Emirom~lemal Y,q/iware, Vol. I I. Nos I 3, pp. 65 72, 1996 Copyright ~) 1996 Elsevier Science Ltd Printed in Great Brflain. All rights reserved 0266 9838/96/S15.00 + 0.0(I PII: S0266-9838(96)00020-2 ELSEVIER

The ANU translator: facilitating computer and data analysis of climate model outputs

D. E. Hyman, a D. R. Whitehouse, b J. A. Taylor? J. W. Larson? D. P. Hansen, a J. A. Lindesay ~

;'Centre for Resource and Environmental Studies, Australian National University, Canberra, Australia hANU Supercomputer Facility, Australian National University, Canberra, Australia ~Department of Geography, Australian National University, Canberra, Australia

Abstract

Computer visualization and data analysis tools are an essential component in the study and evaluation of climate model output. A number of useful visualization and analysis tools already exist, each with its own set of unique strengths. A translation tool is needed to convert data from the large number of model output Ikwmats into formats suitable for use with visualization and analysis tools. Such a translator would allow a researcher using a climate model to easily exploit a large number of visualization and analysis tools. Our goal is to develop a translation tool lk~r computer systems that will be highly portable, very simple to use, and easily expandable. The translation program, The ANU translator will feature a when run on any system with an XII interface. It will work with a number of climate model formats, including: CCMI, CCM2, RegCM2, MM5, CSIRO, and ANU-CTM. It will produce output in a range of data file lormats, including Vis5D, GRADS, and neICDF. NetCDF allows access to a range of analysis and visualization tools, including AVS, HDF, and GMT. VisSD, GRADS, and the tools compatible with netCDF will enable the user to view data statically or in animation, including the following critical options: (1) geographic distribution plots: (2) two-dimensional graphics: and (3) fully' three-dimensional renderings. It is intended that this translation package be made widely available for use by the climate modelling community as a research and teaching aid. Copyright a3 1996 Elsevier Science Ltd

Kevwords." Climate models; scientific visualization

1. Introduction model results must be achieved through the integration of vast amounts of data. Statistical methods are applied The climate modelling community relies heavily to some extent for the purposes of evaluating model upon data visualization for the evaluation and analysis performance, but it is often unclear whether or not of climate model results. Because there are no closed- such methods are strictly applicable to model results. form solutions for climate models, the analysis of Many statistical methods assume a random variable 65 66 D.E. Hyman et al./The ANU translator and/or a normal distribution; this poses a problem strengths. The most inhibiting factor for climate model- when analysing output from General Circulation Mod- lers wishing to access these tools is the problem of els (GCMs) because they are deterministic systems. In converting vast quantities of model output into formats addition to statistical analysis, numerical data analysis suitable for visualization and analysis tools. Our goal is used by climate modellers for preparing data for is to develop a computer program that will convert visualization, e.g. to time average the data before plot- large amounts of data from a number of model output ting. formats into formats suitable for inputting into a variety Scientific visualization is of value in the following of visualization and analysis tools. We aim to create ways: a program that will be highly portable, easily extend- ible, and very simple to use. We believe that such a • it serves as a diagnostic for model performance with data translation tool will be of considerable value to respect to observations; the wider climate modelling community. • it provides the modeller with a quick method of The conversion program, The ANU Translator, is checking that a model run is proceeding in a plaus- being written by Dena E. Hyman and Drew Whitehouse ible manner. This assists the modeller to prevent the at the Australian National University (ANU). The code squandering of valuable computer resources when, is written in ANSI C with a Tcl interpreter. It is due to human or computer error, a run is not designed to be highly portable, capable of running on proceeding as intended; any UNIX platform. The Translator is also designed • it helps the modeller compare the output derived to be simple to use and easily expandable. It interacts from sensitivity experiments with the output derived with the user through a Tcl/Tk user-friendly, graphical from control runs; interface, and can also be run in non-graphical mode • it provides illustrations for scientific articles and if an X11 interface or Tk are unavailable. Our goal is conference discussions; to incorporate the following models into The ANU • it provides visual aids to help scientists meet their Translator in the specified order: CCM1, CCM2, obligations to educate the public about the impor- RegCM2, MM5, CSIRO9, and ANU-CTM. A brief tance and implications of their work; description of each of these models follows in • it allows scientists to examine processes, which tend Section 1.1. Our goal is also to incorporate the follow- to incorporate the time-dependent behaviour of two ing output formats/tools into The ANU Translator in or more variables in concert; and the specified order: Vis5D, GRADS, and netCDF. • it aids scientists in locating displaced climatic fea- NetCDF allows access to AVS, GMT, HDF, and other tures that may be more difficult to detect through tools. A brief description of the output formats and statistical methods alone, e.g. a somewhat displaced tools mentioned here follows in Section 1.2. major circulation cell like the Icelandic Low.

