statistical downscaling & bias correction of climate ... · • gcm predictors (in sdms) are...
TRANSCRIPT
![Page 1: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/1.jpg)
International MISTRALS workshop on "Climate change impacts in the Mediterranean region" Montpellier, France, October, 16-18, 2017
Mathieu Vrac
Statistical downscaling & bias correction of climate simulations:
Main findings of the StaRMIP project
![Page 2: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/2.jpg)
Mathieu Vrac
Motivations
![Page 3: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/3.jpg)
Mathieu Vrac
Motivations
● 30% of the world economic activities are affected by meteorological conditions (source: IPCC)
![Page 4: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/4.jpg)
Mathieu Vrac
Motivations
● 30% of the world economic activities are affected by meteorological conditions (source: IPCC)
● IPCC scenarios of climate change have a coarse spatial resolution !! Not adapted to social and economic scales of impact studies
Ø Social and economic impacts: water resources, hydrology, agriculture, air pollution, human health, etc.
Ø How will climate change interact with existing environmental features at a regional/local scale ?
![Page 5: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/5.jpg)
Mathieu Vrac
Motivations
● 30% of the world economic activities are affected by meteorological conditions (source: IPCC)
● IPCC scenarios of climate change have a coarse spatial resolution !! Not adapted to social and economic scales of impact studies
Ø Social and economic impacts: water resources, hydrology, agriculture, air pollution, human health, etc.
Ø How will climate change interact with existing environmental features at a regional/local scale ?
● Downscaling: To derive sub-grid scale (regional or local) weather or climate using General Circulation Models (GCMs) outputs or reanalysis data (e.g. NCEP)
![Page 6: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/6.jpg)
Mathieu Vrac
Motivations
● 30% of the world economic activities are affected by meteorological conditions (source: IPCC)
● IPCC scenarios of climate change have a coarse spatial resolution !! Not adapted to social and economic scales of impact studies
Ø Social and economic impacts: water resources, hydrology, agriculture, air pollution, human health, etc.
Ø How will climate change interact with existing environmental features at a regional/local scale ?
● Downscaling: To derive sub-grid scale (regional or local) weather or climate using General Circulation Models (GCMs) outputs or reanalysis data (e.g. NCEP)
● Statistical Bias Correction also often needed !!
![Page 7: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/7.jpg)
≈ 250 km
Region, city, fields, station
Coarse atmospheric data Precipitation, temperature, humidity,
geopotential, wind, etc.
How to downscale?: The basics
Local variables (e.g., precip., temp.) (small scale water cycle, impacts – crops, resources – etc.)
How to use the coarse simulations to produce regional/local climate features?
?
![Page 8: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/8.jpg)
≈ 250 km Coarse atmospheric data
Precipitation, temperature, humidity, geopotential, wind, etc.
Dynamical downscaling (RCMs): ● GCMs to drive regional models (5-50km) determining atmosphere dynamics
● Requires a lot of computer time and resources => Limited applications
Local variables (e.g., precip., temp.) (small scale water cycle, impacts – crops, resources – etc.)
Region, city, fields, station
How to downscale?: The basics
![Page 9: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/9.jpg)
≈ 250 km Coarse atmospheric data
Precipitation, temperature, humidity, geopotential, wind, etc.
Dynamical downscaling (RCMs): ● GCMs to drive regional models (5-50km) determining atmosphere dynamics
● Requires a lot of computer time and resources => Limited applications
Statistical downscaling: ● Based on statistical relationships between large- and local-scale variables
● Low costs and rapid simulations applicable to any spatial resolution
● Uncertainties (results, propagation, etc)
Local variables (e.g., precip., temp.) (small scale water cycle, impacts – crops, resources – etc.)
