Probabilistic forecasting using deep generative models
- PDF / 1,871,770 Bytes
- 21 Pages / 439.642 x 666.49 pts Page_size
- 34 Downloads / 275 Views
Probabilistic forecasting using deep generative models Alessandro Fanfarillo1
· Behrooz Roozitalab2,3 · Weiming Hu4 · Guido Cervone1,4
Received: 27 September 2019 / Revised: 2 September 2020 / Accepted: 18 September 2020 / © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract A fundamental problem in Numerical Weather Prediction (NWP) is the generation of ensembles to capture the probability of future states of the atmosphere. This research presents a new methodology to generate analogs using deep generative models, an emerging class of deep learning approaches. The goal is to train a deep generative model using a set of historical forecasts and associated observations, and to use it to entirely or partially replace the need to maintain a potentially very large dataset. In this research, this new methodology is compared with the Analog Ensemble (AnEn) approach, a computationally efficient solution to generate analogs. The proposed approach promises to reduce the amount of memory required to produce the probabilistic forecast by several orders of magnitude. Results show that the generative model solution is constant time without performing any search, saving a considerable amount of time even in the presence of huge historical datasets. Keywords Computational algorithms · Ensemble modeling · Deep learning
1 Introduction Quick and accurate weather prediction is an essential and critical part for decision-making, in particular when human lives are at stake. It has been counted that more than 200 people died in 2018 because of all weather and climate events over U.S. and the damages has been estimated to be more than 90 billion dollars.1 However, these losses would be far higher if scientists had not predicted these events. NWP model forecast is usually used as the main
1 https://www.ncdc.noaa.gov/billions/
Alessandro Fanfarillo
[email protected] 1
Research Applications Laboratory, National Center for Atmospheric Research, Boulder, CO, USA
2
Department of Chemical and Biochemical Engineering, The University of Iowa, Iowa City, IA, USA
3
Center of Global and Regional Environmental Research, The University of Iowa, Iowa City, IA, USA
4
Department of Geography and Institute for CyberScience, Geoinformatics and Earth Observation Laboratory, The Pennsylvania State University, University Park, State College, PA, USA
Geoinformatica
tool for weather prediction. However, its utility is limited as it represents only a single plausible future state of the atmosphere. In fact, imperfect initial conditions and model deficiencies can lead to model errors that grow non-linearly during the model evolution. Lorenz [27] pointed out that, “the errors in estimating the current state of the atmosphere are due mainly to omission rather than inaccuracy”. In other words, the errors can be related to the gaps in science or computational resources and the uncertainty in input data. However, current NWP models are state-of-the-science models that include the most recent discovered science. So, reducing t
Data Loading...