One of the most extensive and widely used analysis 1.1. Model descriptions and visualization packages in climate modelling is the National Center for Atmospheric Research (NCAR) Modular Processor, which is compatible with the 1.1.1. CCM1. The NCAR Community Climate NCAR Community Climate Models (CCMs) (Buja, Model Version 1 (CCM1) is a 12-vertical-layer pseudo- 1992, 1993). The NCAR Modular Processor also reads spectral atmospheric global GCM with spectral resol- observational data written in CCM format, such as the ution R15, corresponding to a grid resolution of 4.44 ° European Centre for Medium Range Weather Forecasts latitude x 7.5 ° longitude. CCM1 can be run with an (ECMWF) data. The NCAR Modular Processor was optional interactive surface hydrology based on a sim- designed for use on the NCAR Crays and is highly ple bucket model. For a complete description of CCM 1, dependent on Cray architecture. see Williamson et al. (1987). It has only been successfully ported to other Crays and to the IBM RS/6000. In Cray binary representation, 1.1.2. CCM2. The NCAR Community Climate the integer, character, and floating point words are all Model Version2 (CCM2) is an 18-vertical-layer equal sized, 64-bit words. The Processor takes advan- pseudo-spectral atmospheric global GCM with spectral tage of this memory structure to manage whole blocks resolution T42, corresponding to a grid resolution of of mixed mode (integer/real/character) memory as a 2.8 ° latitude x 2.8 ° longitude. Compared with CCM1, large group of integers or as a large group of reals the physical parameterizations are improved upon for depending upon the task. the IBM RS/6000 has the a number of key climate processes, including: the rare property of being able to emulate this equal- planetary boundary layer, moist convection, transport, size memory construct, but other high-end processors and clouds and radiation. A diurnal cycle is added. do not. CCM2 can be run coupled to the Biosphere-Atmos- In addition to the NCAR Modular Processor, a num- phere Transfer Scheme (BATS) for more advanced ber of other highly useful scientific visualization and treatment of land surface processes. For a complete analysis tools exist, each with its own set of unique description of CCM2, see Hack et al. (1993). D. E. Hyman et al./The ANU translator 67