Region, city, fields, station
How to downscale?: The basics
![Page 10: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/10.jpg)
StaRMIP:
Statistical Regionalization (downscaling) Models Intercomparison and Hydrological Impacts
Project
Funded by ANR (French National Research Agency), 2013-2017
- Evaluation & intercomparisons of SDMs - Developments of BC & SDMs
Goals (in a nutshell):
![Page 11: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/11.jpg)
• Intercomparisons / Evaluations of SDMs Ø The climate point-of-view (Vaittinada Ayar et al., 2016) Ø The hydrological point-of-view (Grouillet et al., 2016)
Vario
us e
valu
atio
ns
StaRMIP: Main goals & outline
![Page 12: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/12.jpg)
• Intercomparisons / Evaluations of SDMs Ø The climate point-of-view (Vaittinada Ayar et al., 2016) Ø The hydrological point-of-view (Grouillet et al., 2016)
• GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs
(Vrac & Vaittinada Ayar, 2016; Volosciuk et al., 2016) Vario
us e
valu
atio
ns
StaRMIP: Main goals & outline
![Page 13: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/13.jpg)
• Intercomparisons / Evaluations of SDMs Ø The climate point-of-view (Vaittinada Ayar et al., 2016) Ø The hydrological point-of-view (Grouillet et al., 2016)
• GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs
(Vrac & Vaittinada Ayar, 2016; Volosciuk et al., 2016)
• "Spatial" (i.e. field) downscaling Ø Extreme fields of precipitation (SHD, Bechler et al., 2015) Ø Non-stationary spatial SWG for precipitation (See Pradeebane’s prez)
Vario
us e
valu
atio
ns D
evel
opm
ents
StaRMIP: Main goals & outline
![Page 14: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/14.jpg)
• Intercomparisons / Evaluations of SDMs Ø The climate point-of-view (Vaittinada Ayar et al., 2016) Ø The hydrological point-of-view (Grouillet et al., 2016)
• GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs
(Vrac & Vaittinada Ayar, 2016; Volosciuk et al., 2016)
• "Spatial" (i.e. field) downscaling Ø Extreme fields of precipitation (SHD, Bechler et al., 2015) Ø Non-stationary spatial SWG for precipitation (See Pradeebane’s prez)
• BC of precipitation (Stochastic Singular Removal, SSR, Vrac et al., 2016)
Vario
us e
valu
atio
ns D
evel
opm
ents
StaRMIP: Main goals & outline
![Page 15: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/15.jpg)
• Intercomparisons / Evaluations of SDMs Ø The climate point-of-view (Vaittinada Ayar et al., 2016) Ø The hydrological point-of-view (Grouillet et al., 2016)
• GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs
(Vrac & Vaittinada Ayar, 2016; Volosciuk et al., 2016)
• "Spatial" (i.e. field) downscaling Ø Extreme fields of precipitation (SHD, Bechler et al., 2015) Ø Non-stationary spatial SWG for precipitation (See Pradeebane’s prez)
• BC of precipitation (Stochastic Singular Removal, SSR, Vrac et al., 2016)
• Multivariate (multi-sites/variables) BC & SDMs Ø EC-BC (Vrac & Friederichs, 2015) Ø R2BC (Vrac, in prep.) & "Optimal" BC (Robin et al, in prep.) Ø Inter-comparison of Multivariate-BC (François et al., in prep.)
Vario
us e
valu
atio
ns D
evel
opm
ents
StaRMIP: Main goals & outline
![Page 16: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/16.jpg)
• Intercomparisons / Evaluations of SDMs Ø The climate point-of-view (Vaittinada Ayar et al., 2016) Ø The hydrological point-of-view (Grouillet et al., 2016)
• GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs
(Vrac & Vaittinada Ayar, 2016; Volosciuk et al., 2016)
• "Spatial" (i.e. field) downscaling Ø Extreme fields of precipitation (SHD, Bechler et al., 2015) Ø Non-stationary spatial SWG for precipitation (See Pradeebane’s prez)
• BC of precipitation (Stochastic Singular Removal, SSR, Vrac et al., 2016)
• Multivariate (multi-sites/variables) BC & SDMs Ø EC-BC (Vrac & Friederichs, 2015) Ø R2BC (Vrac, in prep.) & "Optimal" BC (Robin et al, in prep.) Ø Inter-comparison of Multivariate-BC (François et al., in prep.)
Vario
us e
valu
atio
ns D
evel
opm
ents
StaRMIP: Main goals & outline
1
![Page 17: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/17.jpg)
• Intercomparisons / Evaluations of SDMs Ø The climate point-of-view (Vaittinada Ayar et al., 2016) Ø The hydrological point-of-view (Grouillet et al., 2016)
• GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs
(Vrac & Vaittinada Ayar, 2016; Volosciuk et al., 2016)
• "Spatial" (i.e. field) downscaling Ø Extreme fields of precipitation (SHD, Bechler et al., 2015) Ø Non-stationary spatial SWG for precipitation (See Pradeebane’s prez)
• BC of precipitation (Stochastic Singular Removal, SSR, Vrac et al., 2016)
• Multivariate (multi-sites/variables) BC & SDMs Ø EC-BC (Vrac & Friederichs, 2015) Ø R2BC (Vrac, in prep.) & "Optimal" BC (Robin et al, in prep.) Ø Inter-comparison of Multivariate-BC (François et al., in prep.)
