Skip to content

AI-Translated Page

This page was translated using artificial intelligence (AI). The content may contain inaccuracies or misinterpretations. For critical or official use, please consult the site administrator to confirm the information.

October 13, 2022

MINUTES 002/2022 - Revision 002

Meeting of the Scientific Committee of the Ocean, Land Surface, and Atmosphere Forecasting Model. At 3:00 pm on October 13, 2022, representatives of INPE (National Institute for Space Research), INMET (National Institute of Meteorology), UFCG (Federal University of Campina Grande), CENSIPAM (Management and Operational Center of the Amazon Protection System), USP (University of São Paulo), LNCC (National Laboratory for Scientific Computing), UFMS (Federal University of Mato Grosso do Sul), FAB (Brazilian Air Force), MB (Brazilian Navy), and MCTI (Ministry of Science, Technology, and Innovation) met virtually with the aim of continuing the work and discussions of the Scientific Committee of the Ocean, Land Surface, and Atmosphere Forecasting Model. These minutes record the meeting and gather the information entered in the chat, such as links and other information relevant to the discussions held. Following the opening of the meeting, led by institutional coordinators Saulo Freitas/INPE and Pedro Dias/USP, this document is oriented according to the guidelines established by Saulo Freitas during his presentation.

Saulo Freitas begins the meeting by welcoming all participants and announces the items to be discussed during the MONAN CC meeting.

Agenda 1 - New Members

Saulo Freitas, in the discussion of the first agenda of the meeting, announced the new members of the MONAN CC: 1. Celso von Randow from the Division of Impacts, Adaptation and Vulnerabilities ( DIIAV ) of INPE; 2. Renato Galante Negri from the Division of Satellites and Meteorological Systems ( DISM ) of INPE and 3. Antônio Marcos Mendonça as representative of MCTI. He also announced that member Daniel Alejandro Vila was invited to be a representative of the World Meteorological Organization ( WMO ), and that, for this reason, he will become an external member of WMO on the MONAN CC.

Agenda 2 - Retrospective and Actions

Saulo Freitas, in the discussion of the second agenda of the meeting, gives a retrospective of the events since the first meeting of the CC, comments on the participating institutions and that the CC currently has 35 members. He talks about the development of MONAN, which is included in INPE's strategic planning for the next 5 years and that the program has already been recognized as an MCTI program, with its own budget and an initial duration of 10 years. He comments on the agreement signed with the European Centre for Medium-Range Weather Forecasts ( ECMWF ) for access to the Open Integrated Forecasting System ( OpenIFS ) codes , access to data, holding seminars and training sessions and the production of a visualization program for unstructured grids using METView . He comments on the Project Opening Term (TAP) for the agreement with the National Oceanic and Atmospheric Administration ( NOAA ).

Still within the retrospective, Saulo Freitas comments that the current year (2022) was very important for the internal reorganization of INPE with a focus on MONAN: he mentions that between August 31st and September 2nd, an internal workshop was held by the General Coordination of Earth Sciences ( CGCT ) of INPE, with the objective of discussing the reorganization of the divisions of Numerical Modeling of the Earth System ( DIMNT ), Satellite and Meteorological Systems Division (DISSM) and Weather and Climate Forecasting Division (DIPTC) of the center around the development of MONAN. He comments that, in this reorganization, transversal groups (called macrogroups) were established, formed with members of all three divisions. He adds that other internal workshops were also held on the topics of atmosphere, oceans and cryosphere and data assimilation , in preparation for the developments of MONAN. He mentions that the organization of the macrogroups was structured and registered in the Electronic Information Service (SEI). Saulo Freitas also comments on the search for resources with the support of the director of INPE, Clézio Marcos De Nardin , and the coordinator of CGCT, Gilvan Sampaio de Oliveira . He mentions the approval of the INPE Computing Infrastructure Renewal (RISC) project, with a budget of R$200 million over 4 years, approved by FINEP . The project has 7 goals, 3 of which are intended for the operation of MONAN using the new supercomputer. Saulo Freitas also comments on the proposal for MONAN as a National Institute of Science and Technology ( INCT ), with a budget perspective of R$7 million over 5 years, managed by 11 members of the MONAN CC. He also comments on another search for resources, a budget request for 2023, with R$5 million per year, being approved as an MCTI program. With this, a proposal was submitted for a CT&I line to the National Fund for Scientific and Technological Development ( FNDCT ) for funding/capital. Another proposal that can be worked on is within the FAPESP Research Program on Global Climate Change ( PFPMCG ), within the theme of changes in land use and occupation and agriculture, strategies for reducing greenhouse gas emissions and adaptation to climate change.

