Deep Learning Weather Prediction Model (DLWP) AI Makes Predictions in Seconds.
New AI-based weather-forecasting research is accelerating global weather predictions. The research, which was just published in the Journal of Advances in Modeling Earth Systems, could help predict extreme weather 2–6 weeks in advance. Communities and vital sectors such as public health, water management, energy, and agriculture will have more time to prepare for and prevent potential disasters if accurate extreme weather predictions are made with a longer lead time.
Climate change is increasing the intensity and frequency of extreme weather events, with storm, heatwave, flood, and drought records set to be broken around the world in 2021. According to a recent NOAA analysis, the United States was hit by 20 climate-related weather catastrophes last year, each causing over $1 billion in damage.
Weather forecasting, both short-term and seasonal, can help reduce the socioeconomic and human consequences of extreme weather. Meteorologists in the Philippines alerted local and national leaders in 2019 that a severe monsoon was approaching in roughly three weeks. The prognosis allowed populations to weatherize structures and evacuate before the Category 4 Typhoon struck, saving lives and lowering overall damage.
Supercomputers are used to process massive amounts of global data such as temperature, pressure, humidity, and wind speed in today’s weather forecasting. These systems require a lot of computing power and take a long time to process.
Furthermore, the capacity to accurately estimate forecasts further out, from several weeks to months, drops dramatically, according to the authors.
In order to improve current weather forecasting, researchers developed the Deep Learning Weather Prediction model, which is computationally efficient and capable of reliably predicting forthcoming weather (DLWP). The DLWP is built on an AI system that learns and recognizes patterns in historical weather data based on worldwide grids. It was first introduced in a paper released in 2020.
The current work refines the DLWP by training a deep convolutional neural network on two additional data points—the temperature at the atmospheric boundary layer and total column water vapor. They also improved the grid resolution at the equator to approximately 1.4°.
Running on a single cuDNN-accelerated TensorFlow deep learning framework on an NVIDIA V100 GPU, the model runs 320 ensemble 6-week forecasts in just 3 minutes. The algorithm can process a 1-week forecast in 1/10th of a second.
The DLWP is capable of producing realistic weather forecasts, such as Hurricane Irma, a Category 4 hurricane that struck Florida and the Caribbean in 2017. While the fast DLWP model equals the performance of current state-of-the-art weather forecasters four to six weeks out, it has limits in predicting precipitation and is less accurate in shorter lead times of two to three weeks.
According to the research, the DLWP could be useful for enhancing spring and summer forecasts in the tropics, a region where existing weather models are challenged.