Land-Atmosphere Feedback in the Sahel

1
Land-Atmosphere Feedback in the Sahel
Randal Koster Global Modeling and Assimilation
Office NASA/GSFC Greenbelt, MD randal.d.koster_at_nas
a.gov
2

• Organization of Talk
• Overview of the processes that control
  land-atmosphere feedback. (Case study North
  America)
• Application of these ideas to the Sahel do the
  observations support the existence of feedback
  there?
• 3. Model study of the controls on Sahelian
  rainfall variability.

3
Warm season precipitation variance is often high
in transition zones between dry and wet
areas. Example North America
July Rainfall Mean
July Rainfall Variance
mm/day
mm2/day2
Observations (Higgins, 50-yr dataset)
Koster et al., GRL, 40, 3004
4
More evidence tree ring data! (360 years of
proxy precipitation data put together by H.
Fritts, U. Arizona)
Jul/Aug precipitation variances at each tree ring
site
Shading Mean annual precipitation (GPCP)
White dots Locations of tree ring sites with
Jul/Aug precipitation variances in top half of
range
5
Q Do we have any reason to suspect that
precipitation variances should be amplified in
transition zones? A Yes. Transition zones are
more amenable to land-atmosphere feedback.
which affects the overlying atmosphere (the
boundary layer structure, humidity, etc.)...
causing soil moisture to increase...
Precipitation wets the surface...
which causes evaporation to increase
during subsequent days and weeks...
thereby (maybe) inducing additional
precipitation
6
Feedback enhances s2P through the enhancement of
P autocorrelation (on timescales of days to
weeks).
Observed s2P
correlates with
Pn
Pn2
means that
Pn
Pn2
correlates with
correlates with
En2
correlates with
wn
wn2
correlates with
7
Feedback enhances s2P through the enhancement of
P autocorrelation (on timescales of days to
weeks).
Observed s2P
correlates with
Pn
Pn2
means that
Pn
Pn2
Breaks down in western US low evaporation
correlates with
correlates with
Breaks down in western US low soil moisture
memory
En2
correlates with
wn
wn2
correlates with
8
Feedback enhances s2P through the enhancement of
P autocorrelation (on timescales of days to
weeks).
Observed s2P
correlates with
Pn
Pn2
means that
Pn
Pn2
correlates with
correlates with
En2
correlates with
wn
wn2
Breaks down in eastern US low sensitivity of
evaporation to soil moisture
correlates with
9
Feedback enhances s2P through the enhancement of
P autocorrelation (on timescales of days to
weeks).
Observed s2P
correlates with
Pn
Pn2
Only in the center of the country (in the wet/dry
transition zone) are all conditions ripe for
feedback
means that
Pn
Pn2
correlates with
correlates with
En2
correlates with
wn
wn2
correlates with
10
We therefore have reason to believe that
land-atmosphere feedback can help explain the
patterns of observed precipitation
variances. Note up to this slide, we havent
looked at any model results! What can AGCMs tell
us?
11
July Rainfall Mean
mm/day
AGCM
AGCM, no feedback
Observations (Higgins, 50-yr dataset)
same plots as before
12
July Rainfall Mean
mm/day
AGCM
bulls-eye in model is definitely induced by
feedback!
AGCM, no feedback
Observations (Higgins, 50-yr dataset)
?The observations show statistics that are
similar in location and timing, though not in
magnitude, to those produced by the GCM. This is
either a coincidence or evidence of feedback in
nature.
13
Central North America, of course, is just one of
the Earths wet/dry transitions zones.
Annual Precipitation
Another is the Sahel
Does nature allow land-atmosphere feedback to
affect rainfall statistics in the Sahel?
14
Precipitation Variances (mm2/day2)
AGCM
The comparison between model results and
observations isnt as clear-cut as it is in North
America, but it is suggestive
