Search Results
climate change ( Schneider et al. 2017 ). Owing to advances in both computing and available datasets, machine learning (ML) is now a viable alternative for traditional parameterization. Viewed from the perspective of ML, parameterization is a straightforward regression problem. A parameterization maps a set of inputs, namely, atmospheric profiles of humidity and temperature, to some outputs, profiles of subgrid heating and moistening. Krasnopolsky et al. (2005) and Chevallier et al. (1998
climate change ( Schneider et al. 2017 ). Owing to advances in both computing and available datasets, machine learning (ML) is now a viable alternative for traditional parameterization. Viewed from the perspective of ML, parameterization is a straightforward regression problem. A parameterization maps a set of inputs, namely, atmospheric profiles of humidity and temperature, to some outputs, profiles of subgrid heating and moistening. Krasnopolsky et al. (2005) and Chevallier et al. (1998
sensitivity of the cost function to the network weights or state space variables, respectively. Table 2. Comparison between typical machine learning (e.g., a deep neural network in TensorFlow) and data assimilation, which underpins most global weather forecasting. To highlight the similarities, NN concepts have been written in a linear algebra style close to typical DA notation. Superscript T denotes the transpose operator, bold lowercase letters are vectors and bold uppercase letters are matrices
sensitivity of the cost function to the network weights or state space variables, respectively. Table 2. Comparison between typical machine learning (e.g., a deep neural network in TensorFlow) and data assimilation, which underpins most global weather forecasting. To highlight the similarities, NN concepts have been written in a linear algebra style close to typical DA notation. Superscript T denotes the transpose operator, bold lowercase letters are vectors and bold uppercase letters are matrices
