A review of carbon cycle work with Hadley Centre models'

1
A review of carbon cycle work with Hadley Centre
models.

• P.L. Vidale,
• M. Roberts, M.E. Demory, J. Donners, A. Clayton,
  G. Robinson
• With big thanks to
• C. Jones, O. Boucher, Hadley Centre
• NCAS Centre for Global Atmospheric Modeling
  (CGAM),
• Univ. of Reading, UK

2
Carbon and Water on land

• Plants eat CO2 for a living
• They open their stomata to let CO2 in
• Water gets out as an (unfortunate?) consequence
• For every CO2 molecule fixed, about 400 H2O
  molecules are lost.

CO2 Farquhar, Collatz
H2O Ball-Berry
The engine room of a 3rd generation land surface
model
3
Biophysics in climate models like it or not, it
is still our business !

• Heat, water, and carbon fluxes are coupled by
  physiology, so that local climate (e.g. Bowen
  ratio) is strongly connected to the carbon cycle,
  via the water cycle
• 2nd generation models mimic the CO2-H2O
  relationship through empirical functions, which
  have been derived for current climate

4
Hadley Centre Coupled Climate-Carbon Cycle
Models HadCM3LC and HadOCC
MOSES2 TRIFFID
NPZD
Source C. Jones, Hadley Centre
5
Scenario Projection 2000-2100

• Climate change is neglected (blue line)
• standard IS92a scenario as used in GCMs (black
  line).
• When climate-carbon cycle feedbacks are included
  (red line) we see much higher rates of CO2
  increase and climate change.
• By 2100 the fully coupled model produces 970 ppmv
  and a global warming of 5.5K (c.f. 700 ppmv and
  4K without climate-carbon cycle feedbacks). Gain
  factor0.27.
• Extra C comes from terrestrial biosphere.
  Increased soil respiration (T) greater than
  increased growth (CO2)

from Cox et al., 2000
6
Change to carbon stored in vegetation (18602100)
Change to carbon stored in soils (18602100)
7
Climate-driven Amazon dieback
1850
2000
2100
Broadleaf tree fraction
Caused mainly by precipitation deficit. Is this a
local effect or something like what we have seen
in 2005 ?
from Cox et al., 2000
8
Assumptions in HC models

• CO2 fertilisation is possible and will affect a
  wide range of plant types, but all equally within
  C3/C4 categories
• We can model a natural ecosystem independent of
  anthropogenic landscape changes
• What are data telling us ?
• Individual C3 and C4 plants show different
  behavior
• Age of plants really matters
• There are signs of acclimation for some plants at
  high CO2 values
• Trade-off between growth and water savings (e.g.
  example in IR picture).

9
European summer ina Hadley Centre A2 scenario
with uncoupled C

• Shift in mean and increase in interannual
  variability
• JJA 2003 looks less special
• Location of change in mean and location of change
  in variability are not the same
• How robust are these model results ?
• What are the active mechanisms ?
• What are the missing ones ?

from Schär et al., 2004
10
The soil moisture/precipitation feedback a
comparison of 11 IPCC AR4 models (C uncoupled)

• are drier models not capable of correctly
  sustaining their hydrological cycle in future
  climates ?
• are we at danger of hitting some thresholds in
  the models, e.g. environmental conditions, in
  which current vegetation would shut-down/die by
  July each year ?

What does this mean for future land and water
resources management ?
from Vidale et al., 2006
11
Climate matters historical simulation 1860-2000

• The simulated present-day CO2 is too high by
  about 15-20 ppmv in a run with CO2 emissions plus
  prescribed GHGs.
• This error is associated with an over-estimate of
  climate warming since 1960.
• A run also including other climate forcings
  (solar, volcanic and manmade aerosols), and
  revised net land use emissions (70 IS92a as
  suggested in IPCC 95), shows a good fit to
  observations.
• Conclusion a good simulation of CO2 change
  depends on a good simulation of climate change.

