Deep learning for climate patterns

Abstract

Abstract Deep learning provides a unique nonlinear tool for calibration, simulation, and prediction in diagnosis and mining of big climate data. It is used to not only discover evolving patterns of climate change, but also predict climate change impacts. The main advantage is that deep learning makes full use of some unknown information hidden in big climate data, although they cannot be extracted directly. The most important deep learning method for climate patterns is the neural network, which employs a massive interconnection of simple computing cells called neurons (or processing units). These neurons have a natural propensity for storing observation knowledge and making it available for use. A neural network resembles the brain in the following two respects: (1) the neural network acquires knowledge from the observation of its environment through a learning process (or learning algorithm); (2) the interneuron connection strengths, which are called synaptic weights, are used to store the acquired knowledge.