The new EGU journal: Geoscientific Model Development
Geoscientific Model Development (GMD) will be an international scientific journal dedicated to the publication and public discussion of the description, development and benchmarking of numerical models of the Earth System and its components. Manuscript types considered for peer-reviewed publication will be: model descriptions, model inter-comparisons, benchmarking papers, and technical papers. European Geosciences Union
The new EGU journal
Geoscientific Model Development
Geoscientific Model Development (GMD) will be an international scientific journal ded-icated to the publication and public discussion of the description, development and benchmarking of numerical models of the Earth System and its components. Manuscript types considered for peer-reviewed publication will be: model descriptions, model inter-comparisons, benchmarking papers, and technical papers.
1 Introduction
Climate science has risen to significant public prominence in recent years, due to the realisation that anthropogenic climate change threatens far-reaching consequences, both for the planet as a whole, and for society. This has coincided with the develop-ment of complex numerical models of the various components of the climate system, enabled by the continuing exponential growth in computing power. Taken together, these last two conditions have made possible the field of Earth System Modelling. In addition to being of intrinsic scientific interest, the outputs of climate and Earth System Models are of interest to policy makers. Therefore, it is of great importance that the whole process of science is undertaken as effectively, rigorously, and as transparently as possible.
Numerical models are the cornerstone of contemporary climate science and Earth System Modelling. Their development is a difficult and time-consuming task; it requires a combination of skills, both scientific, as well as technical. A complex numerical model is rarely ecompletef; in general, new parameterisations and numerical methods are continually being developed. In addition, models require ongoing maintenance; all complex software can be expected to contain errors, and this is true of Earth System Models and their components as much as anything else.
Current journals in the field of geoscientific modelling (e.g. eOcean Modellingf or eCom-puters & Geosciencesf) focus on scientific results but not on the quality of the model description, or the traceability of incremental model versions. It is clear that our present model for scientific publication does not encourage rigor and transparency in model development. In all the sub-fields involved in Earth System Science, including those areas where the complex interactions between the Earth System components are studied, manuscripts which include comprehensive model descriptions are routinely drastically reduced with many details which may be critical to the results being ex-cluded at the review stage. A common request is that model description be minimised and new scientific enoveltyf be maximised. Indeed if there is insufficient novelty then journals are likely to reject the papers as scientifically lightweight (and correctly so, given the journalsf remits).
Yet before scientific discoveries can be made, the models must first be built and tested. In many models, various ingenious techniques are implemented to achieve a eworkingf model. The results of changing a module, or adding a new component, are often unexpected, and fine-tuning may be required to generate acceptable results. The knowledge gained from these adventures in model development goes largely undocu-mented. Occasionally the final model version may be published as an internal techni-cal report, yet it is rare to see proper documentation of the techniques used to achieve this, which could really enhance the field of model development. This may lead to a duplication of efforts by others and thus wasted time. Indeed there is little incentive for scientists to spend time making these reports in any way comprehensive. The pres-sure is always for scientists to produce and publish scientific ediscoveriesf, since this is what drives career progression.
It is with this background in mind that we propose the creation of a new journal specif-ically designed to focus on the description, development and benchmarking of numer-ical models of the Earth System and its components. In encouraging full publication of Earth System Models we have two main goals. The primary goal is to promote the efficient and effective development of the models, through the clear presentation of the techniques from which all other developers can improve their own models. A secondary goal is to provide increased credibility to the Earth System Science field by creating a space within which models can be openly presented and critically dis-cussed, and their results reproduced and validated. A welcome side-effect will be the formal, peer-reviewed, recognition of the work of Earth System Model developers.
2 General description of the journal
The name of the new journal will be eGeoscientific Model Developmentf. It will be
...........................................................................................
available on the internet at and
.............................................................
. Here, the abbreviation eGMDf is used from now on.
The main purpose of GMD is to promote model development as a serious and worth-while activity, by providing a home for papers covering a wide range of aspects of the subject. In particular, we believe the following are currently inadequately provided for in the literature:
. Detailed descriptions of numerical models, covering the processes represented, their mathematical formulation, the numerical methods used, the coding struc-ture, and any other relevant technical information.
. A recognised process for recording the evolution of a model though different ver-sions, with an expectation of being able to distinguish those versions by number or some other label.
. Model benchmarking, including determining the most appropriate benchmarks to use for a given type of model.
. Studies of the dependence of model output and performance on technical con-siderations.
. A forum for the discussion of best practice affecting all these areas.
3 Examples of potential papers
This section gives some examples of papers which would be suitable for publication in GMD. Note that this list is far from exhaustive, and is intended solely to give a fiavour of the sort of papers which would be considered for publication.
3.1 Model description papers
It is anticipated that model description papers will form the backbone of GMD. These will comprehensively describe the underlying science behind the models, and will also include details often omitted from more traditional papers, such as the numerical schemes employed. The papers should be somewhat more advanced than internal technical reports. For example, the inclusion of discussion of the scope of applicabil-ity and limitations of the approach adopted is expected. In order to enable full peer review of the models, evidence of model output should also be provided, with com-parison to standard benchmarks, observations and/or other model output included as appropriate. The publication will potentially consist of three parts: the main paper, a user manual, and the source code (ideally supported by some summary outputs from test case simulations):
. The main paper:
The main paper focuses on the description of the underlying physics and math-ematics of the model. In addition, the model description should also include the following items:
. model name and version number (preferably both in the manuscript title)
. model home page on the web (URL)
. system requirements (e.g. PC, workstation, supercomputer)
. software requirements (e.g. a particular Fortran compiler)
. license info (e.g. GPL, or code available from authors upon request)
. A User Manual: The user manual should contain all information needed for a new model user to install, run, and work with the model. The user manual should be provided by the authors as a separate pdf file. It will be available as an electronic supplement to the main paper. Although the preparation of a user manual is not mandatory, it is strongly recommended.
