PIRCS Boundary Conditions Development

Description and documentation of procedures used in its production
Last revised Sept 30, 1997



There are three domains for which the data are available:

domain (i)
A latitude-longitude grid that includes a limited domain of 101(E-W) x 51(N-S) grid points surrounded by an extended (by 10 grid points) "forcing frame." The borders of this grid were chosen so as to make its coverage similar to domain (ii). The SW corner is at (25 N, 125 W) and the NE corner is at (50 N, 75 W). The grid has a central latitude of 37.5 N, a central longitude of 100 W, and a grid spacing of 0.5 degrees. This configuration is depicted in Figure 2.1. An extended (by 10 grid points) "forcing frame" surrounds grid points spaced regularly at 0.5 degrees. The orientation of the inner domain is:
Central point      (LAT=26, LON= 51) (37.5 N, 100.0 W)
Lower left corner  (LAT= 1, LON=  1) (25.0 N, 125.0 W)
Upper right corner (LAT=51, LON=101) (50.0 N,  75.0 W)
domain (ii)
A grid based on a polar stereographic projection that includes a limited domain of 81(E-W) x 55(N-S) grid points surrounded by an extended (by 10 grid points) "forcing frame." This limited-area domain covers much of North America and it is defined by a polar stereographic grid with a central latitude of 37.5 N, a central longitude of 100 W, and a grid spacing of 60 km. This configuration is depicted in Figure 2.2. The orientation of the inner domain is:
Central point      (X_28, Y_41) (37.50 N, 100.00 W)
Lower left corner  (X_1 ,  Y_1) (23.07 N, 117.79 W)
Upper right corner (X_55, Y_81) (45.41 N,  70.50 W).
domain (iii)
A grid based on a Lambert conformal projection true at 30 N and 60 N that includes a limited domain of 101(E-W) x 75(N-S) grid points surrounded by an extended (by 10 grid points) "forcing frame." This limited-area domain covers much of North America and it is defined by a Lambert grid with a central longitude of 100 W, and a grid spacing of 52 km. The orientation of the inner domain is:
Central point      (X_28, Y_41) (37.50 N, 100.00 W)
Lower left corner  (X_1 ,  Y_1) (23.07 N, 117.79 W)
Upper right corner (X_55, Y_81) (45.41 N,  70.50 W).

In order to create the dataset for domain(ii) and (iii), a domain that encompassed (ii) and (iii) and which was a super-set of domain (i) was used together with bilinear interpolation.

Note that for these domains, the forcing frame is extended by 10 grid points in each direction. Thus for:

  1. domain (i), the forcing frame plus interior simulation domain is a 121(E-W) x 71(N-S) grid with SW corner at (20 N, 130 W), and a NE corner at (55 N, 70 W)

  2. domain (ii), the forcing frame plus the interior simulation domain is a 101(E-W) x 75 (N-S) grid with the same center point and grid spacing as the inner domain.

  3. domain (iii), the forcing frame plus interior simulation domain is a 121(E-W) x 95(N-S) grid with the same center point and grid spacing as the inner domain.
Files are provided which provide the latitude/longitude coordinate of each I,J point for domains (ii) and (iii).

The data are provided at all grid points in the forcing frame and interior simulation domain, and are available every 6 hours of a simulation period. After the initial time, users are requested to use only data in the forcing frame to drive their models, as agreed upon at PIRCS Workshop I.


The PIRCS dataset is derived from the NCEP Reanalysis Project dataset (Kalnay, 1996). The data for 1988 were obtained from the Data Support Service (DSS) at the National Center for Atmospheric Research (NCAR). Information about the NCEP Reanalysis Project dataset appears at the URL ftp://ncardata.ucar.edu/datasets/ds090.0 . A brief description of the NCEP spectral model and assimilation procedures used to create the Reanalysis dataset are described in Kalnay et al. (1996). PIRCS used the 'grbsanl' dataset (a subset of the NCEP Reanalysis dataset). For this subset, the data were on a T62 gaussian transform grid (192x94 grid points) at 28 sigma levels ranging from sigma = 0.00273 to sigma = 0.9985. The longitudinal spacing for the gaussian grid was 1.875 degrees and the gaussian latitudes, as well as the sigma levels, can be obtained from the ds090.0 website. The grbsanl dataset was distributed on 8 mm exabyte tapes, which are labeled as follows:

          Vol S26018 for January-June, 1988,
          Vol S26019 for July-December,1988,
          Vol S26006 for January-June, 1993,
          Vol S26007 for July-December,1993.

