skip to Main Content

Satellite Radar and GIS Volunteers Detect Hurricane Sandy Damage

Sang-Ho Yun1&3*, Alessandro Coletta2, Eric Fielding1, Shoreh Elhami3, Tom Farr1&3, Donald Ferguson4&3, John Helly5&3, Richard Butgereit6, Frank Webb1, Paul A. Rosen1, Mark Simons7, Hook Hua1, Susan Owen1


1 Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA.
2 Italian Space Agency, Rome, Italy.
3 GISCorps Volunteers.
4 US Department of Energy, Morgantown, WV.
5 UC San Diego, San Diego, CA.
6 Florida Division of Emergency Management, FL.
7 California Institute of Technology, Pasadena, CA.


In response to Hurricane Sandy’s devastation on the East Coast of the United States, we produced a damage proxy map of New York City, the place with the highest casualty, using X-band radar data from COSMO-SkyMed satellites and implemented visual inspection on each pixel of the damage proxy map for accuracy. The radar data processing was done with ROI_PAC and the damage proxy map algorithm, both of which were developed at the Jet Propulsion Laboratory and California Institute of Technology. The product contains red (damage) and transparent (no damage) pixels for simplicity. For validation, we overlaid the damage proxy map on a Google Earth or ArcGIS Online base map and imported NOAA’s aerial photographs that were provided by the US Geological Survey’s Hazards Data Distribution System archive. We then compared the damage proxy map, pre-Sandy high-resolution optical images, and the post-Sandy NOAA’s high-resolution optical images, and marked each red pixel of the damage proxy map either as correct detection or false alarm. The damage proxy map algorithm produced about 0.8 percent of red pixels mostly along the coastline, representing changes on the ground. The validation marked each red pixel as correct detection if it appears to be caused by Hurricane Sandy, and as false positive if it seems to be anthropogenic change or change in natural terrain. With no dedicated system or personnel prepared for this major disaster, we were able to produce the damage proxy map on Day 11 and validation products of it on Day 15. This involves manual data discovery, manual data request (via Emails), manual data downloading (via ftp file transfer), and manual coordination of work (via teleconference and Emails). The results were encouraging and demonstrated great potential of future automated system. Given the data acquisition latency was 5 days for strip-map mode data of New York City, an automated future disaster response system would have produced the products within a week.


Hurricane Sandy severely damaged parts of the Caribbean and the Mid-Atlantic and Northeastern United States in late October 2012. The storm was classified as Category 2 at its peak intensity and became Category 1 offshore near the Northeastern United States. Around 8 PM local time (EDT) on October 29, Sandy made landfall near Atlantic City, New Jersey [1]. As measured by its size, being approximately 1,800 km in diameter, Sandy was the largest Atlantic tropical storm on record [2], [3]. Preliminary estimates of losses due to damage and business interruption are estimated at $65.6 billion (USD), which would make it the second-costliest Atlantic hurricane, behind only Hurricane Katrina [4]. At least 253 people were killed along the path of the storm in seven countries, including 131 deaths across 8 states in the U.S., and 43 of them were in New York City [5]. Due to its large extent Sandy’s integrated kinetic energy was greater than Hurricane Katrina at landfall [6]. The large amount of energy, along with the high tide, induced a higher than expected storm surge. In addition, New York’s topography/bathymetry amplified the storm surge at Staten Island, where about half of the deaths in New York City happened with forceful wind and 4.3 meters of storm surge [7], [8].

We implemented a response activity to demonstrate the current capability and future potential of synthetic aperture radar (SAR) sensors and an automated system for disaster response. We ran an end-to-end practice to identify components that need to be improved to reduce the latency in future responses. Scientists and Engineers at the Jet Propulsion Laboratory (JPL) and California Institute of Technology (Caltech) have been working on developing algorithms and a system for rapid response to natural disasters using geodetic observations as well as seismic data. These efforts have been made under Advanced Rapid Imaging and Analysis (ARIA) projects. ARIA team collaborated with the Italian Space Agency (ASI) and GISCorps to implement damage mapping and validation.