Vario
us e
valu
atio
ns D
evel
opm
ents
StaRMIP: Main goals & outline
1
2
![Page 18: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/18.jpg)
• Intercomparisons / Evaluations of SDMs Ø The climate point-of-view (Vaittinada Ayar et al., 2016) Ø The hydrological point-of-view (Grouillet et al., 2016)
• GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs
(Vrac & Vaittinada Ayar, 2016; Volosciuk et al., 2016)
• "Spatial" (i.e. field) downscaling Ø Extreme fields of precipitation (SHD, Bechler et al., 2015) Ø Non-stationary spatial SWG for precipitation (See Pradeebane’s prez)
• BC of precipitation (Stochastic Singular Removal, SSR, Vrac et al., 2016)
• Multivariate (multi-sites/variables) BC & SDMs Ø EC-BC (Vrac & Friederichs, 2015) Ø R2BC (Vrac, in prep.) & "Optimal" BC (Robin et al, in prep.) Ø Inter-comparison of Multivariate-BC (François et al., in prep.)
Vario
us e
valu
atio
ns D
evel
opm
ents
StaRMIP: Main goals & outline
1
2
3
![Page 19: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/19.jpg)
Intercomparison of SDMs and RCMs in CORDEX
From Vaittinada Ayar, P., Vrac, M., Bastin, S., Carreau, J., Déqué, M., Gallardo, C. (2016) Intercomparison of statistical and dynamical downscaling models under the EURO-
and MED-CORDEX initiative framework: Present climate evaluations. Climate Dynamics, 46: 1301. https://doi.org/10.1007/s00382-015-2647-5
Part 1
StaRMIP goal 1: Intercomparisons & guidelines The climate point-of-view
![Page 20: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/20.jpg)
• Main statistical approaches for downscaling
Local variables (e.g., precip., temp.) (small scale water cycle, impacts – crops, resources – etc.)
Coarse atmospheric predictors Precip., temp., humidity, geopot., wind, etc.
StaRMIP goal 1: Intercomparisons & guidelines The climate point-of-view
![Page 21: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/21.jpg)
Transfer f°
Linear Non-linear
Coarse atmospheric predictors Precip., temp., humidity, geopot., wind, etc.
StaRMIP goal 1: Intercomparisons & guidelines The climate point-of-view
• Main statistical approaches for downscaling
![Page 22: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/22.jpg)
Transfer f° Weather generator
Linear Non-linear Stochast. methods
Coarse atmospheric predictors Precip., temp., humidity, geopot., wind, etc.
StaRMIP goal 1: Intercomparisons & guidelines The climate point-of-view
• Main statistical approaches for downscaling
![Page 23: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/23.jpg)
Transfer f° Weather generator
Analogues Weather typing Linear Non-linear Stochast. methods
Clustering
Coarse atmospheric predictors Precip., temp., humidity, geopot., wind, etc.
StaRMIP goal 1: Intercomparisons & guidelines The climate point-of-view
• Main statistical approaches for downscaling
![Page 24: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/24.jpg)
Transfer f° Mod. Output Stat. Weather generator
Analogues Weather typing Linear Non-linear Quantile- Matching
Stochast. methods
Clustering
Coarse atmospheric predictors Precip., temp., humidity, geopot., wind, etc.
