The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
32
named ‘clumping’ (Chen et al., 1997). Many solutions have
been proposed to overcome this clumping bias. For example,
two new instruments have been developed to measure the
between-shoot clumping factor (Q e ): the TRAC developed by
Chen et al. (1997) and the MVI developed by Kucharik et al.
(1997). Furthermore, the boundary and illumination conditions,
data aggregation method, and sampling scheme also influence
the relative accuracy of LAI measurements. Even though
hemispherical photography was believed better than LAI-2000,
it is more suitable for trees instead of prairie regions with low
canopy vegetation.
Although LAI can be directly or indirectly measured by several
ground-based methods, difficulties in deriving it from remotely
sensed data has led to the development of various approaches
and methodologies (especially for LAI determination at
different scales and over diverse types of vegetation canopies)
(Baret and Guyot, 1991; Haboudane et al., 2004; etc.).
Estimating LAI from remotely sensed optical data can generally
be carried out by several methods (Liang, 2003): (1) through
the empirical relationship between LAI and vegetation indices
(LAI-VI); (2) through the inversion of a radiative transfer (RT)
model; (3) the use of look-up tables (LUT), (4) neural networks
(NN), and (5) a hybrid approach. Using remotely sensed
imagery, LAI can be derived from an empirical or modeled
LAI-VI relation. The major limitation of this empirical
approach is that there is no single LAI-VI equation (with a set
of coefficients) that can be applied to remote sensing images of
different surface types. Another limitation of this approach is
the sensitivity of VI to non-vegetation related factors such as
soil background properties (e.g., Huete, 1989), atmospheric
conditions (e.g., Kaufman, 1989), topography (Holben &
Justice, 1980), bidirectional nature of surfaces (Deering, 1989),
and the most important, the spatial and temporal dynamics of
LAI.
Therefore, even though recent research has attempted to
improve LAI estimates through a better description and
sampling of canopy heterogeneity (vertical and horizontal
heterogeneity, clumping, and canopy closure or gaps),
quantifying LAI with high accuracy presents numerous
challenges due to the complex spatial and temporal LAI
variations. Satellite imagery has provided promising results.
Therefore, this study will investigate the spatial and temporal
variations of LAI as well as from the measurement of NDVI.
2. METHODOLOGY
2.1 Study Area
The study area is in St. Denis Wildlife Reserve Area, 40km east
of Saskatoon, Saskatchewan, Canada. The study area is
dominated by rolling landscapes in the mixed-grass prairie
ecodistrict. St. Denis National Wildlife Area has over 200
temporary and permanent wetlands most of which are fringed
by tall grass and shrubs. Blocks of native grassland and aspen
bluffs with willow, serviceberry and chokecherry are distributed
throughout the Wildlife Area. Almost one-half of the previously
cultivated land has been seeded to bromegrass and alfalfa for
nesting cover. The relatively large amount of existing cultivated
land is used for research on the effects of agricultural practices
on waterfowl production.
2.2 Field Data Collection and Satellite Imagery Acquisition
Field data were collected along one transect with 128 samples
with 4.5m interval three time during the growing season of
2007. Variables collected include LAI with LAI-2000 plant
canopy analyzer, reflectance with ASD handheld
spectraradiometer, estimated cover, and digital pictures. SPOT
4 multi-spectral 20m resolution imagery was acquired at
monthly interval in the summer of 2007, which match with field
data collection. Five SPOT scenes were from May, June, July,
August, and September respectively. Images were
geometrically, radiometrically, and atmospherically corrected.
Normalized Difference Vegetation Index (NDVI) was
calculated and LAI values were derived from the satellite
imagery.
O £6 Km
Figure 1. Study area: St. Denis, Saskatchewan, Canada.
2.3 Data Analysis
Normalized difference vegetation index (NDVI) was derived
from ground measured reflectance and SPOT satellite imagery.
NDVI was calculated based on the ratio of the difference
between near infrared and the sum of these two bands.
Wavelet analysis was performed on LAI and NDVI derived
from ground measurements in July as this is the maximum
growing season in Canadian prairies.
3. RESULTS
3.1 Spatial Variation of LAI and NDVI
Wavelet analysis indicated that LAI has several levels of
variations: 4.5m, 12m, and 22.5m. NDVI spatial variation was
mostly corresponding with LAI. The variations are at 4.5m,
12m, 18m, and 22.5m. Clearly, the 18m variation from NDVI
was not found from LAI, indicating that LAI is not the reason
for the variation on NDVI. It might caused by topography (He
et al., 2007). Future analysis is necessary.
3.2 Temporal variation of LAI and NDVI
Figure 3 demonstrates the temporal change of the study area.
July is the maximum growing season for Canadian prairies.
Vegetation greens up in May and senescent in September. The
SPOT images are in standard false color composite (RGB: Near
infrared, Red, and Green). Different tone of red indicates
healthy and dense vegetation, while blue/green is bare ground.