We produced a Damage Proxy Map (DPM) of New York City that shows where most of the damage occurred. The DPM was generated with an algorithm developed under the ARIA project [9], using the radar data acquired from the Italian satellite constellation COSMO-SkyMed. The DPM was validated with aerial photographs produced by the National Oceanic and Atmosphere Administration (NOAA). This report is meant to record a brief history of our response activity, in order to help identifying pain points, guide ARIA’s current algorithm/system development, and promote future improvement in the response effort in general.


The timeline of our response activity and relevant events is summarized in Table I. Immediately after Sandy’s landfall, ASI acquired a number of SAR data of the affected areas, including New York City, from its four identical COSMO-SkyMed (CSK) satellites. Each of the satellites has repeat cycle of 16 days – the satellite visits the same spot on Earth every 16 days, so the nominal revisit time of the four satellites is 4 days. They also acquired wide swath SAR image for fast coverage of a larger area. Figure 1 shows the swath footprints of the SAR scenes acquired by CSK satellites overlaid on a Google Earth basemap. The large rectangles are wide swath data and the narrower rectangles are strip-map mode acquisitions. In this study, we used one of the strip-map mode acquisitions covering New York City. Fortunately, the satellites had been acquiring data of New York City before Sandy’s visit, which made it possible to produce the damage proxy map.

NOAA started acquiring aerial photographs from three days after Sandys landfall. NOAA flew aircrafts at about 1,500 m above the disaster areas, producing thousands of aerial photographs at about 17 cm resolution [10]. On Day 5, COSMO-SkyMed acquired the first post-event strip-map mode radar data over New York City. A week after the disaster, the ARIA team encountered a posting on the Twitter looking for volunteers to interpret SAR images. ARIA responded to the tweet and volunteered to offer SAR image interpretation service. This initiated conversation between ARIA and the GISCorps.

The Urban and Regional Information Systems Association (URISA) is a nonprofit association of professionals using geographic information systems (GIS) and other information technologies to solve challenges at all levels of government. URISA’s GISCorps ( coordinates short-term volunteer GIS services to underserved communities worldwide. Since its inception in October 2003, the Corps has attracted +/- 3,000 volunteers from over 94 countries. To date, GISCorps has deployed 410 volunteers to 112 missions in 46 countries. These volunteers have contributed over 13,000 working hours in disaster and non-disaster response missions including in the United States, Latin America, Asia, Eastern Europe, and Africa.

The GISCorps learned about ARIA’s plan to use CSK data to produce DPMs, and agreed to provide assistance in validating the DPM product. ARIA then discovered the availability of the data on Day 8 through ASI’s e-geos web catalog and issued a data order. On Day 9, ASI delivered the requested CSK data. On Day 11, a DPM was produced with InSAR + DPM processing and was delivered to the GISCorps. After discussions with ARIA for directions, optical data handling, and conversion of DPM raster image into a GIS layer, three volunteers through the GISCorps delivered their DPM validation results on Day 15.

Figure 1. Swath footprints of COSMO-SkyMed data acquisition following the landfall of Superstorm Sandy

The collaboration between ARIA and GISCorps began by chance. Responding to Hurricane Sandy, Chris Vaughan, the Federal Emergency Management Agency (FEMA) Geographic Information Officer, requested the activation of the International Charter, and Brenda Jones, the Emergency Response Coordinator for USGS, put out a call for project manager. Richard Butgereit, the GIS Administrator at the Florida Division of Emergency Management, answered to the call and served as a project manager for the International Charter activation for Sandy. One of the biggest challenges for the International Charter is to get what the Charter calls “value added resellers” to process the data and turn it into something actionable for emergency responders. Hence, Richard reached out to GISCorps for assistance in the evening of 11/5 (Day 7). Then Shoreh Elhami, the founder of the GISCorps called for volunteers through their routine recruitment process by email blast and also broadcasted it via the social media including the Twitter. Mika MacKinnon, a field geophysicist, relayed the call by posting it on her Twitter at 9:53 AM on 11/6 (Day 8). Eric Fielding at the Jet Propulsion Laboratory (JPL) saw the Twitter posting about the GISCorps and emailed it at 12:19 PM to Sang-Ho Yun at JPL, who answered to Shoreh’s call at 1:39 PM on the same day. While Sang-Ho was recruited via the call on Twitter, three other volunteers (Donald Ferguson, John Helly, and Tom Farr) were selected via the regular recruitment process, which included a phone interview.