Agenda 3 - Report of the PAD subcommittee

Luiz Flávio begins his presentation by commenting on the activities carried out by the Scientific Computing Group (GCC), one of which is the analysis of 3 dynamic cores for MONAN. Luiz Flávio comments on the formation of the team, composed of the employees Carlos Renato de Souza , Denis Magalhães de Almeira Eiras , Eduardo Georges Khamis , João Messias Alves da Silva from INPE and the employee Roberto P. Souto from LNCC. He comments that the team is being expanded and that they have dedicated themselves greatly to the work carried out. He mentions that the codes evaluated were the Global Eta Framework ( GEF ), Finite Volume 3 ( FV3 ) and the Model for Prediction Across Scales ( MPAS ), remembering that the FV3 and MPAS models have already been evaluated by NOAA. Luiz Flávio comments that the evaluations performed focused on the source code of the models and did not consider non-functional requirements, i.e., whether the dynamic cores represent the physics well or not. The functional evaluation of the models will be performed by the Model Evaluation Group ( GAM ) of INPE. The aspects of maintainability, portability, performance, scalability and efficiency were addressed for the evaluation. He mentions that all tasks were performed and recorded on GitHub and ZenHub of MONAN. In the first topic, maintainability, the MPAS model was the one that obtained the highest score (22 points), very close to FV3 (20 points). Regarding portability, the MPAS model also scored more points (8.2 points for MPAS and 4.4 for FV3). Luiz Flávio comments that the report with the rules, scores and complete results of this evaluation are available in the document Report of the Computational Evaluation of Dynamic Cores .

Luciano Pezzi asks Luiz Flávio about the negative points attributed to some packages used by the dynamic cores. Luiz Flávio comments that the points are removed because the installation of some of these packages is not trivial, which increases the complexity of using the source code of the models.

Next, Luiz Flávio invites Roberto Souto to explain the results of parallelism and code efficiency (code quality and Performance, Scalability and Efficiency - PEE), considering the MPAS 15 km model, in a 24-hour simulation without the model's physics. With technical problems from Roberto Souto, Luiz Flávio resumes the presentation. Regarding the code quality analysis, in the PEE aspect, the MPAS 15 km model was tested for predictions of only 24 hours due to time limitations and use of Dell's Rattler machine. In processing for Graphics Processing Unit (GPU), the MPAS model was faster than the FV3 model, even being faster than the FV3 model itself when used only with CPU processing. In general, in the PEE aspect, the MPAS model was slightly better than the FV3 model, although the FV3 model is easier to work with (it is simpler from the source code point of view). Luiz Flávio comments that the current version of the MPAS model loses in processing time to the FV3 model.

Saulo Freitas comments on the good work done by the GCC in relation to the dynamic cores evaluation report and then opens the floor for questions.

Pedro Dias asks Luiz Flávio if the documentation for the end user was also evaluated. Luiz Flávio comments that this was one of the requirements of the evaluation, which is the number of documentable lines of code. Pedro Dias adds to the question about the user manual and options for each model and if this was also evaluated. Luiz Flávio responds that for the FV3 model, this type of documentation was not available, but that the community organization is agile when it comes to user questions. Luiz Flávio adds that only the code documentation was evaluated.

Vinicius Capistrano asks about the coupling structure of the FV3 model and whether the evaluations took this into account as well. Luiz Flávio answers yes.

