Land surface contribution to climate variability and predictability

Hervé Douville

Météo-France/CNRM [email protected]

Acknowledgements: B. Decharme, R. Alkama and Y. Peings

WCRP Seasonal Prediction Workshop, Exeter, 1-3 December 2010

2 Outlines Background and motivations 1. Land surface data and statistical studies • • • Global land surface products Data intercomparison and model evaluation Statistical evidence of predictable land surface impacts 2. Numerical sensitivity experiments • • • Pioneering studies Numerical evidence of local land surface impacts Numerical evidence of remote land surface impacts Conclusions, prospects and issues

Seasonal prediction: A question of remote control ?

The forecast The AOGCM The stratospheric component The anthropogenic radiative component The land surface component A « slave » component ?

3

« Need to improve the representation of climate system interactions and their potential to improve forecast quality.

» (WCRP position paper, Barcelona 2007)

GLACE: Global Land-Atmosphere Coupling Experiment (a GEWEX & CLIVAR initiative)

?

4 GLACE-1 multi-model land-atmosphere coupling strength based on the reproductibility of 5-day precipitation (Koster et al. 2006) . Not sufficient to evaluate the impact of land state initialization on seasonal forecast skill => GLACE-2

5 Relevance of land-atmosphere coupling for climate

predictability

: At least 3 conditions 1.

Land surface anomalies must have sizeable (i.e.

potential

predictability) and realistic (i.e.

effective

predictability) impacts on atmospheric variability 2.

Land surface anomalies must be the selected timescale (using dynamical and/or statistical tools)

predictable

at

3.

Real-time

global land surface analyses must be available for initializing the relevant land surface variables (soil moisture, snow mass, …) NB: focus on monthly to seasonal timescale only.

6 (lack of) Land surface data And statistical studies  Global (satellite) land surface observations o o Snow: visible (since 1967), passive micro. (SMMR since 1978, …) Soil moisture: passive & active micro. (AMSR since 2002, ASCAT, …) o Total water storage variations: gravimetry (GRACE since 2002)  Off-line land surface model simulations o o o GSWP-2 (1986-1995): 13 models driven with ISLSCP2 forcing data GLDAS (1979-present): 4 models driven with bias-corrected reanalyses or NOAA/GDAS real-time analyses (since 2000) VIC (Sheffield and Wood 2008) or ISBA (Alkama et al. 2010) driven with 1950-2006 Princeton Univ. (Sheffield et al. 2006)  On-line LDAS systems o o Soil moisture analysis based on the assimilation of screen-level temperature and humidity (e.g. Météo-France, ECMWF, Met Office, …) Assimilation of NESDIS snow extent (e.g. ECMWF since 2004) o Assimilation of ASCAT soil moisture (e.g. Met Office since July 2010)

7 Off-line land surface simulations ISLSCP-2 (1986-1995), Princeton Univ. (1950-2006), … 3-hourly atmospheric forcing Fixed or monthly physiography

LSM

Runoff

RRM

Discharge Evaporation Satellite Data In Situ Observ.

Soil moisture & snow mass climatology

AGCM

T2m et P

8 Land surface data intercomparison Ex: Central Europe vs 1989-1995 climatology • ISBA driven by Princeton University atm. forcings (1950 2006) • ERA-Interim (1989-2010) • ERA40 (1958-2001) • GSWP multi model driven by ISLSCP2 atm. forcings (1986-1995)

Land surface model evaluation ISBA-TRIP vs GRACE and GRDC data Monthly water storage variation (kg/m²/day) anomalies and mean annual cycle ISBA = soil moisture + snow + river Monthly river discharge (kg/m²/day) anomalies and mean annual cycle 7 9 Alkama et al., J. Hydromet., 2010

10 Statistical evidence of land surface contribution to predictability  North American summer temperature (e.g. Alfaro et al. 2006) and precipitation (e.g. Quiring and Kluver 2009)  Sahelian summer monsoon precipitation (e.g. Philippon and Fontaine 2002, Douville et al. 2007)  Indian summer monsoon precipitation (e.g. Blanford 1884, Fasullo 2004, Peings and Douville 2009)  Winter North Atlantic Oscillation (e.g. Cohen and Entekhabi 1999, Hardiman et al. 2008, Cohen et al. 2010)

11 Statistical evidence: North America T2m & P Maps of JJA Tmax prediction skill (cross-validation over 1950-2001 (soil moisture proxy) predictors.

Alfaro et al. 2006 ) using May Pacific SST and/or PDSI Northern Great Plains heavy & light AM snowfall composites ( 1929-1999 ) with interquartile range .

Quiring and Kluver 2009

T2m (

°

C) Cum. P (mm)

Statistical evidence: West African summer monsoon P • Hypothesis: summer monsoon rainfall over the Sahel through a soil moisture memory effect 2nd rainy season over the Guinean Coast affects subsequent (Landsea et al. 1993, Philippon and Fontaine 2002) • But: Stochastic artefact mediated through tropical SST and partly due to multi-decadal variability ? (Douville et al. 2007) 12

Statistical evidence: Indian summer monsoon P • Hypothesis: Winter and spring Eurasian snow cover affects subsequent summer monsoon rainfall over India (Blanford 1884, Fasullo 2004) • But: Such a statistical relationship is neither robust nor stationary in the instrumental record and is not captured by CMIP3 historical simulations (Peings and Douville 2009) 13

