by Lori Dajose, California Institute of Technology
In a simplified depiction of the EnOC algorithm, two dynamical ensemble members are illustrated for clarity. In this scenario, the second (purple) ensemble member is assigned a higher weight due to its proximity to the MISO forecast within the subspace. It is important to note that in the actual implementation, the dynamics in the MISO subspace are reduced to the first two principal components of the MISO mode. Credit: Proceedings of the National Academy of Sciences (2024). DOI: 10.1073/pnas.2312573121
Annually, the South Asian monsoon season brings substantial rainfall to over a billion individuals in the Indian subcontinent from June to September. The precipitation occurs in a pattern of fluctuations: Some weeks experience 1 to 4 inches of rainfall, while others remain predominantly dry. Accurately predicting the timing of these dry and wet periods is crucial for agricultural and urban planning. This foresight allows farmers to determine the optimal harvest times and aids city officials in preparing for potential flooding. Despite the general accuracy of short-term weather forecasts within a day or two, making precise predictions for weather conditions a week or a month in advance remains highly challenging.
A novel machine-learning-based forecast has emerged as a more precise predictor of South Asian monsoon rainfall 10 to 30 days ahead. This advancement marks a significant improvement over current cutting-edge forecasts that rely on numerical modeling rather than artificial intelligence for forecasting. Understanding the behavior of monsoons holds significance as this type of rainfall serves as a prominent atmospheric element in the global climate.
The study, spearheaded by Eviatar Bach, the Foster and Coco Stanback Postdoctoral Scholar Research Associate in Environmental Science and Engineering, was conducted in collaboration with the laboratories of Tapio Schneider, the Theodore Y. Wu Professor of Environmental Science and Engineering and JPL senior research scientist, and Andrew Stuart, the Bren Professor of Computing and Mathematical Sciences.
The details of this innovative approach are outlined in a paper published in the Proceedings of the National Academy of Sciences.
Bach emphasizes, “There is considerable apprehension regarding the impact of climate change on monsoons and other weather phenomena such as hurricanes, heatwaves, and more. Enhancing predictions on shorter timescales is a pivotal aspect of addressing climate change as it is imperative to enhance preparedness for such events.”
A representation illustrating the variation in monsoon rainfall, known as the “monsoon intraseasonal oscillation,” across the Indian subcontinent throughout a single season.
Weather prediction poses challenges due to various instabilities in the atmosphere—for instance, continuous heating of the atmosphere from the earth’s surface leads to the presence of cold, denser air above warmer, less dense air. Additionally, instabilities arise from non-uniform heating and the Earth’s rotation. These factors create a chaotic environment where errors and uncertainties in modeling atmospheric behavior rapidly escalate, rendering long-term predictions nearly impossible.
State-of-the-art models currently rely on numerical modeling, which entails computer simulations of atmospheric conditions based on the physics equations governing fluid motion. Owing to chaos, the maximum predictable timeframe for large-scale weather typically spans around 10 days. While predicting the long-term average atmospheric behavior (climate) is feasible, forecasting weather within the timeframe of two weeks to several months has posed a challenge using numerical models.
In the context of South Asian monsoons, rainfall typically occurs in cycles of intense downpours followed by dry intervals. These recurring patterns are referred to as monsoon intraseasonal oscillations (MISOs). In the recent study, Bach and collaborators integrated a machine-learning element into the existing state-of-the-art numerical models. This integration enabled the researchers to collect data on MISOs and enhance predictions of rainfall on the elusive two-to-four-week timescale. The resultant model exhibited a remarkable enhancement in correlation between predictions and observations, with improvements of up to 70%.
Bach elaborates, “In recent years, there has been a growing interest in leveraging machine learning for weather forecasting. Our research demonstrates that a fusion of machine learning and traditional numerical modeling can yield accurate outcomes.”
The paper, titled “Improved subseasonal prediction of South Asian monsoon rainfall using data-driven forecasts of oscillatory modes,” lists V. Krishnamurthy and Jagadish Shukla from George Mason University; Safa Mote from Portland State University; A. Surjalal Sharma and Eugenia Kalnay from the University of Maryland; and Michael Ghil from École Normale Supérieure in Paris, UCLA, and Imperial College London as co-authors.
More information:
Eviatar Bach et al, Improved subseasonal prediction of South Asian monsoon rainfall using data-driven forecasts of oscillatory modes, Proceedings of the National Academy of Sciences (2024). DOI: 10.1073/pnas.2312573121
Provided by
California Institute of Technology
Citation:
AI improves monsoon rainfall predictions (2024, April 1)
retrieved 2 April 2024 from https://phys.org/news/2024-04-ai-monsoon-rainfall.html
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.
