Summary

Machine learning (ML), a major branch of artificial intelligence, has been advancing environmental science beyond what is possible with the traditional approaches of physics, chemistry, biology, and statistics. ML and statistics are both data science approaches; however, relative to statistics, ML trades off interpretability for prediction accuracy. Poor interpretability initially hindered the acceptance of ML methods in environmental science. ML methods are now widely used in the fields of atmospheric science, oceanography, cryospheric science, hydrology, forestry, agricultural science, and climate science.

The most common ML methods are neural network (NN) models, inspired by biological NNs in animal brains. Deep learning, that is, deep NN models, has become prominent since the mid-2010s, with the number of layers of mapping in deep NN models being much larger than in the earlier NN models.

ML methods were initially introduced into environmental science as nonlinear statistical tools, with no direct relation to numerical models based on physics (“physics” being used in the broadest sense, i.e., physics + chemistry + biology, etc.). The recent merging of the two entirely different approaches, ML and numerical modeling, points to a new future for environmental and climate science.