Inventory sheets for these tapes are attached; see section 8 for instructions regarding extraction of the files from the tape. Once the files are extracted, they have to be 'degribbed' before processing. The degribbing code is on pv14b1.vincent.iastate.edu in /local/users/rwturner/GRIBalpha, and is available, upon request, by anonymous ftp from this machine. Instructions on file retrieval from anonymous ftp sites are also given in section 8. (Throughout this document references are made to files at /local/users/rwturner. These are intended for internal use at Iowa State, but most can be made available upon request). A URL where information about GRIB messages can be obtained is ftp://nic.fb4.noaa.gov/pub/nws/nmc/docs . To degrib (or unpack) the data for each time period, an inventory of the PDS's (product definition sections) for each field you want unpacked has to be created. You can do this by formatting the unpkgrb1.dat file as follows, assuming the time period is 12 UTC 2 July 1988, and the grbsanl file is stored in the directory /PIRCS/NCEPgrbsanl

  FFFFFFFF  00000000  00000000  00000000  00000000  00000000  00000000  0
and executing the command unpkgrb1.x > inventory

The inventory file will contain PDS's for 172 fields of different variables (fields are horizontal slices along constant levels). Stackpole (1994) details the WMO format of the GRIB (edition 1) messages, and this can be consulted to decode the PDS's to determine which GRIB records need to be decoded. For PIRCS purposes the the required fields were T, q, u, and v at all 28 sigma levels as well as surface pressure and surface geopotential. Once the needed PDS's are determined, a new unpkgrb1.dat file has to be created - in which all the necessary PDS's are included. The format of this file is as follows:

  00001C02  0750FF80  0B6B26DE  5807020C  00010000  0A000000  14000001  0
  00001C02  0750FF80  0B6B265D  5807020C  00010000  0A000000  14000001  0
                      Var Sig     Date
  00001C02  0750FF80  01010000  5807020C  00010000  0A000000  14008001  0
  00001C02  0750FF80  07010000  5807020C  00010000  0A000000  14000000  0
  FFFFFFFF  00000000  00000000  00000000  00000000  00000000  00000000  0

It isn't necessary to create new inventory files for all time periods if you edit the date octet as appropriate in the 2nd unpkgrb1.dat file listed above (i.e., the one with all the needed PDS's already listed). Be aware the record is in hexadecimal so 88070212 is 5807020C. Some useful hexadecimal-to-decimal equivalents are:

00H=00D, 01H=01D, 02H=02D, 03H=03D, 04H=04D, 05H=05D, 06H=06D, 07H=07D,
08H=08D, 09H=09D, 0AH=10D, 0BH=11D, 0CH=12D, 0DH=13D, 0EH=14D, 0FH=15D,
10H=16D, 11H=17D, 12H=18D, 13H=19D, 14H=20D, 15H=21D, 16H=22D, 17H=23D,
18H=24D, 19H=25D, 1AH=26D, 1BH=27D, 1CH=28D, 1DH=29D, 1EH=30D, 1FH=31D,
20H=32D, 58H=88D, and 5DH=93D.

To create the output file test1.dat containing all the required fields, simply execute the command
The file test1.dat is then used for input by the routines that create the PIRCS dataset. Note, the programs in /local/users/rwturner/ncargraphics/global.conpack can be used to plot the degribbed reanalysis data.


a) Horizontal Interpolation

Assuming the NCEP reanalysis data have been unpacked correctly for all 28 sigma levels, they are then converted from the T62 Gaussian Grid to spectral coefficients. After the spectral coefficients have been computed the data are converted back to the PIRCS regularly spaced grids, domain (i) grid and sup-(iii), where sup-(iii) is a super set of domain (iii) that encompasses domain (ii). A bilinear interpolation [see Manning and Haagenson (1993) for details] is used to interpolate the data from sup-(iii) to domain (ii). Note, since the dimensions of domain (i) and (iii) are different, two slightly different sets of codes are maintained in two directories (/local/users/rwturner/gsm and /local/users/rwturner/gsm.hi). These codes are adapted from a global spectral model (GSM) that has been documented in Bourke (1974) and Nehrkorn and Hoffman (1985).

