HPC role in forecast and climate research

High performance computing systems (HPC) provide a great help to meteorology and climatology, mainly in numerical simulations for weather forecasting and climate surveys. The data necessary to process the weather forecast come from different sources of detection and observation, from probes and weather stations, radars and so on. The variables collected through these sources are many and heterogeneous: humidity in the air, temperature, atmospheric pressure, intensity and direction of the winds. In addition, the data are compared with time series. Finally, data analysis is based on very complex mathematical models: this is why High Performance computing facilities are required to obtain information in a short time. 

New forecasting paradigms

The demand for increasingly accurate weather forecasts, together with the scientific development of atmospheric models, has led over the years to an increase in the complexity of the physical parameterizations of the models themselves. The consequence was an increase in the resolution of the maps, thus implying an even greater request for processing and storage resources necessary for the management of the simulations.

Furthermore, in order to improve the forecasts with information on their reliability and increase the time interval of the forecast itself, in recent years the processing of probabilistic forecasts, "ensemble forecasting", has been introduced, i.e. the meteorological model is performed several times under conditions slightly different, inserting "small perturbation" elements that modify the initial conditions, given the chaotic nature of the equations that govern the model itself. Also in this case, in order to reduce the execution times, high computational resources are necessary, because all the simulations of the model must be performed as much as possible in parallel mode (several simulations performed simultaneously).

With the appearance of the probabilistic forecasting model, therefore, the amount of data needed to process the predictions has grown by an order of magnitude compared to the deterministic forecasting method. Today the need to use supercomputing resources arises not only in the integration phase of the model equations, but also extends to the next phase of post-processing of the results.

The computing infrastructure to support the weather forecast

The typology of services that are usually requested in the meteorological-hydrological field integrates the typical problems of supercomputing systems, therefore in particular the computing power, with those of highly critical services, therefore in particular speed, availability and high reliability. The operation of a forecasting service that uses supercomputing facilities must work in automatic mode 365 days a year, with rigorous timing and with a remarkably high expectation of reliability and infallibility. To provide a service of this kind, each element of the infrastructure must be extremely reliable and in redundancy. The IT environment must be stable over time and resist any critical event that can compromise its functionality.

Cineca and weather forecasts: a thirty-year collaboration

In order to guarantee such a complex and critical service as that of the weather forecast, therefore, skills and resources of proven reliability and efficiency are required, as the consolidated over time experience. Cineca's expertise in this area is long-standing. For over 30 years, Cineca's supercomputers have been developing and providing detailed weather forecasts for Italy, in particular for the Central Functional Center for Risk Management of the National Civil Protection and its Crisis Units, in collaboration with the Regional Meteorological Service of Emilia-Romagna, ARPA Piedmont and Italian Air Force.

Since 1993 Cineca has been a partner of Arpae in the activities relating to the service of national meteorological forecasts and fulfills the function of supercomputing center for meteorological forecasts.
Moreover, Cineca has been actively collaborating for a long time also with the National Institute of Oceanography and Experimental Geophysics (OGS) of Trieste, which produces data relating to biogeochemical analyzes relating to the Mediterranean sea as part of the "Copernicus: marine environment monitoring service" project : an European service that provides information on the quality of the sea and the environmental conditions of the seas and oceans that line the continent.

Cineca also collaborates with Arpa Piemonte by providing dedicated calculation resources for air quality forecasting models. Thanks to the collaboration with Arpa Piemonte, Cineca actively participated in the development of the parallel version of the FARM air quality model owned by the ARIANET company.

Cineca also provides supercomputing services to the Euro-Mediterranean Center on Climate Change (CMCC) and CIMA research foundation.

Cineca's skills in weather and climate field

The skills of Cineca researchers in the field of meteorological and climatological forecasts concern five main fields:

  • management and administration of supercomputing systems;
  • parallelization, optimization, profiling and debugging of source codes;
  • knowledge of forecast models from an IT point of view;
  • management of high frequency services and high criticality of use in supercomputing environments;
  • management of high frequency and high criticality services in supercomputing environments.

To maintain a high level of reliability, the technological and operational infrastructure has been set up for specific solutions.

The operational workflow

For the management of the weather service, Cineca researchers have designed and developed a customized operating chain, that is a flow of commands, a fully automated workflow, where human intervention is occasional and necessary only in critical cases. The workflow autonomously executes a series of commands that manage the different phases of the flow, i.e. the pre-processing of the input data, the execution of the forecast models, the post-processing of the output data and finally their archiving for processing next. The operating chain has been perfectly integrated with different types of operating systems available on supercomputers.

