International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B-YF. Istanbul 2004
Rather high spatial resolution map of the reference LAI surface
” was rescaled to 1km cell size to compare with the 1km MODIS
LAI map. Figure 4 compares the MODIS LAI and the reference
LAI maps for the whole study area of 28 x 38km. It appears
that the MODIS LAI estimates are higher than the reference
LAI surface. From our fieldwork and experience, the LAI over
the study area showed little variation, at least within the forest.
The MODIS LAI map shows highly variable spatial pattern of
LAL
Figure 4. Comparison between the MODIS LAI (left) and
the reference LAI (right) maps.
Table 2. Simple statistics of MODIS LAI and reference LAI
Fact Class Reference MODIS
Beno = Mean | Std. | Mean Std.
Land Whole area 3.26 0.24 3.52 1.81
an Forest ans [624] 374 [176
cover Grass
: | 53
type & Cropland 3.10 0.17 2.30 LS
RT model 327 0.24 3.40 ].77
Model: | Emprcd | 315 Lael sof om
model
-C D : 3. 67
Cloud Non-cloud 3.25 0.22 41 1.6
Cloud 3.20 0.28 3.81 2.10
The mean LAI value of the MODIS product is about 0.38
higher than the reference map (Table 2). Although the mean
LAI values of whole study area are similar between MODIS
LAI and reference LAI, the spatial variation and pattern of the
LAI values are quite different between two maps. When the
two LAI maps are compared by vegetation type, the MODIS
LAI values were larger than the reference value in forest. On
the other hands, grass and cropland showed the under-
estimation in which the MODIS LAI value was lower than
reference LAI value.
The discrepancy between the MODIS LAI and the reference
LAI could be explained by several factors that were the
accuracy of the input datasets, estimating algorithms and land
cover types. Since the MODIS LAI values are estimated by the
seven cover types of different biomes, it is worthwhile to
examine the MODIS land cover map used for the production of
the MODIS LAI. Figure 5 compares the MODIS land cover
product and the reference land cover map. The classification
accuracy of the MODIS land cover product is 44.72% as
assessed by the reference land cover map (Table 3).
The study area includes large forest areca. The MODIS land
cover product didn’t show well the small water-body,
needleleaf forest, and urban. It was obvious that some portion
of forests were misclassified into grass and cropland on the
MODIS land cover product. Water-body was misclassified to
needleforest and the urban area was misclassified into grass and
cropland. These misclassifications were caused by low spatial
resolution of MODIS land cover product and might be the
causes of the difference between two LAI maps.
ol . = Le
A Les
a-a 42
br. "=
Shrub B Savannah
Needleleaf forest
Water
? Urban
Grass &Crop
Broadleaf forest EH
Figure 5. Land cover map of the MODIS product (left) and
the reference data (b)
Another factors influenced on the MODIS LAI estimation were
the cloud effect. Although the MODIS LAI product is based on
the eight-day maximum value composite to remove/reduce the
Table 3. Classification accuracy of the MODIS land cover assessed by the ETM+ classification (Overall accuracy = 44.72% )
Reference MODIS Water Ga AR Shrubs Savannah Brod Ned Urban Total
Water 0 4 0 0 6 4 0 14
Grasses & Crop 0 71 0 2 16 16 0 105
Shrubs 0 0 0 0 0 0 0 0
Savannah 0 0 0 0 0 0 0 0
Broadleaf forest 0 311 2 0 438 85 2 838
Needleleaf forest 0 98 0 I 60 36 0 195
Urban 0 53 2 14 4 5 78
Total 0 537 4 3 534 145 7 1230
30
