Modeling Gross Primary Production in Semi-arid Inner Mongolia using MODIS Imagery and
Eddy Covariance Data
Ranjeet John1, Jiquan Chen1, Asko Noormets2, Jianye Xu1, Xiangming Xiao3, Nan Lu4, and Shiping Chen5
of Toledo, OH; Southern Global Change Program, North Carolina State University, Raleigh, NC; 3 University of Oklahoma, University of Oklahoma;
4 Research Center for Eco-Environmental Sciences, 5Institute of Botany, Chinese Academy of Sciences, Beijing, People’s Republic of China.
1 University
2
Results
Introduction
(Fig. 2 and 3) Seasonal changes in observed GPPEC and predicted GPPVPM in 2006 & 2007.
Linear regression between observed GPPEC and predicted GPPVPM in 2006 & 2007 (Fig. 4 &
5). Seasonal changes in observed GPPEC as compared with GPPMVPM (Fig. 7).
Latitude
longitude
42.046667 116.283610 1350
42.045556 116.279722 1350
2006
423
410
2007
199
190
43.554444 116.671389 1250
40.563333 108.745000 1020
152.9
227
170.4 7.2
180.4 12.3
7.7
11.5
6
40.451667 108.625000 1160
202.7
180.4 12.3
11
4
a
Eddy covariance GPP
VPM GPP
VPM GPP
6
MODIS GPP
NDVI
EVI
0.2
b
0
6
2
0
c
6
4
2
0
d
6
b
3.5
8
b
1
2
3
4
5
6
7
0.4
4
8
c
9
10
y = 1.34x ‐ 0.02
R² = 0.80
y = 0.53x ‐ 0.03
R² = 0.82
2
0.2
c
8
0
2
0.5
1
1.5
2
2.5
3
0.4
0.2
2
0.4
6
4
4
2
2
0.2
2
y = 0.15x + 0.24
R² = 0.40
0.5
1
0.6
4
0
0
e
d
0
6
y = 1.31x + 0.39
R² = 0.74
0
0.6
4
8
3.5
d
0
6
0
0
6
0.6
6
0
0
4
0
1.5
2
2.5
3
e
8
3.5
e
0
0.6
6
0.4
4
y = 2.41x + 0.53
R² = 0.31
0.2
2
y = 0.31x + 0.20
R² = 0.41
0
0
0
4
0.2
0.4
0.6
Fig 4.
DOY
a
Eddy covariance GPP
0.8
Eddy
1
1.2
1.4
Covariace GPP (g C m‐2
1.6
1.8
Fig 6.
a
VPM GPP
6
MODIS GPP
2
0
40
6
0
c
6
4
2
0
d
6
VPM GPP & MODIS GPP (g C m‐2 d‐1)
2
0.2
0.4
0.6
0.8
1
1.2
1.4
60
40
20
y = 1.04x ‐ 0.02
R² = 0.67
0
0
0.5
1
1.5
2
2.5
y = 1.84x ‐ 0.49
R² = 0.73
c
4
2
y = 0.63x ‐ 0.15
R² = 0.79
0
6
0
0.5
1
1.5
2
2.5
3
3.5
d
2
0
0
e
6
6
0
0
d
20
y = 1.01x ‐ 0.18
R² = 0.68
0
0.2
0.4
0.6
0.8
0
1
1.2
1.4
1.6
1.8
e
2
e
60
40
2
2
20
40
4
4
c
40
60
y = 0.32x + 0.04
R² = 0.55
2
0
60
4
4
4
b
1.6
b
y = 2.42x ‐ 0.27
R² = 0.71
2
6
MVPM GPP
0
0
4
4
a
VPM GPP
20
y = 1.23x + 0.02
R² = 0.43
0
b
6
DOY
Eddy covariance GPP
60
y = 1.01x ‐ 0.02
R² = 0.61
4
MODIS GPP
2
d‐1)
VPM GPP
2
GPP (g C m‐2 d‐1)
3
2
6
2
Fig 2.
y = 0.31x + 0.05
R² = 0.49
y = 0.62x ‐ 0.08
R² = 0.63
20
0
0
Fig 3.
0
DOY
0.2
Fig 5.
0.4
0.6
0.8
1
1.2
1.4
Eddy Covariace GPP (g C m‐2 d‐1)
Fig 7.
DOY
EC flux towers, DO1, D02, X03, K04 & K05 are a, b, c, d, & e, in the figures above.
GPPVPM = εg x fAPARchl x PARi
GEPMVPM = α [ln (GPP MODIS)*(EVI*LSWI*LST)]
2.5
0
4
6
Site
εg = ε0 x Tscalar x Pscalar x Wscalar
2
y = 0.23x + 0.35
R² = 0.55
0
MVPM in particular, offers a cost effective solution in
predicting GPP at remote study sites that lack
infrastructure to set up ground based sensors.