For Arpae and Civil Protection Cineca manages several operating chains, which use over 1000 supercomputer cores. The final data is made available within two hours from receiving the input data:

  • Cosmo-5M: 18 and 72 hour forecasts for the Mediterranean bed on areas with a grid step of 5 km
  • Cosmo-I2: 48-hour forecasts on the Italian national territory, on areas  with a grid step of 2.2 km;
  • SWAN-MED-ITA-RE: forecasts of the state of the sea, up to 72 hours, for the Mediterranean Sea and on five geographical areas close to the Italian coasts.

Furthermore, huge calculation resources are provided (130 dedicated nodes, 4680 cores, on two different clusters, without waiting times) for the management of the other operating chains: LAMI-ENDA: production of frequent analyzes of the current weather situation and very short-term forecasts, also making use of satellite and radar observations using ensemble data assimilation techniques, and LAMI-ENS: operating procedures aimed at probabilistic forecasting using ensemble forecasting techniques (simultaneous execution of different simulations with perturbed configurations).

International collaborations and projects

Cineca's skills in the weather-climate field are internationally recognized, also thanks to the participation in numerous technical and scientific projects relating to the issues of meteorology and climatology. In recent years Cineca has also started several collaborations with international partners.

ICTP and CORDEX project

In recent years Cineca has also started various collaborations in the weather - climate sector with international partners. Recently, the collaboration with ICTP, International Center for Theoretical Physics, a success story for climate research and collaboration in global research. Also in this case, the calculation resources and skills that Cineca made available to climatologists involved in a worldwide collaboration project, called Coordinated Climate Climate Downscaling Experiment (CORDEX) and focused on the downscaling of the regional climate, proved to be decisive. Going from models that consider the planet in blocks of 100 kilometers to regional models, which consider the variation over distances of a few kilometers.
The downsizing of the models is the necessary element to estimate the impact of climate change on dangers and risks in different places. The partnership between ICTP and Cineca was started because the increase in spatial resolution and the inclusion of many more regional details require a lot of processing power and storage space. The elaborations required by the project would not have been possible with the only ICTP calculation capabilities,

Cineca's computing resources were quickly made available to run models and scenarios for countries that don't have compute resources. ICTP has collaborations with working groups in Beijing, Hong Kong and the United States, but in the project it has taken charge of managing domains for developing countries that do not have sufficient processing resources, such as the models for domains of Central and South America and for Africa.
The CORDEX project is important for political decisions relating to floods, droughts, the spread of diseases, public health, agriculture and many others, especially in developing countries that will be most affected by climate change.

In the context of the collaboration with ICTP Cineca has joined the ESGF (Earth System Grid Federation) network, a collaboration that maintains software and IT infrastructure for the scientific data of terrestrial systems. As an ESGF node, Cineca manages and hosts the climatic data produced by the ICTP terrestrial system physics group, providing open access to scientists from all over the world and offers a series of services on data quality, uniform formatting, simplifying use. and comparison with other data. The data of the ICTP CORDEX project are now fully published on the Cineca ESGF node and the global results of CORDEX will be presented in an upcoming special issue of Climate Dynamics.

ECMWF

In 2018, Cineca supported Bologna's candidacy to host the ECMWF (European Center for Medium-Range Weather Forecasts). The data center of the ECMWF weather center, funded with 40 million euros, will definitively make Italy, and Bologna in particular, the European capital of meteorological data. A choice of the European Union that has taken into account the rich fabric of digital infrastructures, research centers and universities in Emilia-Romagna, which allow Italy to discuss internationally on innovation issues.
In addition to supporting Bologna's candidacy, Cineca has also started several collaborations with the prestigious European center.

MISTRAL

Italian initiative that obtained European funding in the Connecting European Facility call in 2018, coordinated by Cineca and in collaboration with the main national stakeholders in the weather sector, such as National Civil Protection, Arpae, Arpa Piemonte and Dedagroup. International partner is ECMWF, which has the task of creating the Italy Flash Flood service, to identify flood episodes.

The goal is to facilitate and encourage the reuse of data sets by the meteorological community. For this purpose, a national platform of open meteorological data has been created to collect and make meteorological data from observational networks, analyzes and historical and real-time forecasts available to professionals and society. The data will be available in terms of grid fields, probabilistic products (such as rain forecasts for the forecast of floods) or punctual time series coming from the modeling chain of Italian operational forecasts and post-processing fields (such as probability of thunderstorm).

Furthermore, the project aims to provide value-added services through the use of supercomputing resources, creating new commercial opportunities. The first version of the platform, where it is possible to access different services created thanks to the collaboration between various partners, is already available.