AGCM with no land feedback
Observations
15
Precipitation Variances (mm2/day2)
AGCM
The comparison between model results and
observations isnt as clear-cut as it is in North
America, but it is suggestive
AGCM with no land feedback
Observations
The dots show where precipitation itself is
maximized
16
Another observational study If land-atmosphere
feedback operates in the Sahel, then realistic
land initialization there should lead to improved
monthly forecasts. Test with comprehensive
forecast study 75 start dates (first days of
each month May to September) 9 ensemble
members per forecast In one set of forecasts,
utilize realistic land ICs In other set,
dont utilize realistic land ICs
Compare
17
Forecast skill resulting from realistic land
surface initialization appears negligible for
precipitation
Differences Added forecast skill from realistic
land ICs
Skill from knowing SST distribution and realistic
land ICs
Skill from knowing SST distribution
Precipitation
Precipitation
Precipitation
Temperature
Temperature
Temperature
18
HOWEVER, locations for which the rain gauge
density is adequate enough to properly initialize
the model are arguably very limited.
Added forecast skill from land initialization
Precipitation
Regions w/adequate raingauge density and model
predictability
Temperature
19
So, for the feedback question, observations are
limited. Consider now a pure model study...
of
Total Exp. simulations Length years
Description
Evaporation efficiency (ratio of evaporation to
potential evaporation) prescribed at every time
step to seasonally-varying climatological means
A 4 200 yr
800 AL 4 200 yr
800 AO 16 45 yr
720 ALO 16 45 yr
720
Prescribed, climatological land
climato- logical ocean
Interactive land, climato- logical ocean
Prescribed, climatological land, interan- nually
varying ocean
SSTs set to seasonally-varying climatological
means (from obs)
SSTs set to interannually-varying values (from
obs)
Interactive land, interan- nually varying ocean
LSM in model allowed to run freely
Koster et al., J. Hydromet., 1, 26-46, 2000
20
Simulated precipitation variability can be
described in terms of a simple linear system
Total precipitation variance
Precipitation variance in the absence of land
feedback
s2ALO
s2ALO s2AO Xo ( 1 - Xo )
s2AO
Fractional contribution of ocean processes to
precipitation variance
Land-atmosphere feedback factor
Fractional contribution of chaotic atmospheric
dynamics to precipitation variance
The above tautology isolates the relative
contributions of SSTs, soil moisture, and
chaotic atmospheric dynamics to precipitation
variability.
21
Contributions to Precipitation Variability
22
Idealized predictability (for 1-month
forecasts, MJJAS) deduced from aforementioned
forecast experiment. (Ability of model to
predict itself.)
Differences Added predictability from realistic
land ICs
Predictability from SST distribution and
realistic land ICs
Predictability from SST distribution
Precipitation
Precipitation
Precipitation
Temperature
Temperature
Temperature
Temperature
23
More AGCM results The GLACE multi-model
experiment. In GLACE, land-atmosphere feedback
was quantified independently in 12 AGCMs. While
the models differ in their feedback strengths,
certain features of the coupling patterns are
common amongst them. These features are brought
out by averaging over all of the model results
24
More AGCM results The GLACE multi-model
experiment. In GLACE, land-atmosphere feedback
was quantified independently in 12 AGCMs. While
the models differ in their feedback strengths,
certain features of the coupling patterns are
common amongst them. These features are brought
out by averaging over all of the model results
The AGCMs tend to agree land-atmosphere feedback
operates in the Sahel.
25