1.5
2
4
6
While VPM needs just four parameters obtained from flux
towers for each of the five ecosystem/LCLU types, the
MVPM is independent of any ground measured
meteorological data. Both models are based on simple
equations that are not computationally intensive like most
process-based models.
1
y = 0.74x + 0.41
R² = 0.67
0
The Vegetation Photosynthesis model (VPM) and modified
VPM (MVPM), driven by Enhanced Vegetation Index (EVI)
and Land Surface Water Index (LSWI) for 2006-2007,
derived from the MODIS surface reflectance product
(MOD09A1), was used to scale up and validate temporal
changes in GPP using EC towers during 2006 & 2007
growing seasons.
0.5
4
4
0
0
0
6
0.6
0.4
2
0
0.8
4
y = 0.46x + 0.18
R² = 0.44
2
a
EC GPP
8
6
y = 1.24x + 0.16
R² = 0.44
2
Fig 1. MODIS derived IGBP classification overlaid with EC flux
tower sites and WWF terrestrial biome boundaries
fPAR MODIS
10
a
MODIS GPP
4
D01
D02
X03
K04
K05
Acknowledgements:
Flux tower
GPPobs (1-12)
GPPobs (5-10)
2006 2007 2006 2007
40.00 24.00 33.20 21.41
77.00 27.65 69.02 22.59
49.62 50.14 42.18 43.67
45.44 39.12 40.61 30.30
30.24 25.40 21.43 21.50
The VPM model predicted the annual GPP
(GPPVPM) reasonably well in 2006-2007 at the
Duolun cropland (R2 = 0.67 & 0.71) and
Xilinhaote typical steppe (R2 = 0.80 & 0.73). The
predictive power of VPM varied in the desert
steppe, at an irrigated poplar stand (R2 = 0.74 &
0.68) and nearby shrubland in Kubuqi (R2 = 0.31
& 0.49).
VPM Model
GPPpre (1-12)
GPPpre (5-10)
2006 2007 2006 2007
69.42 29.43 68.00 28.40
72.96 49.00 71.64 48.02
59.36 70.20 57.72 68.86
74.13 28.51 69.84 26.96
86.28 19.31 81.76 18.13
MVPM
GPPpre (1-12)
GPPpre (5-10)
2006 2007 2006 2007
78.90 62.41 79.08 61.58
74.18 60.29 75.57 59.26
49.50 54.51 49.64 53.33
42.64 34.32 39.52 31.66
53.20 35.95 49.21 34.62
EVI & NDVI
Mean
annual
temperature
(˚C)
2006 2007
8.4
7.4
8.9
7.5
GPP (g C m‐2 d‐1)
Xilinhaote steppe
Kubuqi poplar
stand
Kubuqi shrubland
Altitude Mean annual
rainfall(mm)
(m)
GPP (g C m‐2 8d‐1)
Methods
Duolun steppe
Duolun cropland
Location
(decimal degrees)
VPM GPP & MODIS GPP(g C m ‐2 d‐1)
Semi-arid Inner Mongolia,
D01
P.R.C, is experiencing
D02
climate change with
X03
associated land cover/use K04
K05
change that includes an
increase in irrigated
agriculture and population
growth. We evaluate
temporal scaling up of
carbon fluxes from eddy
covariance (EC) tower
observations in different
water-limited land cover/use
and biome types.
Vegetation
GPP (g C m‐2 d‐1)
Site
The comparison between GPPEC and GPPMVPM
predicted GPP showed good agreement for the
Xilinhaote typical steppe (R2 = 0.84 & 0.70) in
2006-2007, Duolun typical steppe (R2 = 0.63),
and cropland (R2 = 0.63) in 2007. The predictive
power of MVPM decreased slightly in the desert
steppe, at the irrigated poplar stand (R2 = 0.55 &
0.47) and the shrubland (R2 = 0.20 & 0.41). The
results of this study demonstrate the feasibility of
scaling up GPP from EC towers to the regional
scale.
2
Site year
D01 2006*
2007
D02 2006
2007
X03 2006
2007
K04 2006
2007
K05 2006
2007
significance
0.642
0.000
0.014
0.000
0.000
0.000
0.000
0.000
0.013
0.000
SE
0.105
0.033
0.222
0.035
0.540
0.056
0.197
0.085
0.076
0.050
R
.008*
.527
.203
.630
.844
.702
.553
.471
.207
.415
Future Work
We are expanding our flux network in Inner and
Outer Mongolia, thus increasing the number of
measurements / tower-years. OM and IM have
contrasting socio-economic and political systems
and since, 1979 have divergent land use
trajectories with environmental changes taking
place more rapidly in IM than OM.
The regional atmospheric modeling system
(RAMS), calibrated with the EC flux tower/climate
station network & RS data is being used to
forecast future climate change scenarios in the
Mongolian plateau.
This study was supported by the NASA LCLUC Program (NN-H-04-Z-YS-005-N), Natural Science Foundation of China (30928002), the
Outstanding Overseas Scientists Team Project of CAS, and the State Key Basic Research Development Program of China (2007CB106800).