Saulo Freitas comments on the aspect of code complexity, with the difference between the MPAS and FV3 models being quite evident. In this sense, Saulo Freitas mentions the issue of community involvement, for which code simplicity is important. In this sense, he asks Luiz Flávio if the GPU programming aspect will be a difficulty for the community. Luiz Flávio responds by saying that, currently, programming for GPUs is no longer a major challenge and that the Compute Unified Device Architecture ( CUDA ) programming language has advanced enough to use OpenACC , which greatly facilitates programming by inserting directives in the source code, which indicate to the compiler which parts of the code should be processed on the GPU.

Enio Souza asks Luiz Flávio if GPU processing is a requirement of MONAN, and whether this requirement was already defined or not. Luiz Flávio responds that this is an aspect that is still open and that it will not necessarily be used directly. Despite this, GPU processing is more focused on applications that depend on artificial intelligence. He cites the article High-Performance Computing in Meteorology under a Context of an Era of Graphical Processing Units (Nakaegawa, 2022) which surveys the centers that use GPUs for processing numerical models. Most centers are still using CPUs.

Caio Coelho asks about the application of the dynamic cores of the MPAS and FV3 models for subseasonal, seasonal and climatic scales. Saulo Freitas responds that this issue, which involves physics, is yet to be studied and that in the current stage of evaluations, the objective is to test the quality of the dynamic cores to verify their computational efficiency. Pedro Dias comments that, for now, no center is applying the MPAS model for forecasts on these scales. Saulo Freitas adds that, for these scales, it is the physics of the model that dominates the model's forecasting ability and not the dynamics, since first-order errors are dominated by physics. He adds that, if we want a model ready for the seasonal scale, we should choose the FV3 model, but if we want a model to which we can contribute with our legacy and make it work competitively on all scales, we should choose the MPAS model. Antonio Manzi suggests creating a subgroup within the MONAN CC that would maintain contact with the National Center for Atmospheric Research ( NCAR ) staff to follow MPAS developments for the forecast scales that Caio Coelho addressed in his comment. Pedro Dias committed to talking to some of his contacts at NCAR.

Agenda 4 - Planning the Assessment under Meteorological Aspects

Moving on to the next topic, Saulo Freitas invites researcher Ariane Frassoni dos Santos de Mattos /INPE for her presentation on the functional evaluation with the comparison of the results of the dynamic cores of the MPAS and FV3 models.

Ariane Frassoni begins her presentation by addressing dynamic cores from a meteorological and source code quality perspective. She comments that the objective is to determine metrics for evaluating dynamic cores for use in MONAN. The characteristics of the functional aspects of dynamic cores involve suitability, accuracy, interoperability, compliance and security. During her presentation, Ariane Frassoni mentions the requirements for a non-hydrostatic dynamic core and that dynamic cores that do not have this characteristic are considered unsuitable for the intended applications for MONAN.

Saulo Freitas thanks the presentation and opens the discussion. Pedro Peixoto asks if the pre-processing and post-processing, which were used to generate the results presented by Ariane Frassoni, are public on GitHub, since there is an interest in gathering all the information in a single repository. He comments that he has some programs and scripts with which he can also contribute and that it is important to do these developments collaboratively. Saulo Freitas and Luiz Flávio agree with Pedro Peixoto's proposal and Luiz Flávio comments that these codes can be placed in the MONAN repository.

Pedro Dias comments that it is important to have objective diagnostic tools for evaluating numerical models, but that it is also necessary to make comparisons with the real observed data (e.g., radiosonde).

Actions for the Next Meeting

  1. Create a communication channel with NCAR developers to monitor the development of the MPAS model for applications at subseasonal to climatic scales;
  2. Create a collaboration channel between the DIMNT Model Evaluation Group (GAM) and Pedro Peixo's research group for sharing pre-processing and post-processing scripts and programs for MPAS;
  3. Talk to Roberto Souto about what is possible to execute on the LNCC Santos Dumont supercomputer for experiments with the MPAS and FV3 dynamic cores;
  4. João Gerd’s report on the DAS-DIMNT (Data Assimilation) Workshop.
Attachments