from Jones et al., 2003
12
C4MIP
6 GCMs
4 EMICs
13
Cox et al. 2000 revisited and compared
14
C4MIP Simulated atmospheric CO2
Friedlingstein et al., 2005 centennial
simulations with dynamic growth of ecosystems,
under forced CO2 concentrations Results confirm
Cox et al., but to a much lesser extent.
coupled runs
coupled - uncoupled
All models simulate a positive feedback
Friedlingstein et al. (2005)
15
HadCM3, MPI 2.5K
Model climate sensitivity and responses for
doubling of CO2 CO2 concentration clearly
affects the surface climate, but also the ability
to assimilate CO2 Model responses are very
disparate.
CSM-1 1K
Ocean
Land
Uptake
LLNL
FRCGC
Change in uptake
HadCM3
Friedlingstein et al. (2005)
16
C4MIP Sensitivity analysis
LAND ?CL ?L ?CA ?L ?T
OCEAN ?CO ?O ?CA ?O ?T
?L
?O
negative feedback
Land and ocean sensitivity to CO2
positive feedback
?L
?O
Friedlingstein et al. (2005)
Land and ocean sensitivity to climate
17
C4MIP Sensitivity analysis

• The gain factor g will be larger if
• a, the GCM climate sensitivity to CO2, is large
• gL and gO , the ocean and land carbon cycle
  sensitivities to climate change, are large
• bL and bO , the ocean and land carbon cycle
  sensitivities to CO2 , are small.
• All models simulate a positive feedback.
• Uncertainty remains large on feedback strengths.
• More validation of the models needed.

18
C4MIP Sensitivity analysis
climate
CO2
Climate to CO2
19
UJCC GCM configurations for decadal to centennial
time scales
We believe that we need to resolve weather, in
order to study the climate system One of the big
open issues in carbon modelling is the regional
attribution of sources/sinks So, we have a match
made in heaven, but
135 km ATM
135 km ATM
270 km ATM

• 270 km ATM

2.5ox1.25o O
1o-1/3o O
1/3o O
1/3o O
60 km ATM
60 km ATM
90 km ATM
90 km ATM
20
HiGEM July GPP
21
Summary

• Modelling the carbon cycle is difficult, but not
  a challenge we can afford to decline
  carbon-related ideas and mechanisms are in all
  our models, with or without our
  knowledge/consent
• HadCM3 results confirmed by other models, but
  HaddCM3 shown to be an outlier, especially for
  what concerns the sensitivity of carbon uptake to
  surface temperature
• If we want to model long periods of our planets
  climate, we need to make sure that we can
  meaningfully represent the relevant processes
• Moving to higher resolution may reveal important
  local runaway feedbacks which break our initially
  good skills
• How good must the climate be for the land portion
  of the CO2 to be meaningful ? How much
  sensitivity is reasonable ?
• Two important possible responses of land
  ecosystems to eCO2, alone or in combination, for
  which 3rd generation models may contradict 2nd
  generation models
• Water savings
• Growth
• How much should we care about nutrient
  limitations ?

22
HadGEM/HiGEM-ESM The FutureA fully
interactive Earth System Model
Online
External Forcing plus other B.C.s
CLIMATE
Offline
Macro and micro physics Direct Indirect Effects
Greenhouse Effect
Human Emissions
AEROSOLS
GHGs
CH4, O3, N2O, CFC
Fires soot Mineral dust
Oxidants OH, H2O2 HO2,O3
Human Emissions
CO2
N deposition 03, UV radiation
CHEMISTRY
ECOSYSTEMS
Biogenic Emissions CH4,DMS,VOCs Dry
deposition stomatal conductance
Land-use changes
Emissions
(After Peter Cox)
Source Tim Johns, Hadley Centre
23
NPP diagnostics

• NPP summarizes physiologic activity water loss
  for carbon gain
• After less than 1 year, several regions show
  inactive vegetation
• Example of India, which appears to be a source of
  carbon in July.

24
Questions

• How good must the climate be for the land portion
  of the CO2 to be meaningful ?
• Two possible feedbacks
• Water savings
• Growth
• How general is the assumption about CO2
  fertilisation ?
• As a consequence, what is the probability for a
  runaway feedback ?
• 15-20 too much CO2 give us ???? too high
  temperature
• Range of EU models in PRUDENCE shows an earlier
  and deeper access of plant to root zone water in
  scenario experiments versus current climate
  experiments. If this should act in synergy with
  CO2 fertilisation effect, the exhaustion of soil
  water would be much more frequent.
• If CO2 response saturates, this would then
  compensate.

