AI and Traditional Forecasts Deliver Different Economic Benefits, Study Finds

A new World Bank study finds that the real value of weather forecasts depends not just on prediction accuracy but on how effectively they reduce economic losses, improve decision-making, and maintain public trust during extreme weather events. The research shows that repeated forecast failures can sharply increase disaster costs, highlighting the need for governments, development partners, and businesses to evaluate forecasts based on economic impact rather than technical performance alone.

A new study by researchers from Uppsala University, Stockholm University, the Swedish Centre for Impacts of Climate Extremes (Climes), the Centre of Natural Hazards and Disaster Science (CNDS), the World Bank Group, and the Global Facility for Disaster Reduction and Recovery (GFDRR) challenges a common assumption in weather forecasting: that more accurate forecasts automatically create greater economic benefits.

The study argues that the real value of weather forecasts should be measured by how much they help reduce losses from extreme weather events, rather than by accuracy scores alone. This finding comes as governments and businesses increasingly rely on artificial intelligence (AI) to improve weather prediction and disaster preparedness.

Why Traditional Forecast Measures Fall Short

Weather agencies typically assess forecasts using technical indicators such as accuracy, error rates, and prediction skill. While these measures show how well a model predicts weather conditions, they do not reveal whether forecasts help people make better decisions or avoid economic losses.

The researchers note that extreme weather events are becoming more frequent and costly. When multiple disasters occur in a short period, damage can increase rapidly. Infrastructure may already be weakened, emergency services can become overwhelmed, and recovery costs can rise sharply.

The study argues that traditional forecast evaluations fail to account for these "compounding extremes," making it difficult for policymakers to understand the true economic value of forecasting systems.

Repeated Forecast Errors Can Be Costly

A major finding of the study is that repeated forecast mistakes can significantly increase losses.

When dangerous events are missed several times in a row, communities may be left unprepared, resulting in larger economic and social damages. On the other hand, repeated false alarms can reduce public trust in warning systems, making people less likely to respond to future alerts.

To capture these effects, the researchers developed a framework that increases the economic penalty for consecutive forecast errors. They tested scenarios in which costs doubled or even quadrupled with repeated mistakes, reflecting real-world situations where impacts can escalate rapidly.

The results showed that forecast value often falls sharply when these compounding costs are taken into account.

AI Versus Traditional Forecasting Systems

The study compared two forecasting systems developed by the European Centre for Medium-Range Weather Forecasts (ECMWF): the traditional physics-based Integrated Forecasting System High Resolution model (IFS HRES) and the AI-powered Artificial Intelligence Forecasting System (AIFS).

Researchers evaluated five-day forecasts of extreme temperature and wind events in six South Asian cities—Bangkok, Delhi, Dhaka, Hanoi, Islamabad, Mumbai and six Southern European cities, including Athens, Madrid, Milan, Rome, Seville, and Valencia.

For temperature forecasts, AIFS often delivered higher economic value in cities such as Bangkok and Mumbai. However, the traditional IFS HRES model performed better in locations such as Rome and Athens. The findings show that the best forecasting system depends on local conditions, prevention costs, and vulnerability to weather-related losses.

One important result was that when preventive actions are relatively cheap compared with potential disaster losses, forecasts that successfully identify more extreme events generate greater value. When preventive measures are expensive, reducing false alarms becomes more important.

What This Means for Governments and Development Partners

For governments, the study highlights the need to move beyond forecast accuracy when evaluating investments in meteorological services and early-warning systems.

The findings suggest that national weather agencies should assess how forecasts affect emergency planning, infrastructure protection, and disaster response costs. Economic-value assessments could help governments prioritize investments in forecasting technologies that provide the greatest social and financial returns.

For development partners such as the World Bank, regional development banks, donor agencies, and climate funds, the framework offers a new way to evaluate the effectiveness of investments in hydrometeorological infrastructure. It can help identify where improved forecasts are most likely to reduce disaster losses and strengthen climate resilience.

Risks, Opportunities, and the Road Ahead

The study also has important implications for the private sector. Industries such as agriculture, energy, insurance, logistics, aviation, and shipping depend heavily on weather information. Businesses may benefit from evaluating forecasts based on their impact on operations, supply chains, and risk management rather than relying solely on technical accuracy measures.

The researchers found that wind forecasts generally provided less economic value than temperature forecasts. In some cases, especially when the costs of repeated forecast failures were assumed to rise rapidly, wind forecasts produced little or no positive economic value. This highlights the risks of relying solely on forecast accuracy when making investment decisions.

Looking ahead, the authors recommend making economic-value assessments a standard part of forecast evaluation. They also suggest that future AI forecasting systems should be designed not only to improve prediction accuracy but also to maximize economic and social benefits.

As climate change increases the frequency of extreme weather events, the study concludes that the most valuable forecasts will not necessarily be the most accurate ones. Instead, they will be the forecasts that help governments, communities, development partners, and businesses make better decisions and avoid the greatest losses.