Weather significantly impacts society for better and for worse, and improving our ability to forecast and predict weather events is crucial for myriad reasons. An increased notice period for a hurricane could improve safety measures for people at risk, while improved solar predictions can help optimize renewable power production.
Applying artificial intelligence (AI) techniques in conjunction with our physical understanding of the environment can substantially improve prediction of extreme weather events, like hurricanes, and unlock important insights from the climate data that is collected.
“Artificial intelligence and related data science methods have been developed to work with big data across a variety of disciplines,” explained Dr. Ernesto Damiani, Senior Director of the Artificial Intelligence and Intelligent Systems Institute at Khalifa University.
AI techniques can handle large numbers of predictor variables, integrate physical understanding into models efficiently, and discover new knowledge from data, contributing to improved weather predictions and a better understanding of many weather-related phenomena.
“Weather forecasting is the task of predicting the state of the atmosphere at a future time and a specified location,” explained Dr. Damiani. “Traditionally, this is done through physical simulations where the atmosphere is modelled as a fluid. The present state of the atmosphere is sampled and the future state is computed by numerically solving the equations of fluid dynamics at different resolutions: micro-, meso-, and macro-scale.”
In addition to using such physics-based model, in recent years forecasters and researchers have begun to adopt AI techniques much more widely, as they demonstrate their power in a wide variety of applications, including post-model bias correction, processing large datasets, reducing cognitive overload, and unlocking new insights in large datasets.
“The system of ordinary differential equations that govern physics-based models can be unstable under perturbations, and uncertainty in the initial measurements of the atmospheric conditions limit accuracy,” explained Dr. Damiani. “Machine learning is relatively robust to perturbations and does not require a complete understanding of the underlying physical processes that govern the atmosphere. Therefore, machine learning may represent a viable alternative to physical models in weather forecasting.”
While many research groups worldwide have focused on deep learning models, researchers at Khalifa University have been focusing on a multi-view approach, where related groups of sensor data sources provide different views on the phenomenon, to be later compiled into a final classification or prediction stage.
Accurate forecasting is particularly crucial for the UAE’s cloud seeding operations.
Initiated in the late 1990s in the UAE, cloud seeding has become a regular occurrence in recent years, with an average of between 160 and 200 flights per year. The UAE has an arid climate with less than 100mm per year of rainfall, a high evaporation rate of surface water, and a low groundwater recharge rate. Although rainfall in the UAE has been fluctuating over the last few decades in the winter season, most occurs between December and March annually.
The UAE has embraced rain enhancement as an important tool in its arsenal to support the country’s water security efforts. Among the country’s key goals are advancing the science, technology and implementation of rain enhancement, encouraging additional investments in research funding, increasing rainfall, and ensuring water security. Forecasters and scientists have estimated that cloud seeding operations can enhance rainfall by as much as 35 percent in a clear atmosphere, and by up to 15 percent in a turbid atmosphere.
The UAE’s National Center for Meteorology commences cloud seeding drills as soon as meteorologists forecast cloudy weather. To optimize the deployment of the limited budget of the seeding material, these forecasts need to be as accurate as possible, which is where AI can step in.
Many techniques can be applied to improve their forecasting ability, including artificial neural networks (ANNs) – interconnected networks of weighted nonlinear functions that can be connected and trained in multiple layers. ANNs provide the foundation for deep learning methods and have been used in a wide variety of meteorology applications since the late 1980s, including cloud classification and precipitation classification.
Applying modern AI techniques to weather forecasting is improving our ability to sift through the deluge of big data to extract insights and accurate, timely guidance for human weather forecasters and decision-makers, and is playing an indispensable role in the UAE’s efforts to achieve greater water security.