1.1.3. BATS. We plan to give The ANU Translator (DAR). Its resolution in spectral space is R21; in grid the capability of reading BATS output files. The pur- space it has 64 evenly spaced longitudes on the globe poses of BATS as coupled to the CCM are (Dickinson and 28 unevenly spaced latitudes in each hemisphere. et al., 1993): CSIRO9 includes a diurnal cycle and does not include any biosphere modelling. Three layers are used for the • to calculate the net exchange of thermal infrared computation of soil temperatures, with the upper layer radiation of different surfaces and the fraction of being the surface; soil moisture is computed in the incident solar radiation absorbed by them; reservoir method. Ocean grid points adjacent to sea • to compute temperature and moisture values at land ice are modelled as 50 m deep mixed layer ocean and sea ice surfaces; (MLO) points; the location of MLO points changes as • to determine the transfers of moisture, momentum, sea ice extent waxes and wanes. Expansion of the and sensible heat between the atmosphere and the MLO points to include the entire ocean is in progress. surface; and CISRO9 computes screen temperatures, facilitating • to compute values for wind, temperature, and moist- diagnostic comparisons with observational data. The ure at the level of surface observations, within veg- model includes a particle trajectory facility for analys- etation canopies, and in the atmosphere. ing flow patterns and estimating final positions of particles introduced into the atmosphere (e.g. vol- 1.1.4. RegCM2. RegCM2 is an updated version canoes, 1991 Gulf war oil fires). For a complete of the Penn State/NCAR Mesoscale Model Version 4 description of the model see McGregor et al. (1993). (MM4). Compared with a GCM, a mesoscale model has a limited spatial domain (usually continental scale 1.1.7. ANU-CTM. The ANU Chemical Transport or smaller) and relatively high spatial and temporal Model (ANU-CTM) is a three-dimensional Lagrangian resolution. MM4 is a compressible, hydrostatic, primi- model designed to advect air parcels in the atmosphere. tive-equation atmospheric model with a horizontal Car- Each air parcel represents a known mass of a tracer tesian grid and vertical terrain-following sigma coordi- gas in air. The tracers are advected according to a nates. The grid structure incorporates a staggering of stochastic scheme based upon ECMWF observational model variables in the horizontal and the vertical. For wind data on a 2.5°× 2.5 ° grid with seven vertical a complete description of MM4 see Anthes et al. levels. A mean and a time-varying component are (1987). The standard Kuo version of MM4 was included in the wind field. The flux of a trace gas is enhanced to produce RegCM2 by Giorgi et al. (1993) added or removed from air parcels at the lowest vertical who made the model more physically complete and level. Mixing ratios are computed by translating Lag- more computationally efficient. RegCM2 includes the rangian parcel coordinates to Eulerian grid coordinates. following improvements over MM4: The model incorporates diffusive mixing between air parcels (Taylor, 1989; Taylor et al., 1991). ANU-CTM • use of a split-explicit time integration technique; is currently being extended using 14 and 15 level T42 • incorporation of the radiative transfer scheme of ECMWF re-analyses, which cover the period 1985- CCM2; 1994 (Trenberth, 1992). The enhanced model will • use of BATS to parameterize the surface physics; include a boundary layer and stratosphere and the • utilization of an improved formulation of cumulus capability of simulating inter-annual variability. convection; and • inclusion of a lake model. 1.2. Descriptions of visualization and analysis tools