Five regions, Emilia Romagna, Piedmont, Lazio, Campania and Calabria, have joined the project which foresees the development of a platform that aggregates, harmonizes and post-processes all the weather data observed of the ground stations at national level. The data observed by the ground stations of the regions themselves are made available in open access mode. Last but not least in terms of importance is the partnership between the Mistral project and the Highlander project, which is based on excellent conditions.

Highlander

Starting from the end of 2019, Mistral has been joined and integrated by Highlander, another important project dedicated to meteorological and climatic data which will end in 2022. The Highlander project, compared to Mistral, has a more direct link with the territory. In particular, through the use of High Performance Computing, the goal of the Highlander project is to generate climate forecasts that help reduce the risks associated with climate change for better and more sustainable management of natural resources and the territory. (Highlanderproject.eu)
Cineca coordinates the project, and provides the computing power of HPC systems to generate, post-process, host, distribute and make accessible and exploitable, by as many users as possible, both existing and newly generated data. The Cineca team will also have to deal with the development of a platform for the management of data and metadata, the conversion of raw data into a structured format, the integration of deep learning and machine learning tools for image detection and recognition leveraging HPC resources, creating customized applications for the interactive visualization of scientific data.

 

CYBELE

Focused on the themes of agriculture, breeding and precision aquaculture, such as Highlander, the CYBELE project is attentive to the issue of climate change.

CYBELE aim is to generate innovation and create value in the agri-food sector, through industrial use cases. As agriculture is a high volume activity with low operational efficiency, the project aspires to demonstrate how the convergence between HPC, Big Data, Cloud Computing and IoT can revolutionize agriculture, reduce scarcity and increase food supply, bringing social, economic and environmental benefits. In addition, the project aims to ensure that stakeholders have integrated and non-mediated access to a vast amount of data from a variety of sources, and are capable of generating value by providing secure information and direct non-mediated access to HPC infrastructures that support the collection, processing, combination and visualization of data.

As part of the project, specific and generic domain services will be developed in addition to the virtual research environment, to facilitate the acquisition of knowledge from large amounts of data relating to the agri-food sector, addressing the issue of increased reactivity and enhancement the automation-assisted decision, giving stakeholders the opportunity to use resources more environmentally responsible, improve supply, and implement circular economy solutions in the food chain.
In addition to contributing to the development of the platform, Cineca collaborates with GMV Aerospace and Defense and CACV (Cooperatives Agroalimentàries de la Comunitat Valenciana) to develop a pilot project which aims to create an alarm system for extreme weather events (frost, and hail) by combining weather simulations and Artificial Intelligence technologies. The aim is to provide farmers with a tool capable of anticipating the timing of the forecasts, to give sufficient notice time to implement damage reduction measures (such as anti-hail sheets) in order to save the crops.

Future prospects

Artificial Intelligence models for Nowcasting Radar with applications to real-time alert capabilities

The application of Artificial Intelligence models in the weather forecast will be the next step on which Arpae will work together with Cineca and the Bruno Kessler Foundation. The goal is to improve weather nowcasting skills through the use of Deep Learning tools for early warning (15-60 minutes of lead time) and to define integration models between Deep Learning nowcasting algorithms and numerical models, for forecasting real-time up to 3-6 hours, of extreme weather events for the Emilia Romagna region. The system will be based on Artificial Intelligence methods applied to data from multiple radar sources, also in combination with historical and real-time data relating to lightning strikes, rainfall network, and other forecast weather sources.

From a technological point of view, the models will be developed in a high performance computing environment in collaboration with Cineca, and will be made available through a cloud solution that will allow you to experiment with new early warning functions, such as innovative tools for alerting, analysis services. and forecasting of extreme events, in particular of intense rainfall with impact on human and environmental safety (Civil Protection, Agriculture, Tourism, Mobility).

  • During the presentation event of the supercomputer Leonardo (Italian), Stefano Tibaldi and Tiziana Paccagnella, respectively former director and current director of the ARPAE HydroWeather Service, spoke about the role of supercomputing in the context of meteorological forecasts, and the prospects that a resource like Leonardo (at minute 18 and at minute 31).
 
Minimum glossary
  • physical parameterizations: parameterized algorithms that simulate (define) the physics of the (atmospheric) system;
  •  the chaotic nature of the equations: the Navier-Stokes equations that describe the atmosphere are nonlinear equations, therefore in physics in some cases they are called chaotic. This means that the dynamic system of the atmosphere has an exponential sensitivity to small variations in the initial conditions. Therefore a small uncertainty of the initial conditions means that the system can evolve in a completely unpredictable way over time. Two simulations with small uncertainties about the initial conditions can evolve completely differently and obtain completely different results over time.
  •  ensemble predictions probabilistic predictions (learn more - Italian)
\