StaRMIP goal 1: Intercomparisons & guidelines The climate point-of-view
• Main statistical approaches for downscaling
![Page 25: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/25.jpg)
Ø Reference & predictand E-OBS (Haylock et al., 2008) precipitation at 0.44° resolution
Cross-validation set-up
−60 −40 −20 0 20 40 60 80
2030
4050
6070
0
500
1000
1500
2000
2500
E−OBS Domain
Altitude (m)Longitude
Latit
ude
![Page 26: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/26.jpg)
Ø Reference & predictand E-OBS (Haylock et al., 2008) precipitation at 0.44° resolution
Ø Predictors: first PC of SLP, D2, T2, R850, U850, V850, Z850, (ERA-I reanalyses, 1.125°)
Cross-validation set-up
![Page 27: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/27.jpg)
Ø Reference & predictand E-OBS (Haylock et al., 2008) precipitation at 0.44° resolution
Ø Predictors: first PC of SLP, D2, T2, R850, U850, V850, Z850, (ERA-I reanalyses, 1.125°)
Ø Calibration over the period 1979-2008 (2 x 20 years) Ø Evaluation over the period 1989-2008 = CORDEX RCM runs period
Cross-validation set-up
![Page 28: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/28.jpg)
Ø Reference & predictand E-OBS (Haylock et al., 2008) precipitation at 0.44° resolution
Ø Predictors: first PC of SLP, D2, T2, R850, U850, V850, Z850, (ERA-I reanalyses, 1.125°)
Ø Calibration over the period 1979-2008 (2 x 20 years) Ø Evaluation over the period 1989-2008 = CORDEX RCM runs period
Ø 6-month “summer” (15 Apr.-14 Oct.) Ø 6-month “winter” (15 Oct.-14 Apr.) Ø Occurrence threshold = 1mm/day
Cross-validation set-up
![Page 29: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/29.jpg)
StaRMIP goal 1: Intercomparisons & guidelines The climate point-of-view
=> 11 models
![Page 30: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/30.jpg)
![Page 31: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/31.jpg)
![Page 32: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/32.jpg)
Summary
![Page 33: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/33.jpg)
Summary
![Page 34: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/34.jpg)
Summary
![Page 35: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/35.jpg)
Summary
![Page 36: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/36.jpg)
Summary
Ø No model really takes the advantage on the others Ø Ranking is indicator-dependent and therefore
depends on end-users needs ! Ø This work provides ⇒ SDMs simulations within the EURO- & MED-
CORDEX initiative ⇒ a methodology to select the most suited
simulations/models for end-users needs. (Philosophy similar to the COST Action VALUE)
![Page 37: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/37.jpg)
From
Grouillet, B., Ruelland, D., Vaittinada Ayar, P. & Vrac, M. (2016). Sensitivity analysis of runoff modelling to statistical downscaling
methods in the western Mediterranean. Hydrol. & Earth Syst. Sci., 20, 1031‒1047.
Part 2
StaRMIP goal 2: Intercomparisons & guidelines The hydrological point-of-view
![Page 38: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/38.jpg)
Evaluation on 4 Mediterranean non-influenced basins
![Page 39: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/39.jpg)
Low resolution climate data
• NCEP/NCAR daily reanalysis data with a 2.5° spatial resolution
• IPSL-CM5A-MR GCM, regridded at a 2.5° spatial resolution
• CNRM-CM5 GCM, regridded at a 2.5° spatial resolution
(Kalnay et al., 1996)
(Dufresne et al., 2013)
(Voldoire et al., 2013)
![Page 40: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/40.jpg)
Statistical downscaling models (SDMs)
Representing the main families of SDMs: • analogs of atmospheric circulation patterns
(Analog) • the “Cumulative Distribution Function -
transform” approach (CDFt) • a stochastic weather generator (SWG)
![Page 41: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/41.jpg)
High resolution climate series
![Page 42: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/42.jpg)
Hydrological modeling
![Page 43: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/43.jpg)
Runoff simulations
![Page 44: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/44.jpg)
Sensitivity analysis of hydrological responses
![Page 45: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/45.jpg)
Comparison of runoff simulations to provide guidelines
![Page 46: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/46.jpg)
Guideline to select SDMs from a hydrological point of view
Median of the criterion values from the four basins
![Page 47: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/47.jpg)
Guideline to select SDMs from a hydrological point of view
![Page 48: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/48.jpg)
Guideline to select SDMs from a hydrological point of view
![Page 49: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/49.jpg)
Guideline to select SDMs from a hydrological point of view
![Page 50: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/50.jpg)
Guideline to select SDMs from a hydrological point of view
![Page 51: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/51.jpg)
Outcome
CDFt
ANALOG
SWG
Bestresultsbutneversuppliesdownscaledvaluesoutoftherangeofthecalibra6onreferencedataset
Goodperformanceandcouldbeimprovedwithaddi6onalcovariatesaspredictor
![Page 52: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/52.jpg)
Reduce downscaling uncertainty in climate scenarios
• 2GCMsCNRM/IPSL
• 2RCPs
RCP4.5/8.5• 3SDMs
ANALOG
CDFt
SWG
Uncertaintyof12runoffsimula6onsoverthe2041─2060periodbasedon:
![Page 53: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/53.jpg)
Reduce downscaling uncertainty in climate scenarios
Uncertaintyof12runoffsimula6onsoverthe2041─2060periodbasedon:
• 2GCMsCNRM/IPSL
• 2RCPs
RCP4.5/8.5• 2SDMs
ANALOG
CDFt
SWG
![Page 54: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/54.jpg)
Outcome
CDFt
ANALOG
SWG
Bestresultsbutneversuppliesdownscaledvaluesoutoftherangeofthecalibra6onreferencedataset
Goodperformanceandcouldbeimprovedwithaddi6onalcovariatesaspredictor
UnsuitableasisbecauseofthebiasesbetweenNCEP/NCAR(calibra?on)andGCMspredictors.
