A team of High School of Economics and Yandex School of Data Analysis experts is developing a neural network to predict the El Niño climate phenomenon together with the Yandex Cloud cloud platform.
The new algorithm helps to more accurately predict changes in the average temperature of surface ocean waters that can cause natural disasters in certain parts of the world. Now the model predicts El Niño 1.5 years in advance. In the future, scientists plan to increase the forecast period to 2 years.
The neural network models the average temperature in the equatorial region of the Pacific Ocean in perspective. During El Niño, the equatorial part becomes warmer than normal. Along with the decrease in temperature in the ocean, there is also an inverse process – La Niña. Such a replacement cycle occurs every 2-7 years. These fluctuations have a significant impact on weather in many parts of the world and can increase the risk of fires, droughts, floods and crop shortages.
The university research team trained neural networks on thousands of temperature maps with synthetic and real data collected from 1800 to the present. In addition to standard machine learning methods for predicting such events, machine learning experts test the Autoformer architecture in education. In this way, it is possible to process a series of temperature maps in high quality. To preprocess the datasets, the scientists used the Yandex DataSphere ML development service, which has all the necessary tools and dynamically scalable cloud resources for a full machine learning development cycle.
“In projects like the El Niño study, fast and flexible access to services is important for testing different machine learning models. “Every such test with a new architecture is aimed at predicting the phenomenon as early and accurately as possible,” said Anna Lemyakina, Director of National Strategic Projects at Yandex Cloud. It helps,” he said.
Previously at the UN declarationA warming El Niño could replace the La Niña phenomenon in the coming months.