This response activity was made possible with five major components: (1) Pre- and post-disaster COSMO-SkyMed radar data and ASI’s rapid release, (2) Pre-disaster optical imagery through Google Earth and ArcGIS Online, (3) Rapid release of post-disaster NOAA aerial photographs through USGS HDDS archive, (4) Damage proxy map algorithm and radar data processing, (5) Volunteering of GIS experts. However, these components were not pre-determined. Through emails, in-person discussions, and teleconferences, we have explored multiple options at each stage and tried to choose an optimal solution.

A. Algorithm

Interferometric coherence, a measure of similarity between two radar echoes, has been used to estimate the quality of InSAR data. This statistical quantity is calculated as:

where c1 and c2 are complex pixel values of two SAR images coregistered, and < > denotes ensemble average, an average over multiple realizations of a random variable, but practically calculated with spatial average. Coherence from repeat pass interferometry provides a quantitative measure of ground surface property change during the time span of the interferometric pair. Major damage to buildings significantly increases the interferometric phase variance in the impacted resolution element. This change shows up as decorrelation, or a decrease in coherence. InSAR coherence was used to detect ground surface change due to lave flow on Kilauea volcano observed with Space Shuttle Imaging Radar-C (SIR-C) [11]. Decorrelation from ERS radar data was used for mapping surface rupture and ground surface disturbance from intense shaking during the 1999 Mw 7.1 Hector Mine earthquake [12]. Coherence differencing and ratioing techniques were introduced for mapping surface ruptures and damage to the city of Bam, Iran from the 2003 Bam earthquake [13], [14]. Those methods required special imaging geometry and timing conditions. We improved the method by eliminating the requirement for special conditions, and it is now applicable to much broader spectrum of disasters [9]. We used the improved method in this response activity.

B. Results

Figure 2 shows the damage proxy map of New York City derived from COSMO-SkyMed data listed in Table II. The red solid frame is the boundary of our analysis, slightly smaller than the extent of the CSK SAR image.

Within the frame there are red pixels and transparent pixels. The red pixels represent the top 0.8% anomalous changes during the time period of October 22, 2012 – November 3, 2012 compared to the time period of October 18, 2012 – October 22, 2012. The data were provided in single-look SAR image format by the Italian Space Agency. We used the ROI_PAC, the InSAR software package developed at JPL/Caltech [15] for InSAR processing and the damage proxy map algorithm developed at ARIA project at JPL/Caltech [9]. The top 0.8%, or 99.2 percentile, threshold was selected when red pixels started showing up on the sea surface, as we decreased the threshold.

The blowup of the southern part of Figure 2 is shown in Figure 3. Although ARIA’s damage proxy map algorithm can produce a pixel value of full floating point range of [-1 1] or [0 1], this damage proxy map has binary pixel values – either zero or one, to keep the validation process simple. The damage pixels are distributed along the coastline in Staten Island, Brooklyn, and Breezy Point. There are other groups of red pixels along the water and on the docks at harbors, where human activities are expected to have caused changes significant enough to be detected (See Figure 7).


The threshold of the damage proxy map was selected when the red pixels start appearing on the sea, which we know are not signals. However, this may not be an ideal way of choosing the threshold value. Thus, we validated each red pixel in the damage proxy map with visual inspection, comparing pre- and post-event high-resolution optical imagery. The GISCorps provided three GIS expert volunteers for validation of damage proxy map pixels. We explored multiple options to pick the best validation process and optical imagery.