Onesolu6on:correc6onoftheGCMspredictors
![Page 55: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/55.jpg)
What is the influence of bias correcting predictors on statistical downscaling models ?
From
Vrac, M. and P. Vaittinada Ayar (2017) Influence of Bias Correcting Predictors on Statistical
Downscaling Models. J. Appl. Meteor. Climatol., 56, 5–26, https://doi.org/10.1175/
JAMC-D-16-0079.1
Combining BC & SDMs
![Page 56: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/56.jpg)
What is the influence of bias correcting predictors on statistical downscaling models ?
From
Vrac, M. and P. Vaittinada Ayar (2017) Influence of Bias Correcting Predictors on Statistical
Downscaling Models. J. Appl. Meteor. Climatol., 56, 5–26, https://doi.org/10.1175/
JAMC-D-16-0079.1
Combining BC & SDMs
Sorry... Another time...
![Page 57: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/57.jpg)
• SDMs & BC are often univariate (although covariates):
Part 3 Statistical models developments in StaRMIP
![Page 58: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/58.jpg)
• SDMs & BC are often univariate (although covariates):
⇒ May generate unrealistic situations (w/ wrong multidim. properties)
Part 3 Statistical models developments in StaRMIP
![Page 59: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/59.jpg)
• SDMs & BC are often univariate (although covariates):
⇒ May generate unrealistic situations (w/ wrong multidim. properties) ⇒ Needs for inter-sites models (stations or grid-cells)
⇒ Needs for inter-variables models (b/ climate variables) ? MOS ? SWGs?
Part 3 Statistical models developments in StaRMIP
![Page 60: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/60.jpg)
• SDMs & BC are often univariate (although covariates):
⇒ May generate unrealistic situations (w/ wrong multidim. properties) ⇒ Needs for inter-sites models (stations or grid-cells)
⇒ Needs for inter-variables models (b/ climate variables) ? MOS ? SWGs?
⇒ Needs for spatial models: SD even at locations where no data ? Continuous spatial processes (SWGs)
Ø Non-stationary spatial downscaling: Pradeebane’s talk Ø Spatial downscaling of extreme fields of precipitation
(Bechler et al., 2015a, 2015b)
Part 3 Statistical models developments in StaRMIP
![Page 61: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/61.jpg)
• SDMs & BC are often univariate (although covariates):
⇒ May generate unrealistic situations (w/ wrong multidim. properties) ⇒ Needs for inter-sites models (stations or grid-cells)
⇒ Needs for inter-variables models (b/ climate variables) ? MOS ? SWGs?
⇒ Needs for spatial models: SD even at locations where no data ? Continuous spatial processes (SWGs)
Ø Non-stationary spatial downscaling: Pradeebane’s talk Ø Spatial downscaling of extreme fields of precipitation
(Bechler et al., 2015a, 2015b)
Part 3 Statistical models developments in StaRMIP
![Page 62: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/62.jpg)
Multivariate bias correction 2 main philosophies
![Page 63: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/63.jpg)
Multivariate bias correction 2 main philosophies
• Bardossy et al. (2012, “matrix recorrelation”)
• Vrac & Friederichs (2015, EC-BC) • Cannon (2017, MBCn) • etc.
Parallel correction of marginals & (rank) dependence
![Page 64: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/64.jpg)
Multivariate bias correction 2 main philosophies
Successive conditional corrections
• Bardossy et al. (2012, “matrix recorrelation”)
• Vrac & Friederichs (2015, EC-BC) • Cannon (2017, MBCn) • etc.
• Bardossy et al. (2012, “sequential recorrelation”)
• Piani et al. (2012) • Dekens et al. (2017) • etc.
Parallel correction of marginals & (rank) dependence
![Page 65: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/65.jpg)
Multivariate bias correction 2 main philosophies
Parallel correction of marginals & (rank) dependence
Successive conditional corrections
• Bardossy et al. (2012, “matrix recorrelation”)
• Vrac & Friederichs (2015, EC-BC) • Cannon (2017, MBCn) • etc.
• Bardossy et al. (2012, “sequential recorrelation”)
• Piani et al. (2012) • Dekens et al. (2017) • etc.
