PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
September 2020
525
SECTOR
INSIGHT:
.
edu
E
ducation
and
P
rofessional
D
evelopment
in
the
G
eospatial
I
nformation
S
cience
and
T
echnology
C
ommunity
Barry Haack,
George Mason University
Integration of Varied Spatial Resolution Data
Remote sensing has and continues to change extremely
rapidly. Those changes have included varied platforms, new
sensors, and improved data processing methods. With the
current availability of many imagery types, this column
urges renewed educational emphasis on the integration of
varied spatial resolution data for improved spatial analysis.
The changes of platforms and sensors is evident in the
resolutions of imagery available to scientists and decision
makers. Spectral resolution has expanded by the acquisition
of hyperspectral imagery with some sensors providing 512
bands. Temporal resolution has always been frequent with
meteorological sensors but is now high with fine spatial res-
olution systems. The small-sat constellation of the commer-
cial company Planet is able to acquire global imagery daily
with very fine spatial resolution. Radiometric resolution has
increased from the 6 and 7 bits of early Landsats to 12 and
16 bit imagery today.
One of the major changes in resolution has been the avail-
ability of fine spatial resolution imagery from satellites.
Initially the community thought that this imagery would
replace aerial photography but the high initial image costs
did not make that viable. However, there is now fine spatial
resolution satellite imagery available at little or no cost.
However, as with aerial photography, fine spatial resolution
satellite imagery has generally small footprints. The imag-
ery is often 20 km or less per side making it very difficult to
acquire and accurately extract information over large areas.
Most watersheds and local governmental administrative
units would require hundreds of images.
Analysis of medium and coarse spatial resolution satellite
imagery (10 m pixels and greater) is an effective way to
assess regional, continental or global phenomena because of
its synoptic coverage, frequency of image acquisition, large
footprint and often inexpensive cost. The disadvantage of
medium spatial resolution imagery is the lack of detail.
Biases often occur because of the large pixel size of the data.
Less common surface features are often underestimated,
causing a negative bias, and more common components can
be overestimated, causing a positive bias.
The purpose of this column is to establish interest among
remote sensing educators in the integration of imagery at
different spatial resolutions to provide improved spatial sta-
tistics for multiple applications. Many remote sensing scien-
tists believe that the most important information that they
can provide are accurate maps. However, the reality often is
that many decisions are made based upon statistical analysis
such as the extent and rate of wetland loss, deforestation or
urban expansion.
Regression Estimation
Regression estimation is a statistical sampling technique to
combine the synoptic coverage of a coarse spatial resolution
sensor with the improved detail of a finer spatial resolution
sensor (Gallego, 2004). This technique requires the analyst
to map phenomena rst using coarse spatial resolution
imagery. Next, ne spatial resolution images are acquired
for samples within the study area. Phenomena are likewise
mapped with the ne spatial resolution data. A regression
analysis or other statistical procedure is then performed to
determine a correlation or relationship between the two sets
of data. If a good correlation exists, the more accurate, finer
spatially detailed imagery can be used to calibrate the coarse
spatial imagery using a correction factor (Nelson, 1989). In
this manner, regression estimation provides more accurate
statistical information than only using coarse imagery.
This column was prompted in part based upon a review of 12
textbooks in remote sensing to determine if they referenced
regression estimation or similar approaches for surface
inventories. Interestingly, only two had any reference to the
method and they were quite dated, thus the concern that the
procedure is not commonly included in curriculums.
Applications
There are numerous examples of regression estimation. They
generally employ imagery of different spatial resolutions
and footprints but the method can also use field data rather
than fine spatial resolution imagery. It is also possible to
Photogrammetric Engineering & Remote Sensing
Vol. 86, No. 9, September 2020, pp. 525–526.
0099-1112/20/525–526
© 2020 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.86.9.525
