A Novel Bayesian Deep Learning Approach to the Downscaling of Wind Speed with Uncertainty Quantification

Firas Gerges, Michel C. Boufadel, Elie Bou-Zeid, Hani Nassif, Jason T.L. Wang

Research output: Chapter in Book/Report/Conference proceedingConference contribution

4 Scopus citations

Abstract

Wind plays a crucial part during adverse events, such as storms and wildfires, and is a widely leveraged source of renewable energy. Predicting long-term daily local wind speed is critical for effective monitoring and mitigation of climate change, as well as to locate suitable locations for wind farms. Long-term simulations of wind dynamics (until year 2100) are given by various general circulation models (GCMs). However, GCM simulations are at a grid with coarse spatial resolution (>100 km), which renders spatial downscaling to a smaller scale an important prerequisite for climate-impacts studies. In this work, we propose a novel deep learning approach, named Bayesian AIG-Transformer, that consists of an attention-based input grouping (AIG), transformer, and uncertainty quantification. We use the proposed approach for the spatial downscaling of daily average wind speed (AWND), formulated as a multivariate time series forecasting problem, over four locations within New Jersey and Pennsylvania. To calibrate and evaluate our deep learning approach, we use large-scale observations extracted from NOAA’s NCEP/NCAR reanalysis dataset (2.5° × 2.5° resolution), which provides a proxy for GCM data when evaluating the model. Results show that our approach is suitable for the downscaling task, outperforming related machine learning methods.

Original languageEnglish (US)
Title of host publicationAdvances in Knowledge Discovery and Data Mining - 26th Pacific-Asia Conference, PAKDD 2022, Proceedings
EditorsJoão Gama, Tianrui Li, Yang Yu, Enhong Chen, Yu Zheng, Fei Teng
PublisherSpringer Science and Business Media Deutschland GmbH
Pages55-66
Number of pages12
ISBN (Print)9783031059803
DOIs
StatePublished - 2022
Externally publishedYes
Event26th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2022 - Chengdu, China
Duration: May 16 2022May 19 2022

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13282 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference26th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2022
Country/TerritoryChina
CityChengdu
Period5/16/225/19/22

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • General Computer Science

Keywords

  • Climate change
  • Deep learning
  • Time series forecasting
  • Transformer
  • Wind speed

Fingerprint

Dive into the research topics of 'A Novel Bayesian Deep Learning Approach to the Downscaling of Wind Speed with Uncertainty Quantification'. Together they form a unique fingerprint.

Cite this