Statistical evidence: Wintertime N.H. extratropical variability SnowCast Observations JFM 2010 forecasted vs observed temperature anomalies (Cohen et al. 2010) A negative AO/NAO winter preceded by above normal Siberian snow cover • Hypothesis: Fall (i.e. October) snow cover over Siberia affects subsequent winter NAO (Cohen and Entekhabi 1999) • But: Not found in CMIP3 models observed relationship is robust and was verified in winter 2009-2010 (Cohen et al. 2010) (Hardiman et al. 2008) though the 14

15 Further evidence based on numerical sensitivity experiments  Pionneering studies: Land vs SST impact on precipitation variability (e.g. Koster et al. 2000) , dynamical vs non-dynamical feedback (e.g. Douville et al. 2001)  GLACE-2 and related studies (e.g. Douville 2009, Koster et al. 2010, Peings et al. 2010)  Remote impacts of Eurasian snow cover (e.g. Fletcher et al. 2009, Peings et al. in preparation)

Impact of Land vs SST variability on annual mean precipitation (Koster et al. 2000) 16 Control experiment ALO A: Atmosphere only L: Interactive Land Hydrology O: Observed instead of climatol. monthly mean SST  2 ALO /  2 AO

Anom.

Dynamical (P-E) versus non-dynamical (E) soil moisture feedbacks (Douville et al. 2001) E P-E P Sahel dry wet dry wet dry wet South Asia 17 dry wet dry wet dry wet

18 SST vs land surface impacts on monthly T2m predictability over land (Douville 2009) Zonal mean annual cycle of: Stdev Pot. Pred. (ANOVA) Skill (ACC) 75 ° N Control No nudging Obs. SST 55 ° S 75 ° N Nudging Obs. SST 55 ° S 75 ° N Nudging Clim. SST 55 ° S

16-30 days 31-45 days 46-60 days

GLACE-2 coordinated experiments “Consensus” skill due to land initialization

temperature precipitation

 2-months hindcasts initialized on 1 st & 15 th June, July and August => 6 hindcasts x 10 years (1986-1995) x 10 members = 600 runs.

 13 models (“weaker” models are averaged in with “stronger” ones).

 Conditional skill show stronger increase.

19 Koster et al., GRL, 2010 19

3 ensemble experiments: Impact of snow boundary / initial conditions on springtime (MAM) T2m (Peings et al. 2010) Total Stdev Pot. Predictability Skill Control (CTL) Interactive snow cover SBC – CTL Impact of snow relaxation 20 SIC – CTL Impact of snow initialization

21 Remote impact of Siberian snow cover on DJF NAO (Fletcher et al. 2009) a) D SWnet (d1-d15) b) D MSLP (d24-d50) 2 pairs of 100 member ensemble experiments: High minus Low fall snow cover over Siberia A snow-NAO relationship through a stratospheric pathway

Remote impact of Siberian snow cover on DJF NAO (Peings et al. 2011) MSLP (hPa) Zonal mean Z (m) 2 pairs of 50-member ensemble experiments: DSS - CTL

Deep Snow over Siberia

22 DSS* - CTL*

Improved polar vortex climatology through equatorial stratospheric nudging

23 CONCLUSIONS  Growing statistical and numerical evidence of both local and remote impacts of land surface initial conditions on climate predictability (though some of these studies are questionnable);  Such impacts are highly model-dependent , variable across regions and seasons, and sensitive to the magnitude of the land surface anomalies;  Long-range predictability of the land surface hydrology seems limited (mainly by the low predictability of precipitation) but needs further evaluation (i.e. new observations and data assimilation systems);  Land surface impacts do not amount to simple changes in the surface energy budget, but also involve large-scale dynamical and cloud feedbacks ;  Land surface contribution to climate predictability should not be neglected given the weak SST impact on extratropical predictability.

24 PROSPECTS (OPEN FOR DISCUSSION)  Observations: SMOS (L-band, 2010) & SMAP (Soil Moisture Active and Passive, 2015) for upper soil moisture, improved use of passive microwave data for snow (until ESA’s CoReH20 mission), GRACE for total water storage variations, SWOT (Surface Water and Ocean Topography) , …  Land Surface Models & Data Assimilation Systems: increased vertical discretization, simulation of water bodies including floodplains, improved representation of snow under canopy (e.g. SnowMIP), multi-spectral surface albedo and related data assimilation(MSG, MODIS), off-line model inter-comparison without (GSWP-3?) and with (PILDAS?) data assimilation, global & multi decadal (at least since 1989) surface reanalysis, …  Sensitivity experiments: SCM studies, follow-on of GLACE-2 looking at soil moisture but also snow water equivalent and possibly surface albedo, GLACE-type versus state-of-the-art (rather than random) initialization, coupled vs AMIP-type experiments, process-oriented case studies, statistical adaptation of dynamical forecasts using land surface variables, …

(CONTROVERSIAL) ISSUES  What about vegetation ?

Difference in statistical significance of temporal ACCs between two sets of hindcasts of JJA T2m using observed vs climatological vegetation ( red / blue means increased / decreased significance) (Gao et al. 2008)  Statistical benchmarks 25 ACC and RMSS differences between sCast and DEMETER hindcasts of DJF surface temperature (72/73 to 92/93) ( red / blue means sCast has greater / lower skill) (Cohen and Fletcher 2007)

26 (CONTROVERSIAL) ISSUES  Towards decadal predictions ?

Verification of the first

genuine

dynamical decadal prediction by Keenlyside et al. 2008 for global mean temperature (from http://www.realclimate.org) A land surface contribution would be welcome but is unlikely…  Seamless is

not

questionless…

Bekele ESM Bolt NWP

End