![Page 66: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/66.jpg)
Multivariate Empirical Copula – Bias correction (EC-BC)
• Vrac and Friederichs (2015, J. Clim.). Multivariate – Inter-variable, spatial and temporal – bias correction
![Page 67: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/67.jpg)
• Vrac and Friederichs (2015, J. Clim.). Multivariate – Inter-variable, spatial and temporal – bias correction
1. Univariate (classical, 1d) bias correction (BC) w.r.t. reference (training) data
Multivariate Empirical Copula – Bias correction (EC-BC)
![Page 68: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/68.jpg)
• Vrac and Friederichs (2015, J. Clim.). Multivariate – Inter-variable, spatial and temporal – bias correction
1. Univariate (classical, 1d) bias correction (BC) w.r.t. reference (training) data
2. Temporal shuffling of the elements of each 1d-corrected time series s.t., rank(shuffled corrected data) = rank( training data)
Multivariate Empirical Copula – Bias correction (EC-BC)
![Page 69: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/69.jpg)
Training 1d-BC data
• Vrac and Friederichs (2015, J. Clim.). Multivariate – Inter-variable, spatial and temporal – bias correction
1. Univariate (classical, 1d) bias correction (BC) w.r.t. reference (training) data
2. Temporal shuffling of the elements of each 1d-corrected time series s.t., rank(shuffled corrected data) = rank( training data)
Multivariate Empirical Copula – Bias correction (EC-BC)
![Page 70: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/70.jpg)
Training 1d-BC data
• Vrac and Friederichs (2015, J. Clim.). Multivariate – Inter-variable, spatial and temporal – bias correction
1. Univariate (classical, 1d) bias correction (BC) w.r.t. reference (training) data
2. Temporal shuffling of the elements of each 1d-corrected time series s.t., rank(shuffled corrected data) = rank( training data)
Multivariate Empirical Copula – Bias correction (EC-BC)
![Page 71: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/71.jpg)
Training 1d-BC data Shuffled BC data
• Vrac and Friederichs (2015, J. Clim.). Multivariate – Inter-variable, spatial and temporal – bias correction
1. Univariate (classical, 1d) bias correction (BC) w.r.t. reference (training) data
2. Temporal shuffling of the elements of each 1d-corrected time series s.t., rank(shuffled corrected data) = rank( training data)
Multivariate Empirical Copula – Bias correction (EC-BC)
![Page 72: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/72.jpg)
Training 1d-BC data Shuffled BC data
• Vrac and Friederichs (2015, J. Clim.). Multivariate – Inter-variable, spatial and temporal – bias correction
1. Univariate (classical, 1d) bias correction (BC) w.r.t. reference (training) data
2. Temporal shuffling of the elements of each 1d-corrected time series s.t., rank(shuffled corrected data) = rank( training data)
Multivariate Empirical Copula – Bias correction (EC-BC)
![Page 73: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/73.jpg)
• Vrac and Friederichs (2015, J. Clim.). Multivariate – Inter-variable, spatial and temporal – bias correction
1. Univariate (classical, 1d) bias correction (BC) w.r.t. reference (training) data
2. Temporal shuffling of the elements of each 1d-corrected time series s.t., rank(shuffled corrected data) = rank( training data)
Ø Multivariate – temporal, inter-var. & spatial – properties are reconstructed !!
Training 1d-BC data Shuffled BC data
Multivariate Empirical Copula – Bias correction (EC-BC)
![Page 74: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/74.jpg)
• Vrac and Friederichs (2015, J. Clim.). Multivariate – Inter-variable, spatial and temporal – bias correction
1. Univariate (classical, 1d) bias correction (BC) w.r.t. reference (training) data
2. Temporal shuffling of the elements of each 1d-corrected time series s.t., rank(shuffled corrected data) = rank( training data)
Ø Multivariate – temporal, inter-var. & spatial – properties are reconstructed !!
Ø Applicable to any of your favourite 1d-BC or DS method !!
Training 1d-BC data Shuffled BC data
Multivariate Empirical Copula – Bias correction (EC-BC)
![Page 75: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/75.jpg)
Data & Calibration/Projection
• Calibration = 1980-1994
• Projection = 1995-2009
One “summer” season (15 Apr. – 14 Oct.) One “winter” season (15 Oct. – 14 Apr.)
The “winter” results are shown next (but equivalent results for summer)
• Reference data = SAFRAN dataset (S-E of France), 8km x 8km
• Model data to be corrected = ERA-Interim reanalyses, 0.75°x0.75°
1d-BC vs. Cond.BC(2d) vs. EC-BC(3012d) vs. ERA-I vs. Shuffle(ERA-I) vs.
