Rong Jin
Alibaba nowcasting predicts short-term weather
DUBAI, October 24, 2021
Alibaba Damo Academy, the global research initiative by Alibaba Group, has unveiled a cloud-based AI-powered nowcasting platform capable of predicting short-term weather conditions up to six hours in advance.
The Short-term AI Weather Forecasting Platform, now available to Alibaba Cloud’s clients in China, produces high-resolution imagery with one-kilometre grid spacing with updates available every 10 minutes. Capable of tracking rainfall, windspeed, and cloud formation as well as severe weather conditions such as thunder and hailstorms, the platform promises to deliver tangible value to weather-dependent sectors, including agriculture, logistics, transportation and renewable energy.
For farmers, a timely and accurate weather forecast can minimize damage to crops and livestock; couriers can schedule their routes efficiently on rainy days; and photovoltaic power stations can use cloud formation predictions to better prepare their electricity trading plans.
“Nowcasting has proven a critical technology to help various sectors make informed weather-related decisions. Global technology players are working hard to develop technology-based services that utilize reliable climate data from their respective countries,” said Rong Jin, Head of the Machine Intelligence Lab at Alibaba Damo Academy.
“Using our cutting-edge algorithms and cloud technologies, we have significantly advanced our nowcasting capabilities in China. By doing so, we aim to help businesses meet their climate-related challenges and mitigate the risks of unpredictable weather,” Jin added.
The AI-based forecast platform, co-developed by Alibaba Damo Academy and the National Meteorological Center in China, incorporates a convolutional neural network (CNN) model to effectively extract features from radar reflectivity and meteorological satellite images.
A trained machine-learning model is capable of performing highly accurate and close-to-real-time local weather forecasting in minutes, while Generative Adversarial Network (GAN) works to generate forecast images with exceptional clarity and detail. This AI-based prediction model outperforms the traditional physics-based model, for example the Global/Regional Assimilation and Prediction System (GRAPES) which requires hours to generate forecasting data, by increasing the speed and accuracy of reporting. – TradeArabia News Service