25
WRE scenarios

• WRE are a family of scenarios of CO2 level,
  stabilising at 450, 550, 650, 750 and 1000 ppmv
• Wigley, Richels and Edmonds. Economic and
  environmental choices in the stabilisation of
  atmospheric CO2 concentrations. Nature, 1996.
• We run the carbon cycle GCM with these prescribed
  CO2 levels and infer the emissions required to
  achieve them
• Results shown in detail for 550 ppm
• Summary of results for all levels

26
WRE550 Carbon budgets
27
WRE550 Carbon emissions
28
(No Transcript)
29
ESC model matrix the roles of formulation and
resolution
30
MSLP JJA biases
31
(No Transcript)
32
(No Transcript)
33
(No Transcript)
34
Sfc temperature JJA biases
35
(No Transcript)
36
(No Transcript)
37
(No Transcript)
38
(No Transcript)
39
(No Transcript)
40
precipitation JJA biases
41
(No Transcript)
42
(No Transcript)
43
(No Transcript)
44
Sub-domain fluxes in SE-USA and India (D. Clark)
P
P
L

• L

S
S
SE USA
India
s.m.
s.m.
45
Soil moisture evolutions, year 1 of HiGEM
46
HC GCMs summer biases also occur at very
interesting locations
47
Global Land-Atmosphere Coupling Experiment (
)
Impact of all land surface prognostic variables
48
SL diurnal evolution at FIFE site (JA) HadAM3
Criteria for separation of dry and wet days as in
observational work of BB95

• From Lawrence and Slingo 2005
• Model does not seem to be soil moisture limited
• Model seems to abort ML growth soon after local
  noon
• MSE buildup interrupted, with consequences for
  buildup of convection.
• Question can a model with more resolution
  -especially in the vertical-, and more realistic
  vertical mixing, do a better job ?

49
Latent heat fluxes and BL heights in two HiGEM
experiments mean JA diurnal cycles
Betts and Ball 1995
12
6
Latent heat flux
BL height
Local noon

• CTL experiment (top row) has often too warm
  conditions for vegetation activity during the
  peak of solar heating
• VEG experiment can develop the moister and
  shallower BL needed to the indirect
  SM-precipitation feedback

50
SL diurnal evolution at FIFE site (JA) HiGEM
Dry days
Dry days
Wet days
Wet days

• Experiment EACRA (right) has heat-resilient
  vegetation, so that LH flux does not shut down so
  easily at noon.

51
Our potential for ESM work

• What do these error mean for our current ability
  to start running model components such as TRIFFID
  ?

52
Soil respiration in July, yr 1
53
HadGEM family experiments in UJCC
54
Summary

• Experimented with a hierarchy of models
• HadGEM does not scale
• HiGEM1a still ice-crashing
• Important errors still present, also in HadGEM1a
  almost identical in all models analysed,
• Errors in some crucial climate variables are very
  relevant for ESM work.
• We still need for a problem worthy of the ES
  supercomputer moving to N216
• HC VNs need to be longer and include more modern,
  higher-res. data sets, including RS products
• Need concerted effort in solving systematic
  errors.

55
Complementary efforts systematically explore the
role of resolution attack systematic errors.
56
Summary of comparative studies for
HadAM3-HadGEM1-MOSES2 land surface coupling
strength

• For precipitation, extremely low level of
  surface-atmosphere (intra-seasonal) feedback, -as
  compared to other GCMs-, also recently confirmed
  by comparison to observations in Dirmeyer et al.
• Koster et al. have shown that soil moisture
  variability can control surface temperature
  variability in HadAM3
• ET shows a moderate level of coupling (and
  ?E(S)-?E(W)sE(W) is in the high range, so soil
  moisture can affect evaporation)
• However, in Lawrence and Slingo (2005) process
  study
• pointed to low SM variability (also in Vidale et
  al., Climatic Change, 2005)
• excluded soil moisture limitations in P coupling
• pointed to the wrong phase of ML diurnal cycle
  and its unrealistic development
• and too frequent/too weak precipitation too early
  in the day
• Are model biases coincident with regions where
  the coupling strength should be larger ?
• By taking land surface biases into account, are
  the mechanisms near the surface, in the BL or
  above ?

57
SL diurnal evolution at FIFE site (JA) HadGEM1

• 2x resolution in horizontal and vertical (even
  more in BL)
• HadGEM1 uses a non-local turbulent scheme BL
  profile classes also allows interplay of
  convective mass flux parameterization and BL
  scheme
• However, diurnal cycle very similar to that seen
  in HadAM3 by LawrenceSlingo
• Early ML development
• MSE buildup ceases soon after local noon
• Question could vegetation be chronically
  heat-stressed, leading to a positive (dryhot)
  feedback and periods of drought ?