SAFRAN(proj)
![Page 76: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/76.jpg)
EC-BC : spatial evaluation (illustration on T2)
EC-BC
SAFRAN ERA-I
Cond. BC
Shuffle(ERA-I)
1d-BC
![Page 77: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/77.jpg)
SAFRAN
EC-BC : spatial evaluation (illustration on T2)
ERA-I
Cond. BC
Shuffle(ERA-I)
1d-BC
EC-BC
![Page 78: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/78.jpg)
SAFRAN
EC-BC : spatial evaluation (illustration on T2)
SAFRAN
EC-BC
1d-BC
EC-BC
ERA-I
Cond. BC
Shuffle(ERA-I)
1d-BC
![Page 79: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/79.jpg)
EC-BC : Inter-variable (PR-T) evaluation
SAFRAN ERA-I
Cond. BC (T2 given BC PR)
Shuffle (ERA-I)
EC-BC
Cond. BC PR given BC T2
CDF-t (separately)
![Page 80: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/80.jpg)
EC-BC : Inter-variable (PR-T) evaluation
SAFRAN ERA-I
Cond. BC (T2 given BC PR)
Shuffle (ERA-I)
EC-BC
Cond. BC PR given BC T2
CDF-t (separately)
![Page 81: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/81.jpg)
EC-BC : Inter-variable (PR-T) evaluation
SAFRAN ERA-I
Cond. BC (T2 given BC PR)
Shuffle (ERA-I)
EC-BC
Cond. BC PR given BC T2
CDF-t (separately)
![Page 82: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/82.jpg)
EC-BC : Inter-variable (PR-T) evaluation
SAFRAN ERA-I
Cond. BC (T2 given BC PR)
Shuffle (ERA-I)
EC-BC
Cond. BC PR given BC T2
CDF-t (separately)
Good !?
![Page 83: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/83.jpg)
EC-BC : Temporal evaluation (illustration on PR)
SAFRAN
EC-BC
ERA-I
Cond. BC
Shuffle(ERA-I)
1d-BC
![Page 84: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/84.jpg)
EC-BC : Temporal evaluation (illustration on PR)
SAFRAN
EC-BC
ERA-I
1d-BC
SAFRAN
EC-BC
ERA-I
Cond. BC
Shuffle(ERA-I)
1d-BC
![Page 85: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/85.jpg)
EC-BC : Temporal evaluation (illustration on PR)
SAFRAN
EC-BC
ERA-I
Cond. BC (T2 given BC PR)
Shuffle on ERA-I
SAFRAN
EC-BC
ERA-I
Cond. BC
Shuffle(ERA-I)
1d-BC 1d-BC
![Page 86: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/86.jpg)
More evaluations in Vrac and Friederichs (2015, JClim):
Vrac, M., and Friederichs, P. (2015). Multivariate – Intervariable, Spatial, and Temporal—Bias Correction.
J. Climate, 28, 218–237. doi: http://dx.doi.org/10.1175/JCLI-D-14-00059.1
![Page 87: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/87.jpg)
(Partial) Conclusions on EC-BC
• 1d-BC methods (CDF-t/QM) not able to produce multi-dimensional properties
• Cond’l technique only good for inter-var. properties (in present config.)
• EC-BC good for both inter-var. & spatial (& temporal) correlations
ü Preliminary 1d-BC before shuffling is important (“1d-BC + shuffling” => best results)
![Page 88: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/88.jpg)
(Partial) Conclusions on EC-BC
• 1d-BC methods (CDF-t/QM) not able to produce multi-dimensional properties
• Cond’l technique only good for inter-var. properties (in present config.)
• EC-BC good for both inter-var. & spatial (& temporal) correlations
ü Preliminary 1d-BC before shuffling is important (“1d-BC + shuffling” => best results)
• Flexibility: EC-BC is applicable with any 1d-BC or 1d-SDM !!!
![Page 89: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/89.jpg)
(Partial) Conclusions on EC-BC
• 1d-BC methods (CDF-t/QM) not able to produce multi-dimensional properties
• Cond’l technique only good for inter-var. properties (in present config.)
• EC-BC good for both inter-var. & spatial (& temporal) correlations
ü Preliminary 1d-BC before shuffling is important (“1d-BC + shuffling” => best results)
• Flexibility: EC-BC is applicable with any 1d-BC or 1d-SDM !!!
