A new distribution mapping technique for climate model bias correction

Seth McGinnis; Doug Nychka; Linda O. Mearns

doi:10.1007/978-3-319-17220-0_9

A new distribution mapping technique for climate model bias correction

Seth McGinnis, Doug Nychka, Linda O. Mearns

Regional Integrated Sciences Collective

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

We evaluate the performance of different distribution mapping techniques for bias correction of climate model output by operating on synthetic data and comparing the results to an “oracle” correction based on perfect knowledge of the generating distributions. We find results consistent across six different metrics of performance. Techniques based on fitting a distribution perform best on data from normal and gamma distributions, but are at a significant disadvantage when the data does not come from a known parametric distribution. The technique with the best overall performance is a novel nonparametric technique, kernel density distribution mapping (KDDM).

Original language	English
Title of host publication	Machine Learning and Data Mining Approaches to Climate Science
Editors	Valliappa Lakshmanan, Eric Gilleland, Amy McGovern, Martin Tingley
Place of Publication	Cham
Publisher	Springer International Publishing
Pages	91-99
Number of pages	9
ISBN (Print)	9783319172200
DOIs	https://doi.org/10.1007/978-3-319-17220-0_9
State	Published - 2015

Keywords

KDDM
Nonparametric distribution
Oracle evaluation
Quantile mapping
Transfer function

Access to Document

10.1007/978-3-319-17220-0_9

Cite this

@inproceedings{d9347f5d7c9a4c9c912fab5d3d5898c7,

title = "A new distribution mapping technique for climate model bias correction",

abstract = "We evaluate the performance of different distribution mapping techniques for bias correction of climate model output by operating on synthetic data and comparing the results to an “oracle” correction based on perfect knowledge of the generating distributions. We find results consistent across six different metrics of performance. Techniques based on fitting a distribution perform best on data from normal and gamma distributions, but are at a significant disadvantage when the data does not come from a known parametric distribution. The technique with the best overall performance is a novel nonparametric technique, kernel density distribution mapping (KDDM).",

keywords = "KDDM, Nonparametric distribution, Oracle evaluation, Quantile mapping, Transfer function",

author = "Seth McGinnis and Doug Nychka and Mearns, \{Linda O.\}",

year = "2015",

doi = "10.1007/978-3-319-17220-0\_9",

language = "English",

isbn = "9783319172200",

pages = "91--99",

editor = "Valliappa Lakshmanan and Eric Gilleland and Amy McGovern and Martin Tingley",

booktitle = "Machine Learning and Data Mining Approaches to Climate Science",

publisher = "Springer International Publishing",

address = "Switzerland",

}

A new distribution mapping technique for climate model bias correction. / McGinnis, Seth; Nychka, Doug; Mearns, Linda O.
Machine Learning and Data Mining Approaches to Climate Science. ed. / Valliappa Lakshmanan; Eric Gilleland; Amy McGovern; Martin Tingley. Cham: Springer International Publishing, 2015. p. 91-99.

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - A new distribution mapping technique for climate model bias correction

AU - McGinnis, Seth

AU - Nychka, Doug

AU - Mearns, Linda O.

PY - 2015

Y1 - 2015

N2 - We evaluate the performance of different distribution mapping techniques for bias correction of climate model output by operating on synthetic data and comparing the results to an “oracle” correction based on perfect knowledge of the generating distributions. We find results consistent across six different metrics of performance. Techniques based on fitting a distribution perform best on data from normal and gamma distributions, but are at a significant disadvantage when the data does not come from a known parametric distribution. The technique with the best overall performance is a novel nonparametric technique, kernel density distribution mapping (KDDM).

AB - We evaluate the performance of different distribution mapping techniques for bias correction of climate model output by operating on synthetic data and comparing the results to an “oracle” correction based on perfect knowledge of the generating distributions. We find results consistent across six different metrics of performance. Techniques based on fitting a distribution perform best on data from normal and gamma distributions, but are at a significant disadvantage when the data does not come from a known parametric distribution. The technique with the best overall performance is a novel nonparametric technique, kernel density distribution mapping (KDDM).

KW - KDDM

KW - Nonparametric distribution

KW - Oracle evaluation

KW - Quantile mapping

KW - Transfer function

U2 - 10.1007/978-3-319-17220-0_9

DO - 10.1007/978-3-319-17220-0_9

M3 - Conference contribution

SN - 9783319172200

SP - 91

EP - 99

BT - Machine Learning and Data Mining Approaches to Climate Science

A2 - Lakshmanan, Valliappa

A2 - Gilleland, Eric

A2 - McGovern, Amy

A2 - Tingley, Martin

PB - Springer International Publishing

CY - Cham

ER -

A new distribution mapping technique for climate model bias correction

Abstract

Keywords

Access to Document

Fingerprint

Cite this