First, we made the damage proxy map a binary image, where either damage or non-damage are indicated with red and transparent pixels, respectively. This made the validation effort simple. We only handled the red pixels in the damage proxy map. Visiting each red pixel, we marked it as either correct detection (True Positive – TP) or false positive (FP). Note that starting with the damage proxy map, we can reduce the number of pixels to investigate from about 1.5 million down to about 14 thousand pixels. We did not determine true or false negatives, however, for this response activity. For validation, we needed optical imagery to work with, and we tried the following images.


We first tried using the Advanced Spaceborne Thermal Emission and Reflection (ASTER) visible to near IR images (nominally at 15 m posting) that were acquired from the Terra satellite on 2011-09-01 (pre-Sandy) and 2012-11-04 (post-Sandy, see Figure 4a). The one-year separation was chosen to get as close to the same lighting conditions as possible. The images were converted from UTM to lat/lon projection and then truncated to make it easier to register to Google Earth. However, the resolution was too low to be useful for validation. A simple difference of the pre- and post-Sandy images did not produce useful signal for damage detection (Figure 4b).

B. WorldView-1

We then tried using the WorldView-1 optical image (50 cm posting) that was acquired on 2012-11-03 as a post-event scene. Figure 5 shows the WorldView-1 image mosaicked, with file size being approximately 2 GB, and overlaid on Google Earth. This overlay of a large image on Google Earth was done through Google’s Earthbuilder.

Figure 2. Damage proxy map of New York City derived from COSMO-SkyMed data, overlaid on Google Earth.

Unfortunately, the image contained a non-trivial portion of clouds and shadows of the clouds, obscuring the coastlines. There were times when we were confused whether the WorldView image was acquired before or after the devastation, because major damage was concentrated along the coastline, and no obvious sign of damage was seen in the ground under clear sky.

C. NOAA Aerial Photographs

NOAA aerial photographs were the first comprehensive set of optical imagery that turned out to be useful for validation. The high-resolution (17 cm posting) images were made available through the USGS HDDS archive. There was also an online interface where all the NOAA images were overlaid on a base map. However, we wanted to overlay the image on top of a base map that shows a high-resolution pre-Sandy images and the damage proxy map, so we can control the three layers – (1) damage proxy map, (2) pre-Sandy high-resolution optical image, and (3) the NOAA’s post-Sandy optical image – and compare the “before” and “after” images for each red pixel of the damage proxy map. Thus, we decided to download all the images and overlay on our choice of base maps. However, there were over 2400 links, each of which would download one image, spread out over more than hundred webpages. We learned, through instructions provided by USGS, how to implement bulk order up to 250 images at a time.

Figure 3. Damage proxy map of New York City derived from COSMO-SkyMed data, overlaid on Google Earth. The red pixels represent top 0.8 percent anomalous decorrelation

D. Conversion between Remote Sensing and GIS (Raster and Vector Image)

For validation tasks, the damage proxy map, the pre-Sandy image, and the post-Sandy image should be registered to each other. They have to be either a multiple layer stack, so users can turn on and off for comparison (as in Google Earth), or be displayed simultaneously side by side, so users can zoom in/out or pan all at the same time (as in one of the recent features in ArcGIS). User’s input can be recorded by drawing a polygon around the DPM pixels or convert the raster pixels into GIS objects that interact with user’s mouse click. We practiced both in this activity and produced Google Earth polygons and ArcGIS point clouds. Three volunteers worked on the validation tasks, and the results were compiled into one set of point clouds, and then the point clouds were converted back to raster image in the same coordinates of the input DPM for analysis and visualization.

In particular, we first broke up the area of interest (AOI) into three areas of roughly the same size by first splitting into county and then agglomerating the groups to get approximate equal areas (1/3 the AOI). The next step was to remove the “0” value pixels by reclassifying the “0” values to “NULL”. This helped to reduce the number of pixels to deal with for the following steps.

