Why Do Weather Forecasts Fail Even with Supercomputers Worth Hundreds of Billions? [Im Ju-hyung's Tech Talk]

Meteorological Administration's Supercomputer Calculating 'Numerical Weather Prediction Models' for Weather Observation
Increasing Variables Affecting Weather Due to Recent Abnormal Climate
Official from Meteorological Administration: "Difficult to Predict Heavy Rainfall with Current Science and Technology"
UK, Japan, and Others Enthusiastically Introducing New Supercomputers

On the 8th, the embankment of the Seomjin River near Geumgokgyo Bridge in Gwiseok-ri, Geumji-myeon, Namwon-si, Jeonbuk collapsed, causing the river water to overflow onto the road. / Photo by Yonhap News

[Asia Economy Reporter Lim Juhyung] Due to abnormal weather patterns causing repeated inaccuracies in forecasts, distrust in the Korea Meteorological Administration's (KMA) predictions is increasing. In particular, after it was revealed that the KMA invested over 10 billion won in purchasing a supercomputer used for weather forecasting, some critics harshly questioned, "How can they produce inaccurate forecasts despite having equipment worth hundreds of billions?"

However, voices have also emerged pointing out the limitations of existing supercomputers, as the number of variables affecting weather predictions has increased due to various factors such as climate change. Consequently, advanced meteorological agencies in countries like the UK and Japan are actively introducing new supercomputers with even more powerful computing capabilities.

Earlier, in May, the KMA forecasted an "unprecedented heatwave" for this summer, predicting both average temperatures and the number of heatwave days to be higher than usual. At that time, the KMA expected summer temperatures to be 0.5 to 1.5 degrees Celsius higher than average, ranging from 24.1 to 25.1 degrees Celsius, and the number of heatwave days to be more than double the usual, at 20 to 25 days.

However, last month, the national average temperature was 22.5 degrees Celsius, about 2 degrees lower than usual. The number of heatwave days was 3.9, and tropical night days were 2.3, both 2 to 3 days fewer than average.

Rainfall predictions also missed the mark significantly. The KMA initially predicted rainfall to be similar to or less than average, but the monsoon lasted for 49 consecutive days until the 11th, with rainfall in July and August alone reaching 663.3 mm. This is close to the record of 699.1 mm set in 2006, the year with the highest rainfall during the monsoon period.

Given this situation, distrust in the KMA's forecasting ability is growing. Some critics have pointed out that despite spending 17.2 billion won on the state-of-the-art "Supercomputer No. 4," inaccurate forecasts from the KMA continue unabated.

Netizens expressed their dissatisfaction with the KMA's forecasts on social networking services (SNS), writing comments such as, "They spent an enormous amount of tax money on a supercomputer, yet can't even predict the weather a few hours ahead," and "It would be more accurate to just watch CCTV to see if it's raining."

◆The KMA Supercomputer is a Giant Calculator

The Korea Meteorological Administration's 'Supercomputer No. 4', introduced in 2015. It can perform 5,800 trillion calculations per second. / Photo by Korea Meteorological Administration

So, what role does a supercomputer play in weather forecasting? In fact, a supercomputer alone cannot predict the weather. It is more like a "giant calculator" used to compute the equations necessary for weather prediction.

The KMA's weather forecasting begins with collecting as much atmospheric data as possible. The KMA gathers data such as temperature, pressure, humidity, and wind speed from over 110 meteorological observation stations worldwide and weather satellites.

The collected data is input into a "numerical weather prediction model," which the supercomputer then calculates to produce forecast values. Human scientists continuously monitor and compare these computer-generated values with actual weather, refining the forecasts through this process.

The Korea Meteorological Administration uses a combination of the UK-developed numerical weather prediction model "UM" and its own model optimized for Korean climate called "KIM."

Meanwhile, the "Supercomputer No. 4," which calculates these models, is the latest system introduced in 2015. According to the KMA, it uses a hybrid of the XC40 large-scale parallel processor designed by the US supercomputer manufacturer CRAY and Intel central processing units (CPUs), delivering a maximum performance of 5800 TFlop/s (teraflops, or one trillion floating-point operations per second). In other words, Supercomputer No. 4 can perform 5.8 quadrillion calculations per second.

◆The Problem is Climate Change Uncertainty... Increased Variables Lower Accuracy

On the afternoon of the 10th, heavy rain in Yangju-si, Gyeonggi Province caused Yangju Station and nearby roads to be flooded. The photo shows the exterior of the station submerged in water. / Photo by Yonhap News

The problem lies in the fact that climate change has introduced phenomena such as super monsoons and squall line heavy rains, which were not previously observed in Korea's environment, increasing the variables in weather forecasting. Numerical weather prediction models produce results based on vast amounts of data, but even slight changes in input data can lead to completely different outcomes.

For example, if a short, intense rain shower occurs in a specific area and then stops, a large amount of water vapor is generated, increasing the probability of precipitation. This creates new variables in weather forecasting. Because weather conditions are unstable like this, even supercomputers capable of teraflop-level calculations cannot always predict the weather accurately.

Yoon Ki-han, a spokesperson for the KMA, appeared on a radio broadcast on the 11th and explained, "In this monsoon season, there were many localized heavy rains. The fact that these heavy rains move across different regions changes local conditions, making it impossible to capture all data scientifically."

He added, "Reducing the gap between areas with heavy rain and those without is a current technical challenge for the KMA. We continue to treat this as a task and use it as a measure of progress."

◆Advanced Countries Actively Introduce New Supercomputers in the Era of Climate Change

It is not only the Korean Meteorological Administration that struggles with the uncertainty caused by climate change in producing accurate weather forecasts. Other countries are also striving to detect natural disasters early through precise weather forecasting, such as the UK, which issued flood warnings in 634 locations due to Typhoon Dennis in February, and the US, which has suffered damage from successive super hurricanes.

In the UK, the government allocated 864 million pounds (approximately 1.3 trillion won) to the UK Met Office on February 17 to introduce a new supercomputer.

Scheduled to be installed in 2022, this computer will be equipped with the ARM "A64fx" processor, which is far more advanced than the existing XC40 processor. This processor is based on the ARM architecture, a British semiconductor company famous for smartphone CPU designs, and can perform up to 42,000 teraflops, or 42 quadrillion calculations per second.

According to the supercomputer performance ranking site "TOP500," the UK Met Office's new supercomputer ranks around 6th in the world in terms of performance.

The Japan Meteorological Agency also plans to build a system using a new supercomputer that can simultaneously analyze ground humidity and sea vapor inflow, in addition to its existing Japanese numerical weather prediction model "AMeDAS."

The Korea Meteorological Administration is also preparing to introduce a next-generation supercomputer with enhanced performance compared to the existing Supercomputer No. 4. By the end of 2021, it plans to invest 62.8 billion won to build the new "Supercomputer No. 5."