. The model source code: Although not mandatory, it is strongly recommended to include the model source code in the electronic supplement. Although we will not archive large data sets,
summaries of model test cases (such as standard benchmarks and statistics) will also be welcome.
Model description papers will also usually contain an assessment of the modelfs per-formance relative to benchmarks, or standard observational datasets. Likely reasons for discrepancies with the data, and improvements/worsening relative to previous ver-sions, should be discussed in some detail. It is anticipated that subsequent versions of the model, including bug-.xes, will be published in additional papers, and linked to the original model description. For more details, see section 4.2 of this document . Versioning.
3.2 Development papers
Papers which describe technical developments relating to model improvement will be welcome. For example, improvements to the speed or accuracy of numerical integra-tion schemes, or new parameterisations for processes represented in modules (e.g. cloud formation, ocean mixing, biogeochemical processes. . . ). Authors would be en-couraged to provide code to perform test cases described in the paper. Although some of these topics could potentially find homes in existing journals (e.g. International Jour-nal for Numerical Methods in Fluids), we are not aware of any outlet that covers of all Earth System Science and provides the ability to publish code with test cases.
3.3 Benchmarking papers Papers will be encouraged which discuss work on developing new benchmarks for assessing model performance, or novel ways of comparing model results with obser-vational data.
3.4 Technical papers GMD will also publish papers related to technical aspects of running Earth System Models, and the reproducibility of results. For example assessments of their perfor-mance with different compilers, or under different computer architectures.
3.5 Model Inter-comparison project desciption papers In the climate sciences there are now a plethora of MIPs (Model Inter-comparison Projects) such as CMIP, PMIP, C4MIP etc. The aim of these projects is to compare behaviour across a range of models of a certain type by forcing them with common boundary conditions. While the science developed from these projects is readily pub-lishable, reference works which describe in detail the model configurations and exper-iments are rare. Some information is usually available on the website of the projects but not in a form that is readily accessible to the wider community. These papers would be expected to contain description of the experimental details and the project protocol, including: discussion of why particular choices were made; highlighting of differences in the application of the protocol by the different groups; and including sufficient de-scription/.gures of model results to give an overview of the project. The versioning of MIPs could be documented using a similar format to that for the model description papers.
4 Publication model
The editorial board of GMD will consist of several editors covering the different areas of Earth System Modelling. The editors of GMD are expected to edit about 5 manuscripts per year. Depending on the number of editors and submitted manuscripts, this number may vary.
GMD papers will be freely accessible to all via the web (eopen accessf), and financed through page charges. It may be possible to waive these page charges for an initial period until our new journal is established. Page charges are calculated based on the number of pages in the main paper. There is no extra charge for the user manual or model code in the electronic supplement.
4.1 The review process
Like several other EGU journals, GMD will be published in two stages: First, the manuscript goes to the discussion journal (eGeoscientific Model Development Discus-sionsf, GMDD) for public peer review, and after successful completion, a revised paper ..................... will be published in GMD (for details of this publication process, see also ................................................................................................................................................. ). In contrast to other journals, the questions asked to the reviewers are slightly different:
. Is the model description solid, and the user manual understandable?
. The production of ground-breaking scientific discoveries from the model are not a review criterion. Instead, the question is whether the model has the potential to perform calculations that will lead to correct and scientifically important results.
. We cannot expect the reviewers to examine the model code in detail or directly verify model output. Bitwise reproducibility of model outputs across a range of platforms is beyond our scope. The goal should be seen as gscientific repro-ducibilityh rather than gexact reproducibilityh: that is, the reviewer should be con.-dent that the model could be constructed to a level which would enable scienti.-cally equivalent results to be generated. Of course, in simpler cases (or with code which is already widely ported) it would be preferable if the code could be directly tested, and the referee will be invited to comment about the code structure and whether the code is well written or not.
4.2 Versioning
The electronic publication system of GMD should allow model descriptions to be easily updated in a clearly traceable manner. Assuming a standard version of the model has been comprehensively described in GMD in a full model-description paper, the authors will have the possibility to write articles about further model versions. The concept for this is fiexible. Small changes such as bug-.xes can be published as a minor paper which cites and is electronically linked to, the original paper. More substantial changes could be treated as a new version of the original eMajorf paper, highlighting the changes. This concept is similar to that of the eliving reviewsf journal ......................................................................... (), although in our case it is not related to review articles. Note that the papers about the old model versions are not deleted. Once published, everything is permanently archived and citable in GMD. Different model versions can be grouped together into a Special Issue of the journal. On the web pages of GMD, it will be easy to see which publications refer to the same model, and if appropriate, which refers to the latest version.
There is no intention to enforce a version control system, such as CVS, for the model code. Authors should simply be asked to create a zip file of their code and user manual which will then become the electronic supplement of their paper. Some sample output files could also be archived to aid reproducibility.
This document compiled by:
James Annan., Julia Hargreaves., Dan Lunt., Andy Ridgwell., Ian Rutt., Rolf Sander. .Frontier Research Center for Global Change, Yokohama City, Kanagawa, Japan .School of Geographical Sciences, University of Bristol, UK .Department of Geography, Swansea University, UK .Air Chemistry Department, Max-Planck Institute of Chemistry, P.O. Box 3060, 55020 Mainz, Germany