bias correction of climate model data – the golden ... · bias correction of climate model data...
TRANSCRIPT
Bias correction of climate model data – the golden solution for impact models or cursed
black magic?
Stefan Hagemann, Jan O. HaerterMax Planck Institute for Meteorology, Hamburg
Claudio PianiICTP, Trieste
GCI, Bonn, Dec. 2010, Stefan Hagemann
Impact modelling chain
Climate model input: Precipitation, 2m Temperature, other ...
Interpolation to HM resolution
Bias correction required
Hydrology Model (HM)
GCI, Bonn, Dec. 2010, Stefan Hagemann
Quality of observational datasets limits the quality of the bias correction.
It is assumed that the bias behaviour of the model does not change with time, i.e. the transfer function is time-independent and, thus, applicable in the future.
Limitation: Temporal errors of major circulation systems can not be corrected, e.g. onset of monsoon.
Bias correction – main assumptions
GCI, Bonn, Dec. 2010, Stefan Hagemann
Delta change approach (e.g. Hay et al. 2000)
Multiple linear regression (e.g. Hay and Clark 2003; von Storch 1999)Analogue methods (von Storch and Navarra 1999; Moron et al. 2008)Local intensity scaling (Widmann et al. 2003; Schmidli et al. 2006)Quantile mapping (Panofsky and Brier 1968)
Themeßl et al. (2010): Quantile mapping performs best for RCM precipitation over the Alps
Bias correction methods
GCI, Bonn, Dec. 2010, Stefan Hagemann
modeled
Statistical bias correction: Piani et al. (2010), J. Hydrol.
original
corrected
GCI, Bonn, Dec. 2010, Stefan Hagemann
Summary of methodology for precipitation and temperature
In theory: bias correction adjusts all moments of distribution function for each day.In practice: For most regions, a 2-parameter fit to the transform function is used as a good approximation.Specific regions use 3 or 4 parameter transfer functions.Using larger number of parameters may not be adequate as correction needs to be time-independent on climatological time-scales (>10 years).Similar procedure was followed for temperature correction (always 2-parameter fit: linear transfer function).Monthly transfer functions are used, with smooth transitions for temperature.
GCI, Bonn, Dec. 2010, Stefan Hagemann
Observations: Daily WATCH forcing data (Weedon et al. 2010)ERA40 data interpolated to 0.5° with elevation correction to CRU2m temperature: Correction with monthly means CRU data. Precipitation: Correction with monthly GPCC Vs.4 data, and a gauge-undercatch correction according to Jennifer Adam.
Bias correction factors are derived for period 1960-99 and applied to 1960-2100
Precipitation, (Snowfall fraction taken from GCM)Temperature: Tmean, Tmin, Tmax
Diurnal Range: ΔT = Tmax-TminSkewness: σ = (Tmean-Tmin) / ΔT
Data
GCI, Bonn, Dec. 2010, Stefan Hagemann
ECHAM5
IPSL
CNRM
WFD
Bias CorrectedOriginal GCM
Standard deviation of precipitation (1960-99)
GCI, Bonn, Dec. 2010, Stefan Hagemann
GCI, Bonn, Dec. 2010, Stefan Hagemann
Climate Change: Amazon
GCM Precipitation change GCM Temperature change
GCI, Bonn, Dec. 2010, Stefan Hagemann
...
Slope of transfer function ≠ 1 Climate change signal is changed
GCI, Bonn, Dec. 2010, Stefan Hagemann
Annual mean slope of monthly temperature transfer functions
ECHAM5
IPSL
CNRM
GCI, Bonn, Dec. 2010, Stefan Hagemann
Bias correction effectively improves both the mean and the variance of the precipitation and temperature fields in all but a few regions of the globe.Bias correction has an impact on the climate change signal for specific locations and months
Low precipitation amounts (or temperatures) are differently corrected as high amounts (due to different model biases slope of transfer function significantly differs from 1) If distribution between low and high amounts changes in a future climate, bias correction can lead to changes in the analysed signal.
For some regions, the impact of the bias correction on the climate change signal may be larger than the signal itself, thereby uncovering another level of uncertainty that is comparable in magnitude to the uncertainty related to the choice of the GCM or hydrology model. Note that the bias correction has only uncovered but not necessarily caused this extra level of uncertainty within the GCM – hydrology model (or any other impact model) modelling chain.
Summary
GCI, Bonn, Dec. 2010, Stefan Hagemann
Difficult to judge whether the impact of the bias correction on the climate change signal leads to a more realistic signal or not. Giorgi and Coppola (2010) analysed 18 AR4 GCM projections
Projected regional precipitation changes are significantly correlated with the respective regional biases for about 30% of the seasonal/regional cases investigated. For temperature, only a negligible effect of the regional bias on the projected change was noticed.
This suggests at least for precipitation that an impact of the bias correction on the climate change signal may be reasonable. How to handle and possibly reduce the uncovered uncertainty will be subject to future investigations whose outcomes have to be communicated to the impact research communities.
Summary
GCI, Bonn, Dec. 2010, Stefan Hagemann
Precipitation and temperature are correctly independently. Other GCM variables are not corrected.
Bias correction uses mathematical/statistical methods--> No black magicIt is improving but also impacting climate model results, so that it should also be taken with care.
Summary
GCI, Bonn, Dec. 2010, Stefan Hagemann
Thank you for your attention!
Hagemann, S., C. Chen, J.O. Härter, J. Heinke, D. Gerten and C. Piani: Impact of a statistical bias correction on the projected hydrological changes obtained from three GCMs and two hydrology modelsJ. Hydrometeor., submitted
GCI, Bonn, Dec. 2010, Stefan Hagemann
Nile
A = 6 largest Arctic Rivers = Mackenzie, N Dvina, Ob, Yenisey, Lena, Kolyma
Amazon
Congo
Mississippi Yangtze Kiang
Amur
Ganges/Brahmaputra
Danube
Baltic Sea
A AA
A A A
Parana Murray
Large catchments are considered
GCI, Bonn, Dec. 2010, Stefan Hagemann
Spatial distribution of the choice of transfer function typeMostly linear TF used (yellow) additive TF is only option in very dry regions (orange)
Transfer functions
GCI, Bonn, Dec. 2010, Stefan Hagemann
GCI, Bonn, Dec. 2010, Stefan Hagemann
GCM Precipitation change: Ganges/Brahmaputra
GCI, Bonn, Dec. 2010, Stefan Hagemann
Precipitation correction of CNRM: Ganges/Brahmaputra
GCI, Bonn, Dec. 2010, Stefan Hagemann
Regions of an additive CNRM precipitation correction for April
Mapping example
GCI, Bonn, Dec. 2010, Stefan Hagemann
Global modelling chain in WATCH
Climate model input from 3 GCMs: Precipitation, 2m Temperature, other ...
Interpolation to 0.5 degree
Statistical bias correction of P and T fields
Global Hydrology Model
GCI, Bonn, Dec. 2010, Stefan Hagemann
Climate change: Danube
GCM Precipitation change GCM Temperature change
GCI, Bonn, Dec. 2010, Stefan Hagemann
Global Hydrology Models
MPI-HM LPJmLDaily Input P, T P, T, LWn, SW Potential Evap. Thornthwaite Priestley-TaylorRunoff/Infiltration Saturation excess / Saturation excess
Beta functionSnowmelt Degree day Degree day
Refs. for MPI-HM: Hagemann and Dümenil Gates (2003), Hagemann and Dümenil (1998)
Refs. for LPJmL: Bondeau et al. (2007), Rost et al. (2008), Fader et al. (2010)