1.1.5. MM5. MM5 is the latest version of the Penn 1.2.1. Vis5D. Vis5D is a public-domain program State/NCAR mesoscale model. It incorporates a number for UNIX workstations designed for the interactive of improvements and increased options compared with visualization of large 5-D gridded data sets. It was the previous version. It includes the option of running fashioned with numerical weather models in mind. The the model in non-hydrostatic mode. For a complete authors of the most recent version are Bill Hibbard description of MM5 see Grell et al. (1994), and for and Brian Paul of the University of Wisconsin Space information about running MM5, see the following Science and Engineering Center. Vis5D is largely con- URL: trolled through a simple-to-use graphical user interface (GUI). It can be used to make full colour two-dimen- http://www.mmm.ucar.edu/mm5/mm5- sional and three-dimensional static or animated render- home.html. ings, and it has features for making contour line slices, coloured slices, and volume renderings. Vis5D has 1.1.6. CSIR09. CSIRO9 is a nine-vertical-level special provisions for wind vectors and trajectories and atmospheric GCM developed by the Australian Com- for constructing isosurfaces. A number of variables can monwealth Scientific and Industrial Research Organiza- be viewed and animated at once. Several different map tion (CSIRO) Division of Atmospheric Research projections are available including the ability to view 68 D.E. Hyman et al./The ANU translator data on a spherical globe, and images can be easily an interface for accessing the data, and a library that rotated with the mouse. Graphical output may be saved specifies implementation of the interface. A large num- and printed. For more information on Vis5D's require- ber of commercial and free software packages can read ments and capabilities, sample images of Vis5D output, netCDF data, including but not limited to the following: and information on obtaining Vis5D, please see the AVS, GMT, HDF, IDL, NCAR Graphics, and Spyglass following URL: Dicer and Slicer. CSIRO stores GCM results in netCDF form. Many more software packages are currently file://iris.ssec.wisc.edu/pub/www/ being expanded to include netCDF capabilities vis5d.html. (http://www.unidata.ucar.edu/ 1.2.2. GRADS. The Grid Analysis and Display Sys- packages/netcdf/). tem (GRADS) is a public-domain program developed by Brian E. Doty at the Center for Ocean-Land- 1.2.4. AVS. Advanced Visual System's Application Atmosphere Interactions at the University of Maryland. Visualization System (AVS) is a commercially avail- It is an interactive tool for the analysis, manipulation, able, highly generic and customizable, powerful vis- and display of earth science data on UNIX workstations ualization environment. For more information on AVS, and DOS-based PCs. GRADS expects input data sets see the following: to be described by four dimensions, usually: latitude, longitude, level, and time. GRADS includes the follow- 1. http://www.mcnc.org/HTML/ITD/IAC/ ing features: IAC.html, 2. http://www.avs.com/products/avs.html, (http : //grads. iges. org/grads / 3. the file avs.readme/WHAT.ISAVS available from gradover, html ) : the anonymousftpsite avs.ncsc.org. • it can handle station data and gridded data, including 1.2.5. HDF. The National Center for Supercomput- non-linearly spaced grids and variable resolution ing Applications' (NCSA) Hierarchical Data Format data; (HDF) is a public-domain multi-object file format that • different data sets can be both compared analytically expedites the transfer of various data types between and graphically overlaid; different machines and operating systems. HDF data • user-defined or built-in functions can be performed sets can be n-dimensional and are self-defining and on the data interactively; easily extendible. HDF includes utilities to create • graphics capabilities include: bar graphs, line graphs, images scatter plots, contour lines and shaded contours, wind vectors, streamline plots, grid and shaded grid (http://hdf.ncsa.uiuc.edu:8001). boxes, and station model plots. The user can control many aspects of the graphical output; the default 1.2.6. GMT. Generic Mapping Tools (GMT) is a settings are geophysically intuitive. Graphics may package created and maintained by Paul Wessel and be output in Postscript for colour of monochrome Walter H. F. Smith. It is a collection of about 50 printing; UNIX tools for manipulating 2D and 3D data and • GRADS includes an interpreted scripting language. producing illustrations. GMT is freely available to Scripts can be written to include widgets and accept government agencies and non-profit educational organi- mouse input from the user and to automate complex zations worldwide calculations or displays; and • GRADS may be run in batch mode. (http://www.soest.hawaii.edu/soest/ For more information on GRADS, including sample gmt.html). images and instructions on how to obtain the software, please see the following URL: 2. Methods http:://grads.iges.org/grads/ head.html. 2.1. Structure of The ANU Translator

1.2.3. NetCDF. The network Common Data Form The The ANU Translator is being designed with the (netCDF) is a widely used, freely available format, following three goals in mind: (1) portability, (2) ease interface, and library for the creation, access, and of use, and (3) extensibility. Portability is achieved sharing of scientific data. It was created by Russ Rew, by the avoidance of implementation-specific coding: Glenn Davis, and Steve Emmerson at the Unidata ANSI C is used with a standard Tcl/Tk interpreter Program Center. NetCDF provides a machine-inde- (Ousterhout, 1994). Tcl and Tk encapsulate the graphi- pendent format for the representation of scientific data, cal and the non-graphical user interfaces. The GUI is D. E. Hyman et al./The ANU translator 69

Fig. 1. The graphical user interface (GUI) of The ANU Translator. designed so that a user can run the Translator without The C code has four basic components: (1) Introduc- relying significantly on documentation (Fig. 1). The tion, (2) Interrogation, (3) Extraction, and (4) Output. ease of use should assist researchers and greatly facili- The 'Introduction' module receives the user's request tate the introduction of climate modelling to students. for file interrogation from Tcl. It determines whether The extensibility of The ANU Translator is being the user's requested model type is within the Trans- achieved primarily through the modular nature of the lator's capabilities and calls the appropriate 'Interrog- C code and the use of an internal representation type ation' module. There is one 'Interrogation' module for common to all input and output types (Fig. 2). each general type (for coding purposes CCM1 and

INTRODUCTION ] Obtain user's interrogation request. / Determine validity of user s request. / Call appropriate file interrogation command.]

