How AI is changing the world of weather forecastingーNHK WORLD-JAPAN NEWS

Key Concepts

• Artificial Intelligence (AI) in weather forecasting
• Typhoon track prediction
• Machine learning
• Traditional forecasting models
• Accuracy and error rates
• Accountability in AI forecasting
• Hybrid forecasting approaches
• Climate change and typhoon intensity

AI Revolutionizing Typhoon Forecasting

Governments are increasingly focused on improving natural disaster forecasting, with climate change being cited as a factor in the rising frequency and intensity of typhoons. Artificial intelligence (AI) is emerging as a transformative technology in this field, particularly in predicting typhoon tracks.

Global Adoption and Advancements

A recent gathering of approximately 200 typhoon researchers in Yokohama highlighted the significant advancements in AI forecasting. Experts noted a "dramatic improvement" in AI's capabilities in recent years.

Case Study: Typhoon Halong

• Actual Path: Represented by the black line.
• Google DeepMind AI Prediction: Shown in blue.
• Japan Meteorological Agency (JMA) Forecast: Shown in red.

Both AI and JMA predictions were made shortly after Typhoon Halong formed. The AI forecast was notably closer to the storm's actual path. Furthermore, the AI provided a longer-range prediction of 15 days, compared to the JMA's standard 5-day forecast, though the AI's extended forecast was for reference purposes.

Methodological Differences: AI vs. Traditional Models

The divergence in forecasting accuracy stems from the underlying methodologies:

• JMA's Model: Relies on physical laws combined with real-time data such as temperature, pressure, and wind.
• AI Models: Employ machine learning, analyzing massive datasets of past typhoon movements and intensities to predict future outcomes. This data-driven approach allows AI to learn patterns and correlations that may not be immediately apparent in traditional physical models.

Global Trends and Implementation

The shift towards AI in weather forecasting is a global trend:

• South Korea: Plans to introduce AI-based forecasting soon, citing its speed, reduced error rate, and lack of need for human intervention ("bls" likely refers to biases or human limitations). Officials describe this as "the direction of science and the global trend."
• Europe: The European Centre for Medium-Range Weather Forecasts (ECMWF) launched its own AI model in February, providing data to countries worldwide. Forecasts are accessible via their website.

Evidence of AI's Superior Accuracy

A graph illustrating forecast accuracy across numerous typhoons demonstrates that AI predictions consistently exhibit fewer errors compared to traditional methods. Senior researchers have expressed astonishment at these results, acknowledging the necessity of integrating AI into forecasting strategies. This advancement could potentially fuel intense international competition in typhoon track forecasting.

Drawbacks and Challenges of AI Forecasting

Despite its advantages, AI forecasting is not without its limitations:

• Predicting Intensity: AI models still struggle to accurately predict the intensity of typhoons.
• Accountability: This is identified as the "biggest issue."

Case Study: Typhoon Faxai

• Actual Path: Represented by the black line.
• JMA Forecast: Shown in red.
• AI Forecast: Shown in blue.

In the case of Typhoon Faxai, the AI predicted a direct hit on Chiba (Ocar likely a typo for Chiba), but the storm ultimately passed further south. The JMA forecast proved more accurate in this instance.

The JMA's ability to adjust its forecast and explain changes is attributed to its reliance on actual, real-time data. AI models, being primarily statistical, lack this transparency. "You can't really know why our prediction went wrong," highlighting a significant challenge for accountability.

Japan's Strategic Approach

Japan is actively exploring the integration of AI into its forecasting systems:

• Government Initiatives: The government is observing overseas AI usage and plans to develop its own AI model utilizing historical data.
• JMA Recruitment: The JMA aims to hire approximately 30 AI developers in the next fiscal year.

The Ideal Future: Hybrid Forecasting

Experts advocate for a combined approach, integrating traditional methods with AI models.

Quote from Fellan Allet, Research Scientist at Google DeepMind:

"I think that models are certainly going to get larger and ingest more data and probably predict more things as well. I expect that in a couple years time our set of predictions is going to expand and the predictions we're making now are going to become the norm. They're going to be a routine part of of the set of models that forecasters will have at their fingertips."

This suggests a future where AI's predictive capabilities will expand, and current AI predictions will become standard tools for forecasters.

Conclusion: Enhancing Safety in a Warming World

As global warming intensifies, typhoons are expected to become stronger. The ability to accurately predict both the path and intensity of these storms is crucial for ensuring public safety. The ongoing development and integration of AI, particularly in conjunction with traditional forecasting methods, represent a critical step towards achieving this goal.