Estimation of the carbon balance components of heterogeneous agricultural landscape using tall
tower based and remotely sensed data
1
1
1,2
3
Barcza, Z. ., Gelybó, Gy. , Kern, A. , Kljun, N. , Haszpra, L.
1 Department of Meteorology, Eötvös Loránd University, H-1117 Budapest, Pázmány P. sétány 1/A, Hungary
2 Adaptation to Climate Change Research Group, Hungarian Academy of Sciences, H-1117 Budapest, Pázmány P. sétány 1/A, Hungary
3 Department of Geography, Swansea University, Singleton Park, Swansea SA2 8PP, United Kingdom
4 Hungarian Meteorological Service, H-1675 Budapest, P.O.Box 39, Hungary
Using the combination of remotely sensed data, flux tower measurements and state-of-the-art footprint model, we estimate ecosystem productivity of
heterogeneous agricultural landscape in Hungary. Gross Primary Production (GPP) estimation using remotely sensed data is based on the MOD17 GPP
model. The results are compared to flux tower measurements. In order to utilize footprint information obtained with 250 m spatial resolution, the relatively
coarse 1 km resolution MOD17 model is scaled down using NDVI data (MOD13).
Retrieved information on
common grid
Measurements
Typical annual NDVI time
series for winter and
summer crops
Terra & Aqua MODIS NDVI
250 m resolution
Crop fraction maps on 250 m
NDVI grid (below: summer
crop fraction in 2007)
MOD17 model
0.9
Wavelet
transformation
2007
0.8
Crop
classification
NDVI
0.7
0.6
250 m resolution FPAR
0.5
0.4
Eddy-covariance
measurements
250 m resolution crop map
0.3
0.2
0
100
200
300
day of year
Dynamic simulation of tall
tower EC measurements
2007
41
41 pixels
s
l
e
pix
250 m resolution footprint
Hourly source
area detection
250 m resolution land cover
Rasterization
1 km
Rasterized footprint
climatology
250 m NDVI grid
Rasterized CORINE2000
Footprint climatology +
CORINE2000 land cover
(~100m resolution)
Grey:non-agricultural area
Synthesizing
ground based and
remotely sensed data over heterogeneous agricultural land.
The 250 m resolution is comparable with individual field size.
The cutout represents 41x41 pixels that covers the sampled
area.
NDVI signal processing for crop classification.
Determining representativeness of ground based
measurements.
Crop specific fluxes
Retrieve land cover, crop type and footprint
information on the same common (NDVI 250m)
grid for modeling.
Kljun et al., 2004
Barcza et al., 2009
The MOD17 algorithm
emax
MOD17 1km resolution
Tmin, VPD
Official
4000
GMAO GR [MJ m-2 day-1]
30
20
GMAO Tmin [°C]
2000
1000
0
0
0.2
10
20
t [8 days]
30
1000
2000
3000
Tower VPD [Pa]
10
0
-10
40
20
10
0
-20
-30
-30
30
250 m resolution FPAR
0
4000
-20
-10
0
10
Tower Tmin [°C]
20
Land cover type of the
corresponding pixel
Wavelet transformed
NDVI signal from the
corresponding pixel
40
3000
GMAO VPD [Pa]
FPAR
Dynamically changing footprint
location on the NDVI grid
MOD17 tower meteorology
0.8
0.4
GPP
Met.
FPAR quality control and
interpolation. Figure shows FRAP time
series before and after the procedure
0.0
0
e
X
PAR
Rnet
FPAR
0.6
NDVI based downscaling of MOD17
FPAR
10
20
30
-2
-1
Tower GR [MJ m day ]
40
30
GPP algorithm
Six-years-long relationship between local measurements and the GMAO data.
Note that GMAO accuracy varies between years, as well.
Flowchart of the downscaling algorithm
10
- Meteorological input: GMAO reanalysis NTSG
8
1400
6
1200
4
2
version 5.1
- Adapted algorithm: FPAR quality control, GPP
calculations
References:
Barcza, Z., Kern, A., Haszpra, L., Kljun, N. (2009). Spatial representativeness of
tall tower eddy covariance measurements using remote sensing and footprint
analysis. Agric. Forest Meteorol., 149, 795-807.
Kljun, N., Calanca, P., Rotach, M. W., Schmid, H. P. (2004). A simple
parameterisation for flux footprint predictions. Bound.-Lay. Meteorol., 112, 503523.
Acknowledgements:
The research has been supported by the Hungarian State Eotvos Fellowhip
Average annual GPP sums: tower
measurements and model results
GPP [g C m-2 day-1]
FPAR (MOD15)
Tower
MOD17
MOD17 - local met.
GPP [g C m-2year-1]
- Satellite data input: Land cover (MOD12),
Results
GPP [g C m-2 day-1]
Official MOD17 product:
Light use efficiency ( ) based model
10
2001
2002
2003
2004
2005
2006
0
0
50
100
150
200
250
Time [8 days]
GPP time series: tower measurements, model
results (official and local meteorology)
300
1000
800
600
Tower
MOD17
MOD17 - NDVI
8
6
4
2
MOD17
400
2001
MOD17 - met
0
Tower
200
0
MOD17 - NDVI
2002
50
2003
100
2004
150
2005
200
2006
250
300
Time [8 days]
0
2000
2001
2002
2003
2004
Time
2005
2006
2007
GPP time series: tower measurements, model
results (official and downscaled)
Results from the three model setups show that errors in meteorology are not exclusively responsible for occasionally poor
model performance. Improvements in model spatial resolution (primarily determined by FPAR resolution) can significantly
improve accuracy. After eliminating possible sources of errors in input meteorology and spatial resolution, it is clear that the
calibration of the model, even creating new PFTs is the next step toward more accurate results.