~INTERR!GATION~~ Interrogate [ Interrogate [ I Interrogate II Interrogate CCM1 ~'CM2 ] [ Re~42/MM5 CSIRO9 ] [ ANu-crM

Present interrogation information to the userI and obtain any extraction information..I

EXTRA[CTION I Extract Extract Re~2/MM5 CSIR09 ANU-CTM

Internal Representation ou4ptrr f I Write GRADS II wno DF I /

Fig. 2. The structure of The ANU Translator program. 70 D.E. Hyman et al.IThe ANU translator

CCM2 are considered one general model type, as are instructions. We hope that other researchers will add RegCM2 and MM5). The 'Interrogation' modules read modules to the code to increase its range of model the model output files requested by the user and indi- input types and its analysis and visualization output cate what information is present in the files. This types. We believe that The ANU Trimslator will assist information is presented graphically when the program the research and teaching purposes of the climate is being run in graphical mode. The Translator then modelling community. receives any extraction requests from Tcl. An 'Extrac- tion' module exists for each general model type. Data is extracted from each model into a common internal 3. Data and results representation. The 'Output' stage proceeds from this internal representation format, with each different out- The RegCM2 output shown (Fig. 3) is from a run put type having its own 'Output' module. We hope to conducted by Jay Larson on the Fujitsu VP 2200 at include the ability to share output tools, especially ANU. The model run incorporates a 17-vertical-level where a visualization tool does not include adequate grid with a horizontal resolution of 16km, and it numerical analysis capabilities, e.g. data could be aver- begins on radiation day 1 December 1991. The grid is aged in GRADS and then funnelled into Vis5D. centred over Japan at 36°N, 139°E, and is subdivided Rather than having one reading and writing routine into 40 latitudes and 41 longitudes. The output shown for each input-to-output conversion type, the Translator in Vis5D is for the last time step of the second day, is designed so that each input type requires only one i.e. 2 December 1991, 18Z. The topography and the reading routine, and each output type requires only map boundaries are taken from the Vis5D data sets. one writing routine. A researcher wanting to add an The CSIRO9 output shown (Fig. 4) is from a input model type to the Translator should be able to 2 X CO2 Greenhouse run (660ppmv) conducted by do so by adding new 'Interrogation' and 'Extraction' Martin Dix and Ian Watterson of CSIRO DAR on modules and making small adjustments to the 'Intro- the CSIRO Cray Y-MP in August-October 1991. The duction' section and the GUI Tclfrk code. Similarly, 2 X CO2 run was initialized from the state at the end a new output type can be added easily by writing one of the 10-year climatology run described in McGregor 'Output' module.

2.2. Hardware and software requirements

The ANU Translator should run on any UNIX-based system. Because climate model output data are usually extensive, the Translator is being designed in such a way that all of the data being analysed together will not have to be in memory at once. Actual memory limitations and disk space requirements are not known at this time. Compilation of The ANU Translator requires an ANSI C compiler and a Tcl installation. To run the graphical version of the Translator, Tk and an X interface are also necessary. We expect to make the Translator compatible with the most recent non-beta versions of Tcl, Tk, Vis5D, GRADS, and netCDF avail- able at the time of release. Input to the Translator will be in the standard format for the particular model type being input. An installation of one or more of the packages compatible with the Translator's output will be needed for the actual visualization and/or analysis. The C compiler, Tcl, Tk, most of the climate models, and most of the analysis and visualization tools dis- cussed in this paper may all be obtained free of charge via the Internet. Fig. 3. Output from RegCM2 run over Japan displayed in Vis5D using a rectilinear projection. The time step is 18Z 2.3. Availability on day 2 of the run, corresponding to radiation day 2 December 1991. The horizontal surface slice shows model topography (topo) in m. The lightly shaded vertical slice We plan to make The ANU Translator freely and shows the zonal wind (u), in m s-k The dark 3D surface is publicly available via anonymous ftp and the World the temperature (T) isosurface at 230 K. The black lines on Wide Web, complete with easy-to-follow installation the surface indicate political boundaries. D. E. Hyman et al./The ANU translator 71

4. Conclusions

The ANU Translator is a work in progress. We

Acknowledgements

A grant from the Quality Assurance Program of the Department of Employment Education and Training supported this work (DEET R30884). We thank CSIRO Division of Atmospheric Research for providing the CSIRO9 data.

