Human and environmental protection requires accurate predictions of extreme weather events, such as winter storms. Researchers at the Karlsruhe Institute of Technology (KIT) have compared the methods of computation and machine learning for wind wind forecasts with the aim of making more accurate and accurate forecasts. trust. They found that considering other meteorological information and characteristics, such as temperature, significantly improved the quality of the forecast, while using new AI techniques related to neural networks.
Strong winds, such as hurricanes with speeds of more than 65 kilometers per hour, can seriously injure people, animals, and businesses. In order to deliver effective advice, accurate and reliable predictions are needed. “It’s hard to compare wind turbines, because they’re driven by small processes and are limited to land,” said Benedikt Schulz, a medical researcher at KIT’s Institute of Stochastics. “Their predictions with statistical forecasting features used by business services are limited and may not be clear.”
To better compare those uncertainties of predictions, meteorologists make ensemble predictions. Depending on the current weather conditions, they make the sample numbers look like slightly different conditions. In this way, it covers the various developmental features of the future. “Despite the continuous improvement, however, there are various systemic flaws in those ensemble weather forecasts, as different physical characteristics cannot be considered,” Schulz said. “With the help of smart intelligence, we want to correct these systematic errors, improve predictions, and be more reliable in predicting catastrophic events.”
Detailed information and new meteorological variables improve wind forecast
Me Dr. Sebastian Lerch, Schulz for the first time compared the different numbers and AI types for the final correction of the wind wind ensemble predictions. “We’ve analyzed new and innovative techniques for correcting the number after a series of statistical predictions and accurately compared their predictions,” Lerch said. He heads the young research group “AI Methods for Probabilistic Weather Forecasts” supported by the Vector Foundation at KIT’s Institute for Economic Policy Research.
All the latter methods are available to make reliable wind speed predictions. “However, AI methods are better than standard statistical methods and produce the best results, because they allow for better consideration of new sources of information, such as geographical conditions. or other meteorological variables, such as temperature and solar radiation, ”Lerch concludes. “AI techniques reduce the prediction errors of real -time models by about 36 percent on average,” Schulz added. The researchers looked at predictions made with the German Weather Service (DWD) model at 175 monitoring centers in Germany and found that AI methods produced better predictions than 92 percent. of servers. Neural networks can learn complex and non -complex relationships from large data sets available. This plays an important role in correcting systemic errors of ensemble predictions. “Looking at the information about the conditions can draw conclusions about meteorological processes,” Schulz said.
Through their work, the researchers contribute to the development of weather forecasting techniques at the center of statistics and AI. “Weather services may use these methods to improve their forecasts,” Lerch said. “Because of this, we are connected to the German people and other international business services.”
AI prediction: The machine is learning to predict time
Benedikt Schulz et al, Machine Learning Practices for Managing Ensemble Forecasts of Wind Gusts: A Systematic Comparison, The time of the month (2021). DOI: 10.1175 / MWR-D-21-0150.1
Presented by Karlsruhe Institute of Technology
Directions: Researchers use data and AI techniques to correct systemic errors of weather patterns (2022, March 28) Retrieved March 28, 2022 from https://phys.org/news/ 2022-03-statistics-ai-methods-systematic-errors.html
This document is copyrighted. Except for appropriate action for the purpose of personal inquiry or research, no piece may be reproduced without permission. Information is provided for informational purposes only.