The file containing the Fortran code (there are many subroutines in this file) in which the conversion to and from the spectral coefficients is done is called gtend.f. Given the structure of the gtend.f routines, and a thrashing problem (swap space problems due to huge size of data arrays), it was more convenient to perform the vertical interpolations within the subroutine laloop. The vertical interpolation procedures are described in the next section. b) Vertical Interpolation and Extrapolation

Vertical interpolation from the 28 sigma levels to the 42 regularly spaced PIRCS pressure levels (every 25 mb from 25 mb to 1050 mb) is done within the laloop subroutine, and is based on a linear-in-ln p interpolation. The code for this was adapted from the sig2prs (sigma to pressure) subroutine of the INTERP which is an MM5 preprocessing program. This was done for all variables, except for specific humidity, q, which was first converted to relative humidity (RH). RH was then interpolated as for the other variables, except it was subject to the constraint that 0 < RH < 97%. Note, in computing RH, saturated vapor pressures were computed using a look-up tabular method used at NCEP.

For PIRCS pressure levels that were below the analysis surface (i.e., for pressure levels greater than the analysis p_sfc), extrapolation of the variables had to be done. The wind components u and v were set equal to u(sigma=28), and v(sigma=28) respectively. Temperature, T, was extrapolated along the moist adiabat that intersected T(sigma=28), and RH was extrapolated subject to the constraint that q was constant and equal to q(sigma=28). Reasons for the adopting the extrapolation procedures described above are elaborated upon in section 6. Once T and RH were known for all levels, q was calculated. Figures 4.1a (0.0176 MB), 4.1b (0.0175 MB), 4.1c (0.0175 MB), and 4.1d (0.0172 MB), show some initial NCEP reanalysis soundings over the PIRCS domain, and the resultant interpolated/extrapolated soundings. Note, there is some loss of information within the boundary layer and some smoothing of the profiles.


Some participants have requested that geopotential data be provided every 50 mb as part of the data-set. These data will not be distributed with the PIRCS data set, but will be available upon request. The geopotential was computed as follows: the surface geopotential (which is also the analysis topography) was part of the NCEP reanalysis data-set.

Geopotential heights were computed at each of the 28 levels through integration of the hypsometric equation, i.e.,

where "T_v bar" is the average (in ln p) virtual temperature of the k+1 -- k layer. Note; T_v = T (1 + 0.61q), and T and q are known at the sigma levels, and p_k = sigma*p_sfc, where K = 1,..,28 with sigma(1) = 0.0027 and sigma(28) = 0.9985. Note for k=28, geopotential was computed as follows,

In order to compute T_v in the surface to sigma=28 layer, the surface temperature and specific humidity have to be estimated, q is set to the value at sigma=28 and T is linearly interpolated in ln p between the pressure level at sigma=28 and the PIRCS pressure level immediately below the surface.

A linear in ln p interpolation is then done to get at the required pressure levels (i.e., every 25 mb from 25 mb to 1050 mb) above the surface. To get the geopotential at pressure levels below the surface, downward integration of the hypsometric equation is done using the values (previously extrapolated to the pressure levels) of T and q. Finally, after geopotential has been computed everywhere, some horizontal smoothing of the extrapolated values is done. Plots of the 850 mb and 1000 mb geopotential heights for 00 UTC 1 July 1988 are provided in Figures 5.1 (0.0170 MB), and 5.2 (0.0143 MB). When the Shuell method (an NCEP technique) was used to extrapolate temperatures a diurnal bias over the high terrain was evident (in regions of high terrain, heights below the surface were too large when a nocturnal inversion was present)


Surface pressure (Analysis p_sfc) will be distributed as part of the PIRCS analysis data set. Surface pressure was distributed as part of the NCEP reanalysis dataset, and it was fit to the PIRCS domains (i.e., (i), (ii), and (iii)) using the procedures described previously in section 4.a. However, since most PIRCS participants will have their own model topography (or model surface geopotential), which will be more detailed than the analysis topography (i.e., analysis surface geopotential - which is provided by PIRCS), a method for computing a model's surface pressure (Model p_sfc) is needed.