References

Anthes, R. A., Hsie, E. & Kuo, Y. (1987) Description of the Penn State/NCAR mesoscale model version 4 (MM4). NCAR Tech. Note, NCAR/TN-282+STR, 66 pp., Boulder, Colorado. Buja, L. E. (1992) Introduction to the UNICOS CCM Fig. 4. Output from the CSIRO9 2 × CO2 (660 ppmv) dis- processor. NCAR Tech. Note NCAR/TN-383+IA, played in Vis5D. Data are July 30-year averages. A vertical 62 pp., Boulder, CO. slice of temperature (7") in K is shown over the equator. The horizontal uv wind vector slice is in m s '. The isosurface Buja, L. E. (1993) CCM processor users' guide (UNICOS shows the zonal wind (u) at 50 m s '. Version). NCAR Tech. Note, NCAR/TN-384+IA, Boulder, CO. Dickinson, R. E., Henderson-Sellers, A. & Kennedy, P. et al. (1993). The 10-year climatology run relied on J. (1993) Biosphere-atmosphere transfer scheme prescribed sea surface temperatures (SSTs). Surface (BATS) Version le as Coupled to the NCAR Com- heat fluxes from the 10-year climatology run were used munity Climate Model. NCAR Tech. Note, to generate implied ocean heat fluxes (q-flux); these NCAR/TN-387+STR., 72 pp., Boulder, CO. were supplied to the 50 m mixed layer ocean model Giorgi, F., Bates, G. T. & Nieman, S. J. (1993) The included in the Greenhouse run. The 2 X CO 2 run was multiyear surface climatology of a regional atmos- allowed to equilibrate for 27 years and then continued pheric model over the western United States. Jour- tbr 30 additional years to form a climatology. The nal of Climate, 6, 75-95. data shown are the 30-year July monthly averages. Grell, G. A., Dudhia, J. & Stauffer, D. R. (1994) A Several features of Vis5D are illustrated by the description of the fifth-generation Penn State/NCAR RegCM2 and CSIRO9 output displayed here, including: mesoscale model (MM5). NCAR Tech. Note, NCAR/TN-398+STR, 138 pp., Boulder, CO. • horizontal and vertical coloured slices; Hack, J. J., Boville, B. A., Briegleb, B. P., Kiehl, J. T., • two-dimensional wind vector maps; Rasch. P. J. & Wilfiamson, D. L. (1993) Description • three-dimensional isosurfaces; of the NCAR Community Climate Model (CCM2). • ability of the user to choose among three-dimen- NCAR Tech. Note, NCAR/TN-382+STR, 108pp., sional projection types; Boulder, CO. • display of the map and topography data sets built into Vis5D; and McGregor, J. L., Gordon, H. B., Watterson, I. G., Dix, • simultaneous display of multiple variables. M. R. & Rotstayn, L. D. (1993) The CSIRO 9- level atmospheric general circulation model, CSIRO Static, black and white, two-dimensional images do Division ~f Atmospheric Research Tech. Paper not adequately convey the power and flexibility of No. 26, 89 pp., CSIRO, Australia. computer software, especially GUls and visualization Ousterhout, J. K. (1994) Tcl and the Tk Toolkit. 485 pp., packages. At MODSIM '95 we will present a computer Addison-Wesley, Reading, MA. display of The ANU Translator and sample model Taylor, J. A. (1989) A stochastic Lagrangian atmospheric output animated within Vis5D. transport model to determine global CO2 sources 72 D.E. Hyman et al.IThe ANU translator

and sinks--a preliminary discussion. Tellus, 41B, and atlas of 1000 to 10 mb of circulation statistics. 272-285. NCAR Tech. Note, NCAR/TN-373+STR, 191 pp., Taylor, J. A., Brasseur, G. P., Zimmerman, P. R. & Boulder, CO. Cicerone, R. S. (1991) A study of the sources and Williamson, D. L., Kiehl, J. T., Ramanathan, V., Dickin- sinks of methane and methyl chloroform using a son, R. E. & Hack, J. J. (1987) Description of global three-dimensional Lagrangian tropospheric NCAR Community Climate Model (CCM1). NCAR tracer transport model. Journal of Geophysical Tech. Note, NCAR/TN-285+STR, 112 pp., Boulder, Research, 96(D2), 3013-3044. CO. Trenberth, K. E. (1992) Global analyses from ECMWF