Ghent - CarboAfrica

carboafrica.net
  • No tags were found...

Ghent - CarboAfrica

Assimilation of land-surface temperature in

the land surface model JULES over Africa

D. Ghent, J. Kaduk, H. Balzter

Department of Geography, University of Leicester, UK

email: djg20@le.ac.uk

Open Science conference on

“Africa and Carbon Cycle: the CarboAfrica project”

Accra (Ghana) 25-27 November 2008


Research rationale

• Land-surface temperature (LST), which is the radiative skin

temperature of the land, is a critical variable for fuel moisture

content (FMC).

• In land-surface models, air temperature is more commonly employed,

even though LST is more closely related to the physiological activities

of leaves.


Aims and objectives

Aim:

• The aim of this study is to constrain the simulation of FMC, through

the assimilation of remotely sensed LST into a land-surface model.

Objectives:

• Multi-temporal analysis between modelled and observed LST;

• Assimilation into land-surface model of remote sensing product.

Questions:

• How does modelled LST compare with remotely sensed observations?

• Can the assimilation of LST observations into a land-surface model

improve the modelled results?


Methodology: design

Model

• JULES (Joint UK Land Environment Simulator) is the community

version of the UK Met Office Surface Exchange System;

• Coupled with the TRIFFID Dynamic Global Vegetation Model;

• Can be coupled to the HadCM3 general circulation model or driven by

its output.

Methods and data

• Spatial resolution: 1° x 1°;

• Temporal resolution: 1 hour;

• 6-hourly interpolated NCEP reanalysis meteorological data;

• TRMM corrected 6-hourly interpolated NCEP reanalysis precipitation;

• Prescribed IGBP land cover classes;

• Soil texture from the IGBP-DIS 1° map.


Methodology: observation data

SEVIRI MSG1

• Orbit: Geostationary

• Spatial resolution: 3km

• Temporal resolution: 15 minutes

Terra MODIS

• Orbit: Sun-synchronous, near-polar

• Spatial resolution: 1km

• Temporal coverage: Global every 1-2 days

Envisat AATSR

• Orbit: Sun-synchronous, polar

• Spatial resolution: 1km (at nadir)

• Temporal coverage: Global every 3 days


Methodology: assimilation

Perturbed Input Data

JULES

Modelled

Model Error Cov

Obs Error Cov

X 1 X 2 … X N

X 1 X 2 … X N

No

Timestep + 1

Obs?

Yes

Perturbed Obs

Ensemble Kalman Filter

Assimilated


Results: LST comparison

JULES

AATSR

MODIS

SEVIRI

Comparison of mean daytime LST composites for June 2006.


Results: LST comparison

Analysis: June 2006

• Strongest correlations

between remote sensing

observations.

• Largest JULES deviation for

bare soil land-cover class.

ATSR JULES 0.755

ATSR MODIS 0.888

ATSR SEVIRI 0.916

JULES MODIS 0.723

JULES SEVIRI 0.777

MODIS SEVIRI 0.874

Paired samples correlations for

mean June daytime LST (N=2171).

All correlations are significant.


Results: LST comparison

• Mean AATSR uncertainty: < ±0.9K

(Coll et al., 2005).

• Mean MODIS uncertainty: < ±1.0K

(Pinheiro et al., 2007).

Day

• Mean SEVIRI uncertainty ±1.5K

(Sobrino and Romaguera, 2004).

Day deviation (K) Night deviation (K)

JULES -6.16 -3.87

AATSR +1.47 +1.53

MODIS -2.00 -2.20

SEVIRI +0.52 +0.67


Results: LST comparison

AATSR

• Greater proportion of

missing data.

MODIS

• Deviation increases

with viewing angle.

SEVIRI

• Temporal resolution

more appropriate.

• Lowest deviation.

Mean MODIS LST deviation categorised by

viewing angle.


Results: assimilated LST from SEVIRI

Modelled LST vs. SEVIRI

(March 2006)

Mean deviation = -2.48K

Assimilated LST vs. SEVIRI

(March 2006)

Mean deviation = -2.03K

Mean correction of assimilated LST (March

2006).

Mean LST correction = 0.45K


Results: assimilated LST

Modelled vs. assimilated LST for bare soil (16 th -31 st March 2006).


Results: assimilated LST

Modelled vs. assimilated LST for wooded grassland (16 th -31 st March 2006).


Conclusions

• LST modelled with JULES is comparable with remote sensing

observations.

• The modelled LST has a negative deviation in comparison with the

remote sensing LST, which is strongest where no vegetation exists.

• The remote sensing observations are comparable to each other with

strong correlations.

• The assimilation of LST into JULES using an Ensemble Kalman Filter

reduces the deviation; with the bare soil land class displaying the

greatest correction.

• Assimilation of remote sensing products can prove to be a feasible

method of improving model simulation.


Research outlook

• Determination of the optimal ensemble size.

• Assimilation over a range of different time scales.

• Investigation into JULES LST discrepancy for bare soil.

• Investigation of the impact upon surface heat fluxes of assimilated

LST.

• Optimisation of a surface dryness index expressed as a function of

assimilated LST and NDVI.


References

• Coll, C., Caselles, V., Galve, J. M., Valor, E., Niclos, R., Sanchez, J. M. and

Rivas, R. (2005) Ground measurements for the validation of land surface

temperatures derived from AATSR and MODIS data. Remote Sensing of

Environment 97(3): 288-300.

• Pinheiro, A. C. T., Descloitres, J., Privette, J. L., Susskind, J., Iredell, L.,

Schmaltz, J. (2007) Near-real time retrievals of land surface temperature

within the MODIS Rapid Response System. Remote Sensing of Environment

106(3): 326-336.

• Sobrino, J. A. and Romaguera, M. (2004) Land surface temperature retrieval

from MSG1-SEVIRI data. Remote Sensing of Environment 92(2): 247-254.

More magazines by this user
Similar magazines