Figure 6.1: Schematic showing DEL Z.

Participants are urged to adopt the procedure outlined below for the purposes of the intercomparison in order to eliminate unnecessary sources of variation among the models. Code for the procedure is included in the distributed dataset (part of indat_gen.f). If anyone notices a problem with this method or can suggest improvements, please contact us. The procedure is as follows: Letting DEL Z = Model surface geopotential - Analysis surface geopotential (as in Fig. 6.1) and by use of the hypsometric equatioa,n an expression for Model p_sfc can be obtained as:

where Del Z' is given by

Here T_v can be replaced with an average virtual temperature for each layer (as was done in the geopotential calculation). Note, p_k are the PIRCS pressure levels that lie within the interval [Model p_sfc, Analysis p_sfc] and an initial guess for the Model's p_sfc has to be made, so that p_1,...,p_n can be determined and an estimate of the average virtual temperature in the Model p_sfc - p_1 layer can be made. The initial guess for Model p_sfc is assumed to be (Analysis p_sfc-(Del Z/9)).

A similar method is used to compute mean sea level pressure (p_msl). All that is required is that the Model surface geopotential = 0, so DEL Z= - Analysis Surface geopotential and p_1 has to be replaced by 1013.25 in Eq. 6.2. Sample calculations were carried out for domain (ii) with a terrain dataset from MM5. Plots of Model p_sfc and p_msl are for 00 UTC 1 July 1988 are provided in Figures 6.2 (0.0181 MB), and 6.3 (0.0146 MB). Note, for the MM5 terrain dataset, the magnitude of DEL Z was as large as 800 m over the high terrain of the Rockies. For the forcing frame, DEL Z became large over the southern Mexico mountain ranges and over the Canadian Rockies. ( Figure 6.4a(0.0172 MB) shows the analysis topography and Figure 6.4b (0.0173 MB). shows the MM5 topography.) Note, that the necessity of removing diurnal biases in psfc_M and p_msl dictated the extrapolation method for temperature described in section 4b. For example, attempts to preserve nocturnal inversions resulted in unrealistically high values of Model p_sfc (or p_msl) when DEL Z was large and negative (i.e., when the model's topography was well below the analysis topography). This was the reason for extrapolating temperature down the moist adiabat. The mesoscale models will develop their own inversions within the interior of the simulation domain, but it may be of interest to determine the sensitivity of the regional climate simulations to the temperature extrapolation method; i.e., what problems do we encounter if we rigorously preserve boundary layer features of the NCEP Reanalysis? Recall, a regional climate simulation does not have the same constraints as a forecast simulation. For example, contamination by errors in the boundary conditions that affect the timing of rainfall, but not the overall amounts, may not be as important in a climate simulation.


Daily SST boundary conditions were created for each of the PIRCS grids, where a day runs from 00 UTC - 00 UTC. To allow for possibly different land-sea masks among participating models, we give an SST value for every grid point on the PIRCS grid (even though we do not expect anyone to use the "SST" in Nebraska or Nevada!). The PIRCS-supplied land use files in the boundary conditions datastes can be used to provide a land/ocean mask (7 = Ocean/Lake, all other numbers = Land) for those that need it.

