tellite 
large 
lution 
final 
a are 
t can 
IStrial 
gery. 
nship 
nand, 
matic 
ex of 
bidity 
2ases 
green 
s are 
eters 
d by 
MEASUREMENT OF DYNAMIC GEOLOGIC PROCESSES AT SUBPIXEL SCALES 
Robert E. Crippen 
Ronald G. Blom 
Jet Propulsion Laboratory 
California Institute of Technology 
Pasadena, California 91109 USA 
Commission VII, Working Group 4 
KEY WORDS: Dynamic, Geology, Remote Sensing, Subpixel Measurement, SPOT, Image Matching 
ABSTRACT 
Methods have recently been developed for the utilization of remotely sensed image data in the measurement of 
terrain displacements resulting from geologic processes. In optical imagery, measurements precise to a fraction of 
a pixel are achieved by statistical image matching. In radar imagery, measurements precise to a fraction of a 
wavelength are achieved by interferometry. Each method has distinct advantages. Radar interferometry is 
currently more high developed, but the advent of globally available, one-meter optical satellite images will greatly 
increase the utility of optical methods. 
INTRODUCTION 
Terrain displacements related to earthquakes, sand 
dune migration, volcanic activity, glacial motion, and 
gravitational sliding can be measured with precisions 
finer than image resolution by optical remote sensing 
methods as well as by radar interferometry. 
Applications to date have shown both methods to be 
uniquely valuable in the detection, mapping, and 
measurement of geologic and environment processes. 
The two methods also have differing strengths and are 
thus complementary. 
METHODS 
Optical methods involve image cross correlation of 
multitemporal images (Crippen, 1992). A 'before' image 
is used as a reference base upon which a grid is 
delineated. At each node, a neighborhood of pixels 
(e.g. 100 x 100) is sampled and compared to pixels at 
the corresponding location in an 'after' image. The peak 
subpixel correlation point is determined by interpolating 
the 'after' image repeatedly, following the path of 
increasing correlation. This point defines a vector 
relative to the node in the 'before' image reference 
base. By calculating a vector at each node, an evenly 
spaced array of vectors is generated for the entire 
image. Typically, the dominant pattern shown in the 
vector array corresponds to satellite attitude 
differences between the two scenes. However, this 
pattern can be modeled, estimated, and removed 
because it is consistent across the scene, differing 
greatly in spatial frequency from the terrain 
displacement patterns we seek to reveal. 
Radar interferometry has been extensively 
demonstrated and well documented in recent years 
(e.g. Massonnet et al., 1993; Peltzer and Rosen, 1995). 
Measurements require a multitemporal pair of images 
plus an elevation data base (which may also be derived 
by radar interferometric means if an appropriate third 
radar image is available). The measurement is derived 
from the radar phase information, which is independent 
159 
of the radar backscatter measurements usually 
displayed in a radar image. 
COMPARATIVE ADVANTAGES 
Optical methods are two-dimensional, potentially 
providing a complete mapping of both horizontal 
dimensions (assuming the scenes are nadir looking). 
However, they provide no sensitivity to vertical 
displacements Radar interferometry is one- 
dimensional but can detect vertical displacements and 
some horizontal displacements because 
measurements are along the oblique 'slant' path of the 
radar beam. Optical methods are most reliable in 
rugged terrain where image patterns are strong. Radar 
interferometry is most reliable in low relief areas, where 
problems such as layover cannot occur. Both methods 
suffer from temporal decorrelation, which results from 
environmental changes in the scene (e.g., vegetation 
growth). Optical methods require a cloud-free 
atmosphere and consistent sun angles, which are 
irrelevant factors for radar interferometry. Because 
radar interferometric measurements are made relative 
to signal phase, they do not provide absolute 
measurements. Absolute measurements can be made 
only by observing spatial gradients from a known (or 
presumed) value at a geographic reference point. 
Confusion can occur where spatial gradients are too 
steep. In contrast, optical methods provide direct : 
measurements. 
Currently, radar interferometry provides measurement 
precisions on the order of a few centimeters (i.e., on the 
order of a tenth of the signal wavelength). Optical 
methods using spaceborne imagery can measure only 
meter-scale displacements at best (e.g., a tenth of a 
pixel using SPOT panchromatic data). Optical methods 
are limited not only by the spatial resolution of the data 
but by the radiometric resolution as well. If the 
radiometric quantization steps (DNs) do not differ 
substantially from the local image variance, relatively 
precise interpolations are not possible. This is why 
optical methods work best in rugged (shaded) terrain, 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996