The Malaysia Meteorological Department (MMD) works non-stop to save human lives from weather disasters. Located just above the equator, the country’s two parts, Malay Peninsula and the island of Borneo, are surrounded by ocean waters. Weather events in Malaysia can spin up quickly and wreak widespread devastation. While weather data is plentiful and free, the country’s existing Numerical Weather Forecasting Computer system couldn’t produce weather forecasts with enough detail to give adequate warning to any rising threat. A new system was needed that could assimilate weather data efficiently and produce detailed weather forecasts faster in order to provide early warnings to the public in case of inclement weather.
Malaysia’s old computer system generated weather forecasts at an approximate resolution of tens of kilometers apart. That meant there was ample room for error. An emerging weather threat would have to grow large enough for the system to detect it at that range, which meant sizable storms could do considerable damage long before detection.
In order to get a more finite and accurate read on real-time weather conditions in Malaysia’s rainforests, beaches, high-rise commercial and rural areas, as well as aircraft routes, the country needed to upgrade their system to run highly complex numerical weather prediction (NWP) models that require a massive amount of compute power. This is a huge challenge because to double the resolution of the same forecasting area (from 10 to 5 kilometers), an NWP model would require approximately 16 times the compute power to deliver the forecast within the same time period.
Across the South China Sea, about a three-hour flight by jet from Malaysia, lies more than 7600 islands that comprise the Philippines. While weather is an ongoing concern for the Philippine Atmospheric, Geophysical and Astronomical Services Administration (PAGASA) too, that country’s agency also set a dual goal of economic development through predictions of climate change. Being able to precisely pinpoint optimal locations for everything from crop lands, to massive windmill farms, to offshore oil rigs, and other expensive economic engines, is a foundational capability for countries trying to build a 21st century economy.
To successfully simulate regional climate changes, the data scientists at PAGASA adopted a regional Numerical Climate model, producing an ensemble of simultaneous climate forecasts. Each forecast member is initialized with slightly different meteorological data in order to statistically represent the inherent uncertainty of errors in weather data collection and assimilation. The outputs from all ensemble runs are then averaged to produce a probabilistic local climate outlook over a three-month period, rather than just deliver weather expectations over a week.
For example, a perfect location for most crops is somewhere flooding isn’t a routine issue. A seven-day weather prediction cannot reveal the expected number of flooding incidents in a specific geographical area over months or years, but a climate simulation can. Given changes wrought by influences ranging from planet-wide climate change to regional climate cycles such as El Nino, the PAGASA model’s demands on compute infrastructure are extreme and complicated.
Another example is windmill farms which are ideally located where the wind blows more often than not. Being able to predict wind speeds and wind frequency over longer periods of time enable planners to pinpoint such locations. Climate simulations also play a large role in weather predictions by finding and accurately predicting air moisture and warm ocean water pools which are a strong indicator of cyclone or typhoon development.
Both Malaysia and the Philippines were looking for complete systems to tackle these huge computing workloads. Neither country was interested in buying hardware and software piecemeal, and then struggling to make it all work together.
Lenovo answered the challenge by creating two complete supercomputer system solutions that were ready to run these workloads immediately.
For Malaysia, Lenovo, along with its partner, the Numerical Weather Prediction consultants (NWPC LLC) of the US, installed a NextScale water-cooled supercomputer system at the MMD datacenter and a weather forecasting system based on the Weather Research and Forecasting (WRF) weather model of the National Center for Atmospheric Research (NCAR). It operates today in Malaysia at a resolution down to 1 km for all of Malaysia (and 333 meters for Kuala Lumpur), in order to issue overlapping 7-day forecasts, and not have any weather events slip through undetected. This system is first of its kind, and the most powerful supercomputer in Malaysia.
“To run the latest WRF software, we needed the HPC system to benchmark at least 100 TFLOPs performance. We looked at lots of proposals from different vendors, and the Lenovo Scalable Infrastructure (LeSI) came out on top in terms of performance and cost-efficiency – it met all our compute and storage requirements at the best price. This system allowed us to at least double the accuracy of the weather forecasts in Malaysia.” Dr. Wan Azli B Wan Hassan, Deputy Director General (Strategic & Technical) at MMD.
For the Philippines, Lenovo built a powerful system uniquely suited to the enormous task of producing climate simulations as well as weather predictions. This supercomputer system is based on the Lenovo ThinkSystem SD530 and it uses Intel’s XEON Scalable Systems processors. Lenovo partnered with NCAR and NWPC to deliver a climate solution that is also based on the WRF model at climate resolutions down to 3 kilometers.
“As the National Meteorological and Hydrological Center in the country, PAGASA acquired a first-in-country High Performance Computing System for Operational Climate Forecasting from Lenovo. This system will be able to produce both, short and long-term local climate outlooks to aid various decision-making agencies, while planning activities for socio-economic sectors in the Philippines.” Ana Liza S. Solis,. Chief, Climate Monitoring and Prediction Section, PAGASA.