Offline data assimila-on using pollen data, LOVECLIM and BIOME4 - - PowerPoint PPT Presentation

Offline data assimila-on using pollen data, LOVECLIM and BIOME4 models

Joël Guiot, Raj Rani (CEREGE, Aix-en- Provence) With the help of H. Goosse, F. Klein (ELI, LLN), N. Barrier (Labex OT-Med)

Objec-ves and method

• Using pollen data, climate model and vegeta-on

model

• Propose a new reconstruc-on of climate for the

Holocene in the Mediterranean region

• Taking into account the « reality » of the data

and the physical coherency of the models

• by off-line paleodata assimila-on based on fast

climate + vegeta-on models

• Proxy model used in a forward mode

Data and method

• Pollen data from European Pollen

Data (286 sites), pollen types converted into Plant func-onal types (PFT)

• CRU climatology resolu-on 10’
• Intermediate complexity climate model: LOVECLIM 1.3

(Renssen et al 2005)

– Atmosphere: ECBILT (Opsteegh et al, 1998) – Ocean sea ice: CLIO (Goosse and Fichefet, 1999) – Vegeta-on: VECODE (Brovkin et al, 2002)

Diagram of the method

Assimila-on (par-cle filter) Inversion (MCMC) Proxy model BIOME4 Canonical analysis Uninforma-ve prior pollen pollen MCMC

The forcings

Time-series of forcings: (a) total solar irradiance (w/m2) Steinhilber et al. (2012) (b) CO2 (ppmv) (c) CH4 (ppbv) (d) N2O (ppbv), used in the Holocene experiment from ice cores for the Holocene (Indermuhle et al 1999) Completed by Muscheler et al (2007) for the last 1200 yrs Volcanism not considered a b c d

LOVECLIM simula-ons

• The model is applied to the whole Holocene
• Outputs: -me sclices (by step of 1 year) of monthly

temperature, precipita-on, cloudiness

• We calculate mean maps by steps of 10 years

Sta-s-cal downscaling

• Outputs of LOVECLIM (low resolu-on, >300km) are

interpolated to the CRU 10’ resolu-on grid

• Generalized Addi-ve Model (Vrac et al, 2007; Levavasseur et

al 2011)

• Physical predictors are LOVECLIM outputs
• Geographical predictors are coordinates of CRU data
• Predictands are CRU monthly variables
• Applied to the 10-yrs LOVECLIM outputs

BIOME4

• BIOME4, equilibrium vegeta-on model

(Kaplan et al, 2003)

• Carbon and water flux cycles
• Applied to the downscaled 10 years simulated

and downscaled climate fields

• inputs: monthly temperature, precipita-on,

cloudiness, CO2, soil texture

• Outputs: net primary produc-vity of 13 PFTs

Canonical analysis

• The PFT NPP of BIOME4 are not exactly the

pollen PFTs

• Canonical analysis between the 13 model

variables and the 9 pollen PFT

• Transforma-on of both sets at the same scale:

two independent canonical components

• Es-ma-on of these CC on the basis of 16746

pollen samples from the Med region

• Canonical correla-on: CC1: 0.80, CC2: 0.74
• Pollen data are averaged by 100 years steps

Data assimila-on

• par-cle filter technique (Goosse et al, 2010)
• for each 100 yr pollen step,
• Poten-al par-cles (or analogues) are 10 yrs

downscaled climate outputs

• Posterior density func-on (PDF) is a weighted mean of

the 10-yrs climate par-cles

• Weights = distance between CC1(model) and CC1

(pollen) taking into account the respec-ve errors

• These CC1 are given with error bars
• This is done by an itera-ve process of resampling

where the low weight par-cles are excluded

Time-Series

Loveclim alone Downcaled Assimilated Correla-ons

• L/D = 0.53
• L/A = 0.12
• D/A = -0.02

Correla-ons

• L/D = 0.99
• L/A = -0.22
• D/A = -0.21

Comparison assimila-on with previous construc-on (BIOME4 inversion)

Case 1: a rela-vely good agreement between model and data: 6200 yr BP

Reconstruc-on (inversion) compared with Loveclim and assimila-on Results constrained by model

Case 2: a low agreement between model and data: 900 yr BP

Reconstruc-on (inversion) compared with Loveclim and assimila-on Results constrained by data

Conclusions

• First results, but encouraging
• Difference of scale between GCM and ecosystem:

importance of downscaling the GCM outputs

• Necessity to beper cope with the uncertain-es
• BIOME4 is a simple model, but appropriate for

paleo studies (e.g. CO2 effect)

• Offline assimila-on: possibility to extend the data

a posteriori