scale. These models range in complexity from regressions that
estimate association between climatic variables and the estimates
of biosphere trace gas fluxes to quasi-mechanistic models that
simulate the biophysical and ecophysiological processes. The
CASA
model was chosen from these options due to the following
factors: (a) inclusion of different formulation for
NPP
down regu-
lation, controlled by environmental stress scalars (temperature
and precipitation), (b) usage of constant versus variable maxi-
mum
LUE
for different vegetation types, and (c) the
CASA
Model
variables can be inferred from remote sensing images, which
make
CASA
suitable for estimation of
NPP
over large areas.
The
CASA
model was first developed by Potter et al. (1993)
and then expanded to incorporate the combination of ecologi-
cal principles, satellite
NDVI
data, and surface data (temperature
and water scalars) by Field
et al.
(1995). Subsequently this
model has been successfully implemented in the estimation of
regional and continental patterns of
NPP
and
CO
2
sink in other
parts of the world (Ruimy
et al.
, 1996; Goetz and Prince, 1999;
Cramer
et al
., 1999; Bian
et al
., 2010). According to Monteith
(1977),
NPP
is directly proportional to the absorption of photo-
synthetically active radiation (
APAR
) and the maximum
LUE
of
green plants. Since
LUE
is characterized as an indicator of plants
ability for fix solar energy (Bian
et al
., 2010), it is also adjusted
for spatial temporally varying stress scalars such as tempera-
ture and precipitation (or moisture stress). In contrast to the
estimation of global
NPP
, exploring the spatial distribution and
temporal pattern of national
NPP
can be beneficial for success-
fully sustainable resource management. To meet this goal, the
following questions need to be addressed: (a) What is the
NPP
of
the terrestrial ecosystems in Mongolia?; (b) Are there any tem-
poral and/or spatial patterns in
NPP
distribution?; (c) How does
NPP
react to temperature or precipitation changes?; and (d) Does
the
NPP
increment during a growing season show any spatial
patterns across a sample scale of 5-years? This study attempts
to answer these questions using the
CASA
model driven by me-
teorological data and remotely sensed vegetation parameters on
data collected in Mongolia between 2000 and 2004.
Materials and Methodology
Study Area and Materials
Mongolia experiences extreme diurnal (30°C) and annual
temperature fluctuations (min –45° to –53°C, and max. 40° to
45°C). Due to the country’s latitude (approximately 41°30'N
–51°53'N) and high elevation (approximately 516 to 4,484 m),
average annual temperatures are quite low, ranging from −7°
to −5°C in the Northwest and 2° to 4°C in the Southeast. The
total annual precipitation in the mountainous regions averages
about 400 mm, in the steppe 150 to 250 mm and in the desert-
steppe less than 100 mm (Dagvadorj
et al.,
2009). About 85 to
90 percent of the precipitation falls during the summer season
(June/August) (Shiirevdamba, 1998). The number of rainy days
decreases from north to south. There is very little precipitation
at the beginning of the growing season (generally from April to
October) but much more in the second half of the season when
cool air starts to spread across the country. This variation has
considerable effects on the growth of several spring plants.
The precipitation in summer, autumn, and winter is a source
of soil moisture but it is insufficient for vegetation to thrive
(Shiirevdamba, 1998).
The parameters used in the
CASA
model for
NPP
estimation
are satellite
NDVI
data, meteorological data such as temperature
and precipitation, and vegetation types. The multi-temporal
NDVI
images used in this study were acquired from the standard
product of
MOD
13Q1 Vegetation Indices with 250 m spatial
resolution and 16-day temporal resolution. Meteorological data
used in this study were monthly total precipitation, monthly
mean temperature and monthly total solar radiation data,
obtained from the National Agency for Meteorology, Hydrology
and Environmental Monitoring (
NAMHEM
) and monthly gridded
temperature, precipitation, and solar radiation maps at the
same resolution as
NDVI
. A land cover map (Plate 2a) displaying
water and the four types of terrestrial ecosystems, i.e., forest,
typical steppe or grassland, desert steppe, and desert was used
as the input of vegetation types in the
CASA
model. This map
was produced by Dugarsuren
et al
. (2011) using
MODIS
13
Q
1
data that was collected in the summer of 2000. The supervised
maximum likelihood classification was carried out on the veg-
etation indices and the four reflectance bands of Blue, Red,
NIR
,
and
MIR
based on the
IGBP
classification scheme (Gao and Yu,
1998) with an overall accuracy of 89.6 percent.
