Geoscience Reference
In-Depth Information
The orientation of the camera can be assessed by noting the position of the large tree bole that
originates from the five-o'clock position in the image. The April 9, 2002, image places the bole
closer to the 4:30 position. However, as mentioned previously, camera orientation does not affect
whole-image calculations of LAI or canopy openness. Orientation was important only if it became
necessary to match TRAC data with a particular sector of the hemispherical photograph.
L
values derived by hemispherical photography increased over the course of phenological
development at the Hertford site. A decrease in L
e
from 2.13 to 1.88 was observed between the
July 25 and August 5, 2002, images. The decrease may have partly been a result of the understory
removal operation that occurred between July 25 and 30, 2002. However, decreases of this mag-
nitude were observed at other APB sites in mid to late summer, when no understory canopy removal
was performed. The Hertford site was primarily coniferous forest. Needle loss due to the extreme
drought conditions experienced in the APB study area may partially account for the observed
decrease in L
e
.
e
4.4.3
Satellite Remote Sensing Issues
spatial resolution. Inherent in this product
were a number of spatial factors that may contribute to uncertainty in the final accuracy of this
analysis. MODIS pixels were nominally 1 km
The MODIS LAI product was produced at 1-km
2
at nadir but expanded considerably as the scan
moved off nadir toward the edges of the 2330-km-wide swath. As a result, off-nadir pixels sampled
a larger area on the ground than near-nadir pixels. The compositing scheme partially compensated
for this by preferentially selecting pixels closer to nadir. Mixed pixels contained more than one LC
type. In the APB study region, the landscape exhibits varying degrees of fragmentation, producing
a mosaic of parcels on the ground. Within a 1-km
2
block, agricultural, urbanized, and forested LC
types may be mixed to such a degree that assigning a single LAI value is questionable. There were
also angular effects to consider. The NDVI and LAI products were adjusted for the bidirectional
reflectance distribution function (BRDF; MODIS product MOD43). Still, angular effects produced
by variable viewing geometry may have degraded the accuracy or interpretability of the results.
An important issue was that of spatial scaling from
2
in situ
reference data measurements (m
)
2
to MODIS products (1 km
). ETM
data provided the link between
in situ
measurements and
2
+
MODIS measurements. Quadrants correspond to a ground region of a 3
¥
3 ETM
pixel window
+
(90
60 m).
The Landsat data were precision-registered to ground coordinates using ground control points,
providing an accuracy between 0.5 and 1 pixel. Once the
¥
90 m), and the subplots correspond to a region of approximately 2
¥
2 pixels (60
¥
in situ
data and Landsat image were
coregistered, an ETM
LAI map was generated by establishing a regression relationship between
+
in situ
LAI and Landsat NDVI. Then, the Landsat LAI map could be generalized from a 30-m
resolution to a 1-km
resolution. Overall accuracy was influenced by accuracy of coregistered data
sets, interpolation methods used to expand
2
in situ
measurements to ETM
NDVI maps, and the
+
spatial coarsening approach applied to scale the ETM
imagery to the MODIS scale of 1-km
pixels.
+
2
4.5 SUMMARY
Research efforts at the U.S. Environmental Protection Agency's National Exposure Research
Laboratory and National Center for Environmental Assessment include development of remote
sensing methodologies for detection and identification of landscape change. This chapter describes
an approach and techniques for estimating forest LAI for validation of the MODIS LAI product,
in the field using ground-based optical instruments. Six permanent field validation sites were
established in the Albemarle-Pamlico Basin of North Carolina and Virginia for multitemporal
measurements of forest canopy and biometric properties that affect MODIS NDVI and LAI prod-
