., 1979: High resolution models of the planetary boundary layer. Advances in Environmental Science and Engineering, Vol. 1, No. 1, J. Pfafflin and E. Ziegler, eds., Gordon and Breach, pp. 50-85.
Bloom, S. C., L. L. Takacs, A. M. Da Silva, and D. Ledvina, 1996: Data assimilation using incremental analysis updates. Mon. Wea. Rev., 124, 1256-1271.
Doggett, M. K., M. Squires, and R. Kiess, 1999: An evaluation of the quality of local climate statistics generated from the output of a 3-D mesoscale atmospheric model. Preprints, 11th Conference on Applied Climatology, Dallas, TX, Amer. Meteor. Soc.
Hoke, J. E., and R. A. Anthes, 1976: The
initialization of numerical models by a dynamical initialization technique. Mon Wea. Rev., 104, 1551-1556.
Kaplan, M. L,
J. W. Zack, V. C. Wong, and J. J. Tuccillo, 1982: Initial results from a
mesoscale atmospheric simulation system and comparison with the AVE-SESAME-1
dataset. Mon. Wea. Rev., 110, 1564-1590.
Manobianco, J.,
J. W. Zack, and G. E. Taylor, 1996: Workstationbased real-time mesoscale
modeling designed for weather support to operations at the Kennedy Space Center
and Cape Canaveral Air Station. Bull. Amer. Meteor. Soc., 77, 653-672.
MESO, 1998:
MASS Ver 5.11 Reference Manual, MESO, Inc. Troy, NY 12180, 118 pp.
Therry, G. and P.
Lacarrere, 1983: Improving the eddy-kinetic energy model for planetary boundary
layer description. Bound.-Layer Meteor.,
25, 63-88.
Zack, J. W., K. T. Waight III, M. D. Bousquet, C. E. Graves, S. Yalda, and G. E. Van Knowe, 1996: An evaluation of local climate statistics generated from the output of a 3-d mesoscale atmospheric model.. Preprints, 11th Conference on Numerical Weather Prediction, Norfolk, VA, Amer. Meteor. Soc., 379-381.