The National Institute for Space Research (Inpe) has implemented a weather forecasting computational model called the Ocean, Land, and Atmosphere Forecast Model (Monan). This system began operating this month, utilizing the Jaci supercomputer located at the agency's data center in Cachoeira Paulista, São Paulo.
Monan will be gradually introduced to replace Inpe's previous forecasting system, known as the Brazilian Global Atmospheric Model (BAM). According to information provided by the Ministry of Science, Technology, and Innovation (MCTI), the new model will raise the accuracy of Brazilian forecasts to the level of international agencies such as the UK Met Office, the National Oceanic and Atmospheric Administration (NOAA), and the U.S. National Weather Service (NWS).
Digital Perspective
Saulo Ribeiro Freitas, a physicist and researcher who heads the Numerical Modeling Division of the Terrestrial System (DIMNT) at Inpe, was interviewed about Monan. Freitas led the development of this model, which is already operational. An Inpe operations team ensures the issuance of forecasts twice daily, at midnight and noon, with a projection capacity of up to ten days.
This project represents an effort that started four years ago and incorporates several innovations. Among them, the community nature of Monan stands out, which includes the collaboration of Brazilian universities and science and technology institutes. Furthermore, it functions as a unified model, capable of supplying different time and space scales relevant to meteorological information, also attracting interest from entities outside Brazil, including Latin America, Central America, and the Caribbean.
Advantages of having a national model
Inpe based its development on foreign models, notably those from the Center for Weather Forecasting and Climate Studies (CPTEC), which have been improved over the years. Currently, Inpe uses its own model that capitalizes on more than thirty years of experience to create a more modern and integrated system. This single model can generate various climatic and meteorological products: severe weather forecasts, short-term forecasts (next few hours), medium-term forecasts (five to ten days), sub-seasonal forecasts (thirty days), and seasonal forecasts (next season).
Each of these products serves specific audiences. For example, seasonal forecasts are crucial for agriculture and for the management of water and energy resources. Medium-term forecasts benefit the general public and Civil Defense, assisting in the prevention and mitigation of natural disasters caused by atmospheric phenomena.
Challenges in model development
Regarding the four-year development time, Freitas explained that the process was relatively fast considering the complexity of atmospheric modeling. Weather forecasting is classified as one of humanity's most difficult techno-scientific challenges due to the chaotic behavior of the atmosphere, making the detailed representation of its movements and evolution extremely complex. Building an accurate and physically realistic model requires significant involvement from the modeling division, involving over 35 doctoral professionals.
This three-to-four-year period is considered standard for global meteorological centers to develop enhanced modeling versions. The models are written in computer language to run on supercomputers and provide timely forecasts; the code for such models can reach a million lines, demonstrating its high complexity, comparable to satellite development.
The complexity of weather forecasting is equated to simulating the human brain. From a mathematical point of view, it is the initial condition problem: although the mathematical rules for atmospheric evolution are known, it is necessary to know the current state of the atmosphere to project the future. This implies knowing the temperature, humidity, and wind speed at all points on the planet.
This data collection is carried out by a vast constellation of satellites launched by the most advanced countries, which monitor thermodynamic properties and send measurements to processing centers. Additionally, an extensive network of surface stations collects data on temperature, pressure, humidity, and wind globally. All this immense amount of data is processed to create a 'snapshot' of the current atmosphere, which is then input into the mathematical model running on the supercomputer, generating successive future images.
After generating the forecast, which is a massive volume of data, meteorologists interpret and disseminate the bulletins, with media outlets being responsible for the final distribution. Freitas emphasized that this daily routine depends on a colossal infrastructure, encompassing satellites, supercomputers, international data traffic, and fifty years of scientific research.
The new model promises greater accuracy and reliability. The main advance is a resolution four times higher than the previous CPTEC model, which operated in 20x20 kilometer grids, moving to 10x10 kilometers. This improvement increases spatial accuracy in locating events and estimating precipitation, in addition to allowing higher quality forecasts over longer time horizons.
Freitas clarified that models are inherently imperfect because perfect simulation of the atmosphere is impossible, and forecast quality declines with increasing time horizons. However, the milestone achieved after four years results in a more precise product than what is currently available on mobile devices.
Next steps for Monan
Weather forecasting development is a trajectory of continuous evolution spanning almost a hundred years, marked by constant performance gains. Freitas emphasized that this is an ongoing process ('work in progress'), as imperfections will persist despite improvements in observation, modeling, and computing.
Currently, Monan delivers medium-term forecasts, covering up to ten to fifteen days. However, there are plans to make new products available across different time scales. Sub-seasonal (30 days) and seasonal (three months) forecasts are still under development and are scheduled to be delivered to society within two years, according to the timeline. The Monan project, initiated in 2022, has a total duration of ten years, extending until 2031, when a complete model is expected to be available, capable of providing everything from torrential rain forecasts for the next six hours.