A new dust cycle model that

includes dynamic vegetation

Sarah Shannon

Dan Lunt & Sandy Harrison

Department of Geographical Sciences, University of Bristol, UK

Talk outline

� Introduction – Why include dynamic vegetation in a dust cycle model ?

� Model description

� Model tuning � Model tuning

� Case study: Can variability in the Barbados dust record be explained by vegetation changes in the Sahel ?

Introduction

� Dust cycle models describe vegetation using

� remote sensing data

� BIOME4/BIOME3 models

� Using BIOME4 it is not possible to test if changes in

the atmospheric dust loading is caused by vegetation the atmospheric dust loading is caused by vegetation

changes or other processes

� To test this we require dynamic vegetation

The LPJ-dust model

LPJ-DGVM

(Sitch et al. 2003)

CRU

precipitation,

temperature

and cloud cover

(monthly)

Calculate monthly dust

source areas

Annual CO2

soil texture

ERA-40 large

scale &

convective

precipitation, low

& medium cloud

amounts (6hr)

0.5 x 0.5

Surface emissions

(Tegen et al. 2002)

TOMCAT

(Chipperfield 2006)

ERA-40 10m

wind speed (6hr)

Soil texture

ERA-40 3D wind

speed,

temperature &

specific humidity

(6hr)

Sub cloud

scavenging & dry

deposition (1hr)

Output dust fields

amounts (6hr)

T42

Calculate monthly dust source areas from LPJ

BIOME mapDominant FPC

• Scheme used to reduce high latitude dust emissions

• Use annual mean FPC, GDD5 and tree height to make BIOME map (Joos et al. 2004)

• Prohibit dust emissions from polar desert

• Calculate monthly bare ground fraction using simulated FPAR, soil moisture and snow

depth. Grasses vary seasonally, shrubs protect surface all year around

Mean data for the year 1988 to 2002 using CRU 2.1 data.

Tuning the LPJ-dust model

� Tune threshold limits for surface

emissions

� Threshold friction velocity

� FPAR limit

� Soil moisture limit

� Snow depth limit

� Use Latin Hypercube sampling to

choose select values choose select values

� Tune sub-cloud scavenging

parameterisation

� Λ is independent of rain droplet

size – Brandt et al. 2002

� Λ is dependent on rain droplet

size, 0.5µm and 2µm - Slinn 1983

• Number of tuning experiments

5 x 4 = 20

+1

21 x 3 = 63

• Run model for 1987 to 1989

Tune to observations� DIRTMAP deposition (excluding loess data)

� Ginoux et al., 2001 deposition

� University of Miami surface concentrations (annual mean data -1989)

� Create a skills score

� Calculate T, the value which minimises the NRMSE

� Error = NRMSEDIRTMAP + NRMSEGINOUX + NRMSEMIAMI

� Best experiment uses Slinn Dp=0.5µm, FPARlim=0.37, SMlim=7.79mm, SDlim=0.01m, TFRSF=0.55

DIRTMAP (blue), Ginoux data (Pink) and

University of Miami data (Red).

Slinn Dp=0.5µm

Slinn Dp=2µm

Fixed scav

Experiment skill

Measured and modelled deposition rates for the un-tuned and tuned model

results

Greenland

Antarctica

North Pacific

South Pacific

North Atlantic

South Atlantic

Arabian sea

DIRTMAP DIRTMAP

Arabian sea

Ginoux Ginoux

North Pacific

North Atlantic

South Pacific

1.French Alps

2. Spain

3.Tel Aviv

4.Taklimakan

Model validation: Simulated and observed surface concentrations

Model validation: Simulated and observed surface concentrations

� Good agreement in the

North Pacific– spring

peak simulated

Model validation: Simulated and observed surface concentrations

� Good agreement in the

North Pacific– spring

peak simulated

� Relatively good

agreement in the north

Atlantic

Model validation: Simulated and observed surface concentrations

� Good agreement in the

North Pacific– spring

peak simulated

� Relatively good

agreement in the north

Atlantic

� Poor agreement at � Poor agreement at

sites effected by

Australian dust

emissions- model

simulates peak Aug-

Dec when LPJ predicts

minimum vegetation

cover

Case study: Can vegetation changes explain the

cause of high dust concentrations at Barbados

during the 1980s?

60

80

gm

3

Barbados 13.17N 59.43W40

Ob

se

rva

tio

ns µ

gm

3

Mahowald et al. 2002

suggested high dust

concentrations during the

1980s were caused by

expansion of the Sahara or

Simulated and measured annual mean surface concentrations at

Barbados. The correlation coefficient for data between 1965 and 1978 is

r=0.82

1965 1970 1975 1980 1985 1990 1995 20000

20

40

Mo

de

l µg

m

1965 1970 1975 1980 1985 1990 1995 20000

20

Ob

se

rva

tio

ns

land use changes.

Southward shift of the Sahara-Sahelian boundary line in 1984

• Vegetation shifts southwards in 1984 relative to 1966

• Emissions in Sahel (Latitudes 10oN -20oN, longitudes 17oW – 40oE)

double from 1.1Mtyr-1 in 1966 to 2.2Mtyr-1 in 1984

Conclusions

� The LPJ-dust model can predict the expansion and contraction of

dust source regions by changes in vegetation cover which was not

possible using the BIOME4 model

� Tuning the LPJ-dust model improved estimates of deposition rates

to the North & South Pacific, North Atlantic and Arabian seato the North & South Pacific, North Atlantic and Arabian sea

� The case study showed there was a southward movement of the

Sahara in 1984 relative to 1966 but this is not sufficient to account

for the high dust concentrations measured at Barbados.

sarah.shannon@bristol.ac.uk

Extra slides

The LPJ-dust model

Calculate monthly dust source areas from LPJ: Step 2

� Calculate bare area using monthly FPAR, soil moisture and snow cover.

� Grass biomes

� Shrub biomes

<−

=otherwise

FPARFPARifFPAR

FPAR

Aveg

0

limlim

1

� Snow cover reduces exposed area

� Abare=Aveg. Asnow.Iθ where Iθ =1 if SOILMOISTURE < SOILMOISTURElim ELSE 0

<−

=otherwise

FPARFPARifFPARAveg

0

limmax1

<−=

otherwise

SNOWDEPTHSNOWDEPTHifSNOWDEPTH

SNOWDEPTH

Asnow

0

limlim

1