The SST are based primarily on SST in the NCEP reanalysis, supplemented by

Supplementary SST were used to fill gaps in the ocean/lake mask of T62 grid used by the reanalysis. Weekly AVHRR SST were assigned to each day of the week covered. When there were temporal gaps in the supplemetary SST, we interpolated linearly in time between known values if the gap was less than one month. If the gap occurred at the start/end of the PIRCS period and was less than one month, we assigned all missing days the value of the first/last value in the time series. Gaps greater than one month were treated as missing data. Although AVHRR retrievals are available every 18 km, we used AVHRR SST at just four points:

This restriction was used to keep potentially large numbers of AVHRR retrievals from dominating the SST analysis. We further required that SST for PIRCS grids in the Gulf of California use only input data east of Baja California, whereas SST for PIRCS grids in the nearby Pacific Ocean used only input data west of Baja, so that the relatively warm Gulf of California could be distinguished from the cooler Pacific.

SST on the PIRCS grids were computed using a Cressman scheme with 5 iterations. The radius of influence started at 1800 km and was halved on successive iterations. The large initial radius was chosen so that we could assign an "SST" to every grid point in the PIRCS domains.

The initial soil-moisture fields are provided in the form of a surface moisture availabilty (m). This gridded soil moisture data were obtained from the June 1988 NCEP reanalysis volumetric soil-moisture content in the uppermost 10 cm layer of the soil. The algorithm to obtain m from the volumetric soil-moisture content (eta) is as follows:

where (eta_s) is the volumetric soil moisture at field capacity and (eta_a) is the volumetric soil moisture at wilting point. The soil-moisture availability fields are in the files named SMAP60k.88060100 and SMAP0p5.88060100, and the field for PIRCS domain (iii) is shown in Figure 7.2(0.00833 MB).

NOTE: When participants initialize their model, they must impose a constant soil moisture profile.


The following individuals are thanked for their help: Chi-Fan Shih and Dennis Joseph of the DSS at NCAR, Masao Kanamitsu at NCEP, James Caveen and Clement Chouinard (Canada - COMPARE) Partial suupport was provided by the International Institute for Theoretical and Applied Physics (IITAP), by the Center for Global and Regional Environmental Research (CGRER) at the University of Iowa, and by the Electric Power Research Institute (EPRI).


Bourke, W., 1974: A multi-level spectral model. I. Formulation and hemispheric integrations. Mon. Wea. Rev. , 102, 687-701.

Brown, J.W., O.B. Brown, and R.H. Evans, 1993: Calibration of Advanced Very High Resolution Radiometer Infrared Channels: A New Approach to Nonlinear Correction. Journal of Geophysical Research, 98, 18257-18268.

Brown, O.B., J.W. Brown, and R.H. Evans, 1985: Calibration of Advanced Very High Resolution Radiometer Infrared Observations. Journal of Geophysical Research, 90, 11667-11677.

Kalnay, E., M. Kanamitsu, R. Kistler, W. Collins, D. Deaven, L. Gandin, M. Iredell, S. Saha, G. White, J. Woollen, Y. Zhu, M. Chelliah, W. Ebisuzaki, W.Higgins, J. Janowiak, K. C. Mo, C. Ropelewski, J. Wang, A. Leetmaa, R. Reynolds, R. Jenne, D. Joseph, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 76, 437-471.

Giorgi, F., C. S. Brodeur, and G. T. Bates, 1994: Regional climate change scenarios over the United States produced with a nested regional model. J. Climate 7,375-399.

Nehrkorn, T., and R. Hoffman, 1985: Documentation of the general truncation version of the global spectral model. AER technical memorandum-1.

Stackpole, J. D., 1994: GRIB (Edition 1) The WMO format for the storage of weather product information and the exchange of weather product messages in gridded binary form. Office note 388. US Dept. of Commerce, NOAA, NWS, NMC.

Takle, E. S., 1995: Project to Intercompare Regional Climate Simulations (PIRCS), Preliminary Workshop, 17 - 18 November, 1994. Bull. Amer. Meteor. Soc., 75, 1625-1626.

Welch, T. A., 1984: A technique for high performance data compression. IEEE Computer, 17,#6, 8-19


URL's of related websites.

- NCEP Reanalysis dataset information
- GRIB documentation

Original page created by Richard Turner (rwturner@iastate.edu)
25 June 1996

Return to PIRCS Boundary Conditions page

Copyright/Trademark Legal Notice