Procedures for the Parametric Image Generation
MODIS
products downloaded from
NASA
were in the Sinusoi-
dal projection. These original images were reprojected to the
geographical latitude and longitude projection in
WGS
84 datum
in which pixel value of the output image was determined using
nearest neighbor resampling method. The individual scenes
(originally in 10° × 10° tiles) were mosaicked and subsetted as
a seamless product covering the entire area of Mongolia. The
dynamic range of
NDVI
is from −1 to +1.
MOD
13
Q
1 250-meter VI
product was generated using the daily
MODIS
Level-2G (L2G)
surface re ectance, pointer le, geo-angle le and 1-km state le
by
LP DAAC
,
NASA
. Once all 16 days were collected, the
MODIS
VI
composite data was produced using filter-based Maximum Value
Composite (
MVC
) and/or Constrained View angle - Maximum
Value Composite (
CV-MVC
) algorithms to remove the noise that
consisted of cloud pixels and extreme off-nadir sensor views
(Solano
et al
., 2010).
Temporal temperature and precipitation data collected
from all of the 53 meteorological stations in Mongolia were
used to produce monthly average temperature and monthly
accumulated precipitation images using inverse distance
weighted method and then resampled using a nearest neigh-
bor method to generate 250-m/pixel raster images. After the
data preprocessing, the
NPP
of the study site was calculated
using the parameters demonstrated in Figure 1.
The CASA Model
In the
CASA
model, the
NPP
for a given location is determined
using the absorbed photosynthetically active radiation at the
grid
x
for a particular month
t
(APAR
x,t
, MJ m
–2
mo
–1
) and the
actual light use efficiency at the same grid and month (
ε
x,t
,
gC MJ
–1
) using Equation 1. APAR
x,t
can be determined by the
product of the incident photosynthetically active radiation
(PAR
x,t
) and the fraction of incoming PAR intercepted by green
vegetation (fPAR
x,t
). Since approximately half of the total solar
radiation (TSR
x,t
, MJ m
–2
, which is from the database of the
Institute of Meteorology, Hydrology, and Environment of Mon-
golia) is from the region of the PAR wavelength (0.40-0.70 µm),
the APAR
x,t
therefore may be determined using Equation 2.
Assuming a strong linear relationship exists between fPAR
and
NDVI
(Fensholt
et al.
, 2004), fPAR
x,t
can be solved using
the maximum and minimum
NDVI
values for each vegetation
type
i
(
NDVI
i,max
and
NDVI
i,min
) and the corresponding maximum
and minimum fPAR values using Equation 3 where fPAR
max
and fPAR
min
are independent of vegetation types and are set to
be 0.950 and 0.001 respectively based on the least and most
developed state of vegetation (Stöckli
et al.
, 2011).
NPP
(x,t)
= APAR
(x,t)
×
ε
(x,t)
(1)
APAR
(x,t)
= PAR
(x,t)
× fPAR
(x,t)
= 0.5 × TSR
(x,t)
× fPAR
(x,t)
(2)
fPAR
(x,t)
=
(NDVI
(x,t)
– NDVI
(i, min)
)(fPAR
max
+ fPAR
min
)
(NDVI
(i, max)
– NDVI
(i, min)
)
+ fPAR
min
(3)
588
July 2015
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