Dry days
Wet days
58
Uncertainty in the definition of optimal
environmental temperature for physiology and
biophysics
Sellers et al. 1997

• Models exhibit huge range (about 18K), both
  linked to vegetation classes and to model
  formulation
• Shape of curves is also very variable
• Previous experiments with CHRM (PL Vidale,
  Europe) and with HadAM3 (C. Taylor, Sahel)
  indicated strong sensitivity
• Is a positive feedback really so easy to trigger
  at these GCM scales?

MOSES 2
C4 plants
C3 plants
59
SL moisture and heat fluxes in the two HiGEM
experiments mean JA diurnal cycles
2m specific humidity
Sensible heat flux

• CTL experiment (top) dries up SL during early
  morning SH fluxes always high
• VEG experiment can develop the moister and lower
  Bowen ratios seen in obs.

Betts and Ball 1995
12
6
Local noon
60
What happens to the diurnal evolution of the BL
in HadGEM/HiGEM ?
HadGEM/Higem have the (Lock et al. 2000) BL
scheme, which uses a non-local closure for the
unstable BL, an evolution of Holtslag-Boville
(1993). It also uses pre-defined
situations/profiles and allows mixing due to BL
top radiation and due to convection.

• A CRAZY
• PL-HYPOTHESIS
• (B. Holstlag did not completely jump out of his
  chair -well, almost- but I am still alive and he
  still talks to me)
• Lock et al. tell us that if conditions for
  cumulus are diagnosed, convective
  parameterization takes over the determination of
  exchange coefficients from the BL scheme, but
  since we allow a mass flux scheme to operate
  inside the BL, moist air can escape the BL at
  activation and dry air is entrained. Is this what
  is happening to MSE buildup ?

61
Summary

• Investigation of BL processes related to land
  surface-atmosphere feeback in HadGEM1-HiGEM
  showed a model behavior very similar to that
  observed within HadAM3
• Despite new BL resolutionformulation
  HadGEM1/HiGEM produce a mean diurnal cycle
  similar to that in HadAM3
• Modeled vegetation is not soil moisture limited
  over FIFE, but it could be heat stressed
• Tests with heat-resilient vegetation indicate
  that variability of summer BL development/growth
  could be improved
• Conditions for direct and indirect
  (shallowermoister ML) soil moisture-precipitation
  feedbacks could be met if positive dryhot
  feedback is avoided (stomatal suicide)
• However, it may well be that convective mass flux
  is activated too early and steals BL air,
  interrupting MSE buildup and creating shallow
  convection
• Are we sure that we are still low in the level
  of coupling range ?
• Need to repeat coupling strength ensemble
  experiments and sensitivity studies with new
  AGCM, so as to have better sample, and to expand
  analysis to other sites.
• BUT one more thing

62
For people who say mod more than once a day

• The land surface coupling strength test described
  in the introduction relies upon the ability to
  perform seasonal ensemble runs, in which members
  are initialized at the end of each spring and run
  for 60-90 days. This only makes sense if we can
  spawn a reasonable response to initial
  perturbations.

CCM2
HadGEM
63
Global Land-Atmosphere Coupling Experiment (
)
On the other hand, consensus shows which are the
regions of high level of coupling and those are
the regions where HadAM3/HadGEM1 has the largest
summer biases (dryhot)
64
precipitation JJA biases
65
Increasing complexity and resolution at the same
time is extremely expensive. Climate Modelling
and Prediction requires huge supercomputers
66
First results from ESC
HadGEM 1

• 5-10 years of coupled simulations, using a chain
  of HC models with increasing complexity and
  resolution
• Impacts on well-known Pacific cold bias and
  precipitation in the warm pool
• HiGEM can reproduce ENSO, which is virtually
  absent in HadGEM1
• Some work still ahead with sea-ice and land
  surface biases
• Preliminary work of this kind is needed to
  confirm portability/reproducibility and to
  identify the best model for ESM building.

HadGEM 1a
HiGEM 1
67
HadAM3 also had a problem with precipitation
frequency
68
Land-atmosphere coupling strength diagnostic
W(rite) - 16-member ensemble forced with June 1
initial conditions from each year of a 16-year
climatological SST control run. Soil moisture
from W1 experiment recorded. R(ead) - 16-member
ensemble where, at every timestep, simulated soil
moisture is discarded and replaced with values
from W1 experiment.
O measure of time series similarity between
ensemble members
R ? SST and Soil moisture
W ? SST
OP(W) 0.07
OP(R) 0.85
69
Land-atmosphere coupling strength (HadAM3-MOSES2)