How Large Tech Companies Respond to Climate Change

2023-07-13

Google, Microsoft, Huawei, and other major tech companies are investing in artificial intelligence models for weather forecasting. But are they as reliable as traditional methods? More and more tech companies are leaving their mark in the field of climate by harnessing the transformative power of advanced artificial intelligence/machine learning models. Recently, Huawei released its latest AI model, Pan Gu Weather, which has higher forecast accuracy compared to traditional numerical weather prediction methods. This model utilizes deep learning technology and 43 years of historical data, predicting speeds 10,000 times faster than traditional methods. Pan Gu Weather successfully forecasted the path of Typhoon "Mawar" near the eastern islands of Taiwan with an accuracy of up to 5 days in advance. Google Research has developed a deep learning model called MetNet-2 to address weather and climate-related issues using artificial intelligence. This model has high accuracy in precipitation prediction, with spatial resolution up to 1 kilometer, temporal resolution up to 2 minutes, and duration up to 12 hours. Microsoft and DeepMind have also established their own artificial intelligence models, ClimaX and Graphcast, respectively. Meanwhile, NVIDIA is researching the Earth 2 model in collaboration with climate researchers and policymakers. By using Modulus and FourCastNet, machine learning models that mimic global weather patterns dynamically, NVIDIA has made significant improvements in weather trajectory prediction with unprecedented speed and accuracy. They are able to generate weather trajectories for 1,000 model members for a duration of 21 days in a fraction of the time previously required for a single trajectory. IBM has been in this field for some time now. In 2016, IBM acquired The Weather Company, a subsidiary of the Weather Channel. This model combines information from over 100 weather forecast models worldwide using IBM's Watson framework and artificial intelligence. Zeus AI, a startup founded by former NASA scientists, utilizes a large amount of data from government satellites, including information on atmospheric winds, water vapor, temperature changes, and cloud cover, which affect weather and global weather patterns. Other companies addressing weather forecasting issues include Tomorrow.io, Atmo.io, Jua.ai, and Zeus.ai. Limitations: Using machine learning models for weather forecasting has its limitations. One of the issues is the limitation of training data. Rare and extreme weather conditions pose challenges for training and testing. Additionally, data availability is another problem. In NWP models that use satellite data, missing values are interpolated. However, using this interpolated data for artificial intelligence models may result in concept drift and inherent biases. According to a paper on machine learning for weather and climate modeling, neural networks may need explicit guidance on the relationships between certain variables. Short to medium-term datasets are insufficient for models to understand long-term changes, such as El Niño phenomena or any form of climate change. The New Era of AI Forecasting: Weather forecasters primarily rely on numerical weather prediction (NWP) models. These models use data from weather stations, weather balloons, and satellites to understand the current state of the atmosphere. By solving equations related to air motion, these models can accurately predict most weather patterns. However, smaller weather events, such as localized thunderstorms or predicting which side of a town will experience heavy rain during a thunderstorm, pose challenges. This method also requires expensive computational power, which is where artificial intelligence models excel. Predictive methods using machine learning models analyze a large number of previous weather maps to understand typical weather patterns, rather than obtaining models through solving complex physical equations like traditional NWP models. By training on historical data, artificial intelligence models utilize current weather information to make future predictions. However, this method also lacks the ability to forecast local weather conditions. Compared to traditional forecasting methods, machine learning systems require significantly less computational power. NVIDIA's Earth-2 model reduces energy consumption to approximately one-thousandth. The Indian Meteorological Department (IMD) has started experimenting with artificial intelligence in real-time and short-range weather forecasting (from 3 hours to 7 days). Artificial intelligence models can provide accurate and tailored predictions for specific user needs by examining subtle details in data that traditional methods may overlook. However, given the current challenges, AI models are unlikely to replace traditional methods. Source: https://analyticsindiamag.com/how-big-tech-are-tackling-climate-change/

RECENT AI NEWS

RECENT AI TOOLS