• Easiness of coding + flexibility + fast application + quality
= EC-BC is a good candidate for many (multivariate) BC applications
• Package R (ECBC) already available (upon request to me)
![Page 90: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/90.jpg)
Perspectives on Multivariate BC
![Page 91: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/91.jpg)
• GCMs/RCMs instead of reanalyses • Ensemble (i.e., multiple models) approaches
ü Multi-1d-BCs / Multi-MBCs/ Multi-RCMs
• Comparisons with multivariate/spatial SDMs • Consequences in terms of “impacts” (e.g., hydro)
Perspectives on Multivariate BC A
pplic
atio
n
![Page 92: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/92.jpg)
• GCMs/RCMs instead of reanalyses • Ensemble (i.e., multiple models) approaches
ü Multi-1d-BCs / Multi-MBCs/ Multi-RCMs
• Comparisons with multivariate/spatial SDMs • Consequences in terms of “impacts” (e.g., hydro)
App
licat
ion
François et al.(in prep)
Perspectives on Multivariate BC
![Page 93: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/93.jpg)
• GCMs/RCMs instead of reanalyses • Ensemble (i.e., multiple models) approaches
ü Multi-1d-BCs / Multi-MBCs/ Multi-RCMs
• Comparisons with multivariate/spatial SDMs • Consequences in terms of “impacts” (e.g., hydro)
App
licat
ion
Prac
tice
& th
eory
François et al.(in prep)
Perspectives on Multivariate BC
![Page 94: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/94.jpg)
• GCMs/RCMs instead of reanalyses • Ensemble (i.e., multiple models) approaches
ü Multi-1d-BCs / Multi-MBCs/ Multi-RCMs
• Comparisons with multivariate/spatial SDMs • Consequences in terms of “impacts” (e.g., hydro)
• So far, the spatial dep. structure does not change (=obs) ⇒ Conservative approach: “safe” projections ⇒ Should we make it evolve? (How much do you trust GCM?)
App
licat
ion
Prac
tice
& th
eory
François et al.(in prep)
Perspectives on Multivariate BC
![Page 95: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/95.jpg)
• GCMs/RCMs instead of reanalyses • Ensemble (i.e., multiple models) approaches
ü Multi-1d-BCs / Multi-MBCs/ Multi-RCMs
• Comparisons with multivariate/spatial SDMs • Consequences in terms of “impacts” (e.g., hydro)
• So far, the spatial dep. structure does not change (=obs) ⇒ Conservative approach: “safe” projections ⇒ Should we make it evolve? (How much do you trust GCM?)
• More flexibility (e.g., only inter-var & spatial BC, no temporal BC)
App
licat
ion
Prac
tice
& th
eory
François et al.(in prep)
Perspectives on Multivariate BC
![Page 96: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/96.jpg)
• GCMs/RCMs instead of reanalyses • Ensemble (i.e., multiple models) approaches
ü Multi-1d-BCs / Multi-MBCs/ Multi-RCMs
• Comparisons with multivariate/spatial SDMs • Consequences in terms of “impacts” (e.g., hydro)
• So far, the spatial dep. structure does not change (=obs) ⇒ Conservative approach: “safe” projections ⇒ Should we make it evolve? (How much do you trust GCM?)
• More flexibility (e.g., only inter-var & spatial BC, no temporal BC) • From deterministic to stochastic EC-BC
App
licat
ion
Prac
tice
& th
eory
François et al.(in prep)
Perspectives on Multivariate BC
![Page 97: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/97.jpg)
Perspectives on Multivariate BC
• GCMs/RCMs instead of reanalyses • Ensemble (i.e., multiple models) approaches
ü Multi-1d-BCs / Multi-MBCs/ Multi-RCMs
• Comparisons with multivariate/spatial SDMs • Consequences in terms of “impacts” (e.g., hydro)
• So far, the spatial dep. structure does not change (=obs) ⇒ Conservative approach: “safe” projections ⇒ Should we make it evolve? (How much do you trust GCM?)
• More flexibility (e.g., only inter-var & spatial BC, no temporal BC) • From deterministic to stochastic EC-BC
⇒ Variants (R2B2) of EC-BC (Vrac, in prep.)
⇒ Different approach(es): “Optimal” BC (Robin et al., in prep.)
App
licat
ion
Prac
tice
& th
eory
François et al.(in prep)
![Page 98: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/98.jpg)
Thank you
Sophie Bastin (LATMOS)
Julie Carreau (HSM)
Denis Ruelland (HSM)
Pradeebane Vaittinada Ayar (LSCE)
Benjamin Grouillet (HSM)
& many thanks to the StaRMIP team
![Page 99: Statistical downscaling & bias correction of climate ... · • GCM predictors (in SDMs) are not perfect (!) Ø Influences of Bias correcting climate predictors for SDMs (Vrac & Vaittinada](https://reader035.vdocuments.us/reader035/viewer/2022070108/60280069e59d60492b17aa33/html5/thumbnails/99.jpg)