• To summarize
• Organization of Talk
• Overview of the processes that control
  land-atmosphere feedback. (Case study North
  America)
• Application of these ideas to the Sahel do the
  observations support the existence of feedback
  there?
• 3. Model study of the controls on Sahelian
  rainfall variability.

26

• To summarize
• Organization of Talk
• 1. Overview of the processes that control
  land-atmosphere feedback. (Case study North
  America)
• 2. Application of these ideas to the Sahel do
  the observations support the existence of
  feedback there?
• 3. Model study of the controls on the West
  African monsoon.

We think we understand the impact of
land-atmosphere feedback on the statistics of
precipitation in North America. Through
feedback, precipitation memory and variance are
increased in the transition zones between wet and
dry areas. The observations appear to support
this.

27

• To summarize
• Organization of Talk
• 1. Overview of the processes that control
  land-atmosphere feedback. (Case study North
  America)
• 2. Application of these ideas to the Sahel do
  the observations support the existence of
  feedback there?
• 3. Model study of the controls on the West
  African monsoon.

Observations are too sparse in the Sahel
(relative to North America) for an equally clear
indication that land atmosphere feedback operates
there. Nevertheless, the available observations
are not inconsistent with feedback.
28

• To summarize
• Organization of Talk
• 1. Overview of the processes that control
  land-atmosphere feedback. (Case study North
  America)
• 2. Application of these ideas to the Sahel do
  the observations support the existence of
  feedback there?
• 3. Model study of the controls on Sahelian
  rainfall variability.

The NSIPP model (and indeed most of the models
participating in GLACE) show the Sahel to be a
region of strong land-atmosphere feedback.
29
The above modeling results may, of course, be
model dependent. A new, upcoming experiment may
provide a clearer look at the controls on monsoon
dynamics
WAMME West African Monsoon  Modeling and Evaluation
See website http//wamme.geog.ucla.edu/ A
Spring AGU (Acapulco) session addresses the
experiment
30
(No Transcript)
31
(No Transcript)
32
Experiment Design
W Simulations Establish a time series of surface
conditions
time step n1
Step forward the coupled AGCM-LSM
Step forward the coupled AGCM-LSM
Write the values of the land surface prognostic
variables into file W1_STATES
Write the values of the land surface prognostic
variables into file W1_STATES
(Repeat without writing to obtain simulations W2
W16)
All simulations are run from June through August
33
Experiment Design (cont.)
R(S) Simulations Run a 16-member ensemble, with
each member forced to maintain the same time
series of surface (deeper) prognostic variables
time step n1
time step n
Step forward the coupled AGCM-LSM
Step forward the coupled AGCM-LSM
Throw out updated values of land
surface prognostic variables replace with
values for time step n from file W1_STATES
Throw out updated values of land
surface prognostic variables replace with
values for time step n1 from file W1_STATES
34
Participating Groups
Contact
Model
Country
McAvaney/Pitman
1. BMRC with CHASM
Australia
Kanae/Oki
2. U. Tokyo w/ MATSIRO
Japan
Dirmeyer
3. COLA with SSiB
USA
Kowalczyk
4. CSIRO w/ 2 land schemes
Australia
Oleson
5. NCAR
USA
Canada
Verseghy
6. Env. Canada with CLASS
Gordon
7. GFDL with LM2p5
USA
Sud
8. GSFC(GLA) with SSiB
USA
Taylor
9. Hadley Centre w/ MOSES2
UK
Lu/Mitchell
10. NCEP/EMC with NOAH
USA
Koster
11. NSIPP with Mosaic
USA
Xue
12. UCLA with SSiB
USA
35
W GFDL Scale goes from 0 to 1
S GFDL Scale goes from 0 to 1
Differences GFDL Scale goes from -0.5 to 0.5
36
Another pure model study (no observations)
monsoon rainfall
What controls the timing of the monsoon?
Quantify importance of

• Average solar cycle.
• Interannual SST variations
• 3. Interannual soil moisture variations

Region considered
37
Illustration of W diagnostic (not for African
monsoon region)
Precipitation time series produced by different
ensemble members under the same forcing
All simulations in ensemble respond similarly to
boundary forcing W is high
Simulations in ensemble have no coherent
response to boundary forcing
W is low
38
The contributions of the different boundary
forcings to the agreement (between ensemble
members) of monsoon structure is established by
analyzing the outputs of various experiments
NSIPP model
solar, SSTs, soil moisture
solar, SSTs
solar, SSTs
solar,
W
(Middle two bars differ because they were derived
from different experiments, with different
assumptions.)
39
The contributions of the different boundary
forcings to the agreement (between ensemble
members) of monsoon structure is established by
analyzing the outputs of various experiments
NSIPP model
solar, SSTs, soil moisture
solar, SSTs
solar, SSTs
solar,
W
In this model, soil moisture variations have a
major impact on monsoon evolution
(Middle two bars differ because they were derived
from different experiments, with different
assumptions.)