<p><strong class="journal-contentHeaderColor">Abstract.</strong> A new model is presented for multisite statistical downscaling of temperature and precipitation using convolutional conditional neural processes
(convCNPs). ConvCNPs are a recently developed class of models that allow deep-learning techniques to be applied to off-the-grid spatio-temporal
data. In contrast to existing methods that map from low-resolution model output to high-resolution predictions at a discrete set of locations, this
model outputs a stochastic process that can be queried at an arbitrary latitude–longitude coordinate. The convCNP model is shown to outperform an
ensemble of existing downscaling techniques over Europe for both temperature and precipitation taken from the VALUE intercomparison project. The
model also outperforms an approach that uses Gaussian processes to interpolate single-site downscaling models at unseen locations. Importantly,
substantial improvement is seen in the representation of extreme precipitation events. These results indicate that the convCNP is a robust
downscaling model suitable for generating localised projections for use in climate impact studies.</p>
Описание
GMD - Peer review - Convolutional conditional neural processes for local climate downscaling
%0 Journal Article
%1 vaughanconvolutional
%A Vaughan, Anna
%A Tebbutt, Will
%A Hosking, J. Scott
%A Turner, Richard E.
%D 2022
%I Copernicus GmbH
%J Geoscientific Model Development
%K ML downscaling stats
%N 1
%P 251-268
%R https://doi.org/10.5194/gmd-15-251-2022
%T Convolutional conditional neural processes for local climate downscaling
%U https://gmd.copernicus.org/articles/15/251/2022/gmd-15-251-2022-discussion.html
%V 15
%X <p><strong class="journal-contentHeaderColor">Abstract.</strong> A new model is presented for multisite statistical downscaling of temperature and precipitation using convolutional conditional neural processes
(convCNPs). ConvCNPs are a recently developed class of models that allow deep-learning techniques to be applied to off-the-grid spatio-temporal
data. In contrast to existing methods that map from low-resolution model output to high-resolution predictions at a discrete set of locations, this
model outputs a stochastic process that can be queried at an arbitrary latitude–longitude coordinate. The convCNP model is shown to outperform an
ensemble of existing downscaling techniques over Europe for both temperature and precipitation taken from the VALUE intercomparison project. The
model also outperforms an approach that uses Gaussian processes to interpolate single-site downscaling models at unseen locations. Importantly,
substantial improvement is seen in the representation of extreme precipitation events. These results indicate that the convCNP is a robust
downscaling model suitable for generating localised projections for use in climate impact studies.</p>
@article{vaughanconvolutional,
abstract = {<p><strong class="journal-contentHeaderColor">Abstract.</strong> A new model is presented for multisite statistical downscaling of temperature and precipitation using convolutional conditional neural processes
(convCNPs). ConvCNPs are a recently developed class of models that allow deep-learning techniques to be applied to off-the-grid spatio-temporal
data. In contrast to existing methods that map from low-resolution model output to high-resolution predictions at a discrete set of locations, this
model outputs a stochastic process that can be queried at an arbitrary latitude–longitude coordinate. The convCNP model is shown to outperform an
ensemble of existing downscaling techniques over Europe for both temperature and precipitation taken from the VALUE intercomparison project. The
model also outperforms an approach that uses Gaussian processes to interpolate single-site downscaling models at unseen locations. Importantly,
substantial improvement is seen in the representation of extreme precipitation events. These results indicate that the convCNP is a robust
downscaling model suitable for generating localised projections for use in climate impact studies.</p>},
added-at = {2022-02-11T10:36:20.000+0100},
author = {Vaughan, Anna and Tebbutt, Will and Hosking, J. Scott and Turner, Richard E.},
biburl = {https://www.bibsonomy.org/bibtex/2d658347cb765878decf3e8e5c793835e/simon.brown},
description = {GMD - Peer review - Convolutional conditional neural processes for local climate downscaling},
doi = {https://doi.org/10.5194/gmd-15-251-2022},
interhash = {9df4df6bb31c56e37264af980bbf7744},
intrahash = {d658347cb765878decf3e8e5c793835e},
issn = {1991-959X},
journal = {Geoscientific Model Development},
keywords = {ML downscaling stats},
number = 1,
pages = {251-268},
publisher = {Copernicus GmbH},
timestamp = {2022-02-11T10:36:20.000+0100},
title = {Convolutional conditional neural processes for local climate downscaling},
type = {misc},
url = {https://gmd.copernicus.org/articles/15/251/2022/gmd-15-251-2022-discussion.html},
volume = 15,
year = 2022
}