Search Results

You are looking at 1 - 10 of 106 items for :

  • Deep learning x
  • Monthly Weather Review x
  • All content x
Clear All
Ryan Lagerquist, Amy McGovern, Cameron R. Homeyer, David John Gagne II, and Travis Smith

( Cintineo et al. 2018 ). Convolutional neural networks (CNN) are specially designed to learn from spatial grids and often contain many layers, which qualifies them as a deep-learning method (section 1.1.4 of Chollet 2018 ). In traditional ML, spatial grids must be transformed into scalar features, which become the direct inputs to the model. Examples are principal components, spatial statistics (such as means and standard deviations), and raw gridpoint values (where each value in the grid is treated as

Restricted access
David John Gagne II, Sue Ellen Haupt, Douglas W. Nychka, and Gregory Thompson

localized variations in temperature and moisture. HAILCAST has demonstrated skill in diagnosing hail size from proximity soundings ( Jewell and Brimelow 2009 ) and convection-allowing model environments ( Adams-Selin et al. 2019 ). In this paper, we demonstrate that incorporating both vertical profile and spatial information into a deep learning hail size diagnostic model can provide both increased hail size analysis skill and insight into important factors for hail growth. The importance of storm

Full access
Yang Liu, Laurens Bogaardt, Jisk Attema, and Wilco Hazeleger

to forecast Arctic sea ice with a statistical model. This brings contemporary machine learning techniques into scope. Machine learning approaches, especially deep learning, are widely embraced by many fields and are increasingly used to deal with problems like clustering, classification, and regression ( LeCun et al. 2015 ). Benefiting from large volumes of data of Earth system ( Knüsel et al. 2019 ), those deep learning methods may be appropriate for the weather and climate domain ( Reichstein

Open access
Anthony Wimmers, Christopher Velden, and Joshua H. Cossuth

1. Introduction Deep learning (DL) is a newly popular, powerful and often confounding computational tool for developing predictive models in the sciences. It builds on a long legacy of neural network modeling, with a key feature being the organization of neural connections into multiple layers of nonlinear operations, enabling models to apply high levels of abstraction in their tasks. New hardware innovations, particularly in accessing graphical processing units (GPUs), have enabled DL

Full access
Jing-Yi Zhuo and Zhe-Min Tan

Abstract

A deep learning-based method augmented by prior knowledge of tropical cyclones (TCs), called DeepTCNet, is introduced to estimate TC intensity and wind radii from infrared (IR) imagery over the North Atlantic Ocean. While standard deep learning practices have many advantages over conventional analysis approaches and can produce reliable estimates of TCs, the data-driven models informed by machine-readable physical knowledge of TCs could achieve higher performance. To this end, two approaches are explored to develop the physics-augmented DeepTCNet: (i) infusing the auxiliary physical information of TCs into models for single-task learning; (ii) learning auxiliary physical tasks for multi-task learning. More specifically, augmented by auxiliary information of TC fullness (a measure of the radial decay of the TC wind field), the DeepTCNet yielded a 12% improvement in estimating TC intensity over the non-augmented one. By learning TC wind radii and auxiliary TC intensity task simultaneously, the model’s wind radii estimation skill is improved by 6% over only learning four wind radii tasks, and by 9% over separately learning a single wind radii task. The evaluation results showed that the DeepTCNet is in-line with the Satellite Consensus technique (SATCON) but systematically outperforms the Advanced Dvorak Technique (ADT) at all intensity scales with an averaged 39% enhancement in TC intensity estimation. The DeepTCNet also surpasses the Multi-platform Tropical Cyclone Surface Wind Analysis technique (MTCSWA) with an average improvement of 32% in wind radii estimation.

Restricted access
Simon Veldkamp, Kirien Whan, Sjoerd Dirksen, and Maurice Schmeits

adaptive moment estimation (Adam; Kingma and Ba 2014 ), a variant of stochastic gradient descent that is very popular in deep learning. We use early stopping to determine the number of epochs (the number of times the training data are used during training). The neural networks used in this research were programmed using Keras ( Chollet et al. 2015 ), with TensorFlow as backend ( Abadi et al. 2015 ). Adam was employed using default options for all parameters other than the learning rate decay parameter

Restricted access
John Bjørnar Bremnes

relations to the ensemble forecasts can be allowed for. By using the continuous ranked probability score as loss function and an estimation method developed for deep-learning problems ( Goodfellow et al. 2016 ) they demonstrated that the parameters could be efficiently estimated and skillful probabilistic forecasts could be made. The method proposed in this article can be seen as a generalization of the work by Rasp and Lerch (2018) and also as a way to deal with challenges in quantile regression

Open access
Stephan Rasp and Sebastian Lerch

parameters without having to specify appropriate link functions, and the ease of adding station information into a global model by using embeddings. The network model parameters are estimated by optimizing the CRPS, a nonstandard choice in the machine learning literature tailored to probabilistic forecasting. Furthermore, the rapid pace of development in the deep learning community provides flexible and efficient modeling techniques and software libraries. The presented approach can therefore be easily

Open access
Gregory R. Herman and Russ S. Schumacher

1. Introduction Machine learning algorithms have demonstrated considerable utility in many scientific disciplines, including computer vision (e.g., Rosten and Drummond 2006 ), natural language processing (e.g., Collobert et al. 2011 ), and bioinformatics (e.g., Larrañaga et al. 2006 ). Machine learning has also been used with considerable success in a wide range of future prediction scenarios, from financial market analysis (e.g., Cao and Tay 2003 ) to election forecasting (e

Full access
Filipe Aires, Francis Marquisseau, Catherine Prigent, and Geneviève Sèze

.8, 31.4, and 89 GHz. It is a cross-track scanning radiometer, with ±48.3° from nadir with a total of 30 earth fields of view of 3.3° per scan line, providing a nominal spatial resolution of 48 km at nadir. The swath is approximately 2000 km and the instrument realizes one scan in 8 s. The AMSU-B microwave radiometer is designed to measure the atmospheric water vapor profile, with three channels in the H 2 O line at 183.31 GHz plus two window channels at 89 and 150 GHz that enable deeper penetration

Full access