Author: Leslie Goldman
© 2013 National Ecological Observatory Network, Inc. All rights reserved. The National Ecological Observatory Network is a project sponsored by the National Science Foundation and managed under cooperative agreement by NEON, Inc. This material is based upon work supported by the National Science Foundation under the following grants: EF-1029808, EF-1138160, EF-1150319 and DBI-0752017. Any opinions, findings, and conclusions or recommendations
expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.
The National Ecological Observatory Network (NEON)
is constructing an Airborne Observation Platform
(AOP) to provide high resolution LiDAR, aerial, and
hyperspectral data for NEON sites across the United
States. This includes the NEON imaging spectrometer
(NIS) that provides more than 420 bands of high
resolution data across a spectral range of 380 nm to
2,510 nm. AOP will provide a collection of normalized
difference vegetation indices (NDVIs), which use data in
the red and
near infrared
portions of
the electro-
magnetic
spectrum to
estimate
vegetation
density and
health. A
challenge
in deriving
NDVIs with
hyperspectral sensors is band selection and comparability
to other sensors. This study analyzed data collected in
2013 from the San Joaquin Experimental Range, CA.
Background
Contact Information:
www.neoninc.org
General Sensor Comparison
Internship Project: A comparative analysis of Landsat, AVIRIS, and NIS normalized
difference vegetation indices in Domain 17, the Pacific Southwest
Madeleine Ball (Senior, Tufts University) Mentors: Nathan Leisso (AOP), Shelley Petroy (Data Products), Leah Wasser (Education)
1. NIS NDVI is sensitive to spectral band selection.
Therefore, it is important to specify bands selected for
use in the NDVI calculation to minimize variation across
instruments
2. Spatial resampling is required to effectively compare
high spatial and spectral resolution NIS data to lower
resolution sensors such as AVIRIS and Landsat.
3. Subsequent studies could further validate results by
analyzing other NEON sites with different vegetation
types and by incorporating in-situ data
4. In the future, data users can use this information to
understand the differences between sensors and make
informed decisions when selecting data sources
Conclusions and Future Directions
Figure 2. Images demonstrating the differences in
spatial resolution between 1) Landsat at 30m
resolution, 2) AVIRIS at 14.8 m resolution, and
3) NIS at 1 m resolution
Figure 6. NDVIs from 1) NIS (bands 62 and 94), 2) AVIRIS (bands 36
and 53), and 3) Landsat (bands 5 and 4). They are challenging to
compare due to spatial and spectral resolution differences.
NIS Spectral and Spatial Resampling
Landsat AVIRIS NIS
Spectral
Resolution
7 bands
224 bands
426 bands
Spatial
Extent
7501 x 7311 pixels
225 x 219 km
1156 x1175 pixels
17.11 x 165.4 km
936 x 7172 pixels
.936 x 7.172 km
1
2
3
Objective 1 Results: NIS NDVI Band Selection
Figure 4. Matrix demonstrating band selection
process for the NEON Imaging Spectrometer (NIS)
To determine an optimal band
combination for deriving NIS NDVIs,
matrices were created (Figure 4).
Results showed the greatest contrast
between areas of high NDVI returns
(healthy vegetation) and areas of low
returns (unhealthy vegetation or a
lack of vegetation) in the upper left
quadrant. This indicates NDVI
sensitivity to hyperspectral band
selection. Highest vegetation
contrast is obtained in the upper
left quadrant.
Low
Band 58 (.668 µm)
Mid
Band 62 (.688 µm)
High
Band 66 (.708 µm)
Low
Band 90
(.828 µm)
High
Band 98
(.868 µm)
Mid
Band 94
(.848 µm)
Red Wavelengths
Near-Infrared Wavelengths
Objective 2 Results: Comparing NIS, Landsat, AVIRIS NDVIs
To compare with Landsat, NDVI
outputs from NIS and AVIRIS were
spatially resampled on a 30 m grid
(Figure 5). Spatial resampling
decreased sensitivity of NDVI to
hyperspectral band selection and
decreased differences between NIS,
AVIRIS and Landsat. This
demonstrates that when comparing
with other sensors, users should
spatially and/or spectrally resample
NIS data (see flowchart below).
Figure 5. Matrix of spatially resampled NIS and AVIRIS
NDVI maps overlaying Landsat NDVI.
Low
Band 58 (.668 µm)
Mid
Band 62 (.688 µm)
High
Band 66 (.708 µm)
Low
Band 90
(.828 µm)
High
Band 98
(.868 µm)
Mid
Band 94
(.848 µm)
Red Wavelengths
Near-Infrared Wavelengths
1. Determine NDVI sensitivity to NIS spectral band
selection
2. Compare NIS NDVI to Landsat and AVIRIS
Figure 1. Images collected in 2013 from the SJER site:
a) NEON site location, b) 2013 Landsat RGB bands
4,3,2; AVIRIS RGB bands 29,20,12 and c) NEON
Imaging Spectrometer (NIS) RGB bands 52,34,19
Table 1. Shows the inherent spectral resolution and
spatial extent differences between the Landsat,
AVIRIS and NIS sensors
NIS NDVI products cannot be directly compared to Landsat and AVIRIS due to significant differences in spatial and spectral resolution. NIS data
can be spectrally and spatially resampled to match the resolution of Landsat and AVIRIS to make the NDVI product comparable between sensors.
Incomparable images:
What are NDVIs?
=
+
Eq. 1
Normalized Difference Vegetation Indices
(NDVIs) are used to characterize vegetation, an
important aspect of understanding ecosystems.
NDVIs are derived using a ratio (Equation 1)
between reflectance in the red and near infrared
regions of the electromagnetic spectrum. This
ratio highlights areas of photosynthesizing
vegetation. NDVI values will be closer to 1 in
areas of healthy vegetation and closer to 0 in
areas with unhealthy vegetation or a lack of
vegetation.
Figure 3. NIS true color image with NIS NDVI
overlaid. Green trees correctly appear white in
the NDVI whereas dead grass appears black.
2
1
3
a
b
c
Resampling NIS (1 m resolution, 426 bands) to Landsat 8 (30 m resolution, 7 bands):
Original NIS
Spectral Resampling
To Landsat
NDVI
Spatial Resampling
To Landsat
Final Resampled NDVI
Comparison
Landsat NDVI
Spectral Resampling
To AVIRIS
NDVI
Spatial Resampling
To AVIRIS
Final Resampled NDVI
Comparison
Original NIS
AVIRIS NDVI
Resampling NIS (1 m resolution, 426 bands) to AVIRIS (14.8 m resolution, 224 bands):
Objectives