Figure 4. (a) ASTER image with RGB channel being assigned with R: Green part of the spectrum, G: Red, B: Near IR. (b) ASTER difference image in red spectrum (20121104 – 20110901). The linear artifacts running in NW-SE direction are due to clouds and cloud shadows

Next the binary DPM raster image was split into the three areas and then each raster was “vectored” by converting each pixel to point vectors with attribute values comparable to the pixel value of the binary DPM raster (1 for predicted positive). Point values were then reclassified as “0” to represent non-processed points. Then points were assigned to the three evaluators. From this point every evaluator did something different. One used the points in ArcGIS and another drew polygons in Google Earth. One used the points for the visual comparison and classified true positives as “1” and false positives as -1, using “0” value to track progress. Once all of the points were processed, reclassified “-1” points were reclassified as “0” to represent false positives. In retrospect we could have started with predicted positives as “-1” instead of “0”. Finally, the point vectors were rasterized to pixels with the appropriate values (“1” – true positive and “0” – false positive).

E. Results

The true positives from our validation are shown in Figure 6. The total number of red pixels in the DPM is 13,723 and the number of confirmed damage pixels is 1,139, about 8.3% of the DPM pixels. Each red cross represents the location of the red pixel in the damage proxy map that was validated to be correct detection. The size of the red crosses is much bigger than the size of the red pixels for visibility. Figure 7a shows the DPM of southeastern part of Staten Island, and Figure 7b shows the validation task results. The red cross marked on top of the red pixels mean correct detection, and the blue stars on the red pixels represent false positives. The group of red pixels in the right-hand-side of Figure 7a indicates severe damage on the houses. In fact, along the coast near Cedar Grove Avenue, almost all the houses were swept away, which can be confirmed in Figure 7c and 7d.

Figure 5. WorldView-1 image of New York City mosaicked and overlaid on Google Earth. The red boundary indicates the extent of the mosaicked image.

There are two major sources of false positives. The first type of major false positives is vegetation change. The central part of Figure 7 is an example of false positives due to vegetation. Radar coherence is highly sensitive to vegetation and vegetation change because it contributes to volume decorrelation and temporal decorrelation at the same time. Plants can easily look different to radar at different time, because they sometimes sway in the wind and continuously grow. In case of natural disasters, plants do get damaged, which can also be the reason for them to appear as red pixels in DPMs. In most cases, however, the changes in plants are not our interest for disaster response.

Figure 6. Damage confirmed with visual inspection starting from the damage proxy map (Figure 3. Each red cross represents the location of the red pixel in the damage proxy map that was validated to be true positive

The second type is human activities that do not involve structure damage. For example, the distribution of cars at parking lots change every day. The distributions of merchandise containers and ships on and around the docks at harbors change on daily basis as well. This type of change appears as change in the recorded radar echo because of the difference in the way the radar signal reflects from the different configuration of objects. For example, the two parallel docks in the city of Bayonne produced a number of red pixels in the damage proxy map (Figure 8a), but we confirmed that most of the signal is due to the change in the objects placed on the docks as can be seen in Figure 8b-8d. Although we did not track false negatives, we saw occasional false negatives through our work. For example, the houses at the upper right corner of the before image (Figure 7c) are gone in the after image (Figure 7d), but the red pixel group in the DPM did not stretch all the way up to the boundary.


The times of the radar data acquisition are not always closely aligned with those of the optical images. Thus, there may not be one-to-one match in this radar-optical image comparison. Also there may be uncertainty in optical imagery due to cloud cover and different illumination condition causing different shadows.

Figure 7. (a) Damage proxy map, (b) validation result of it, and (c) before and (d) after image on Google Earth of southeastern part of Staten Island. Near Cedar Grove Ave along the beach in the right-hand-side of the image are houses swept away by the storm surge (True Positives). Vast majority of the pixels are transparent, which means no damage. In many cases, this is true as shown in the Google Images (True Negatives). The house damage in the upper right corner of the image was not detected however (False Negatives). There are grassland in the middle of images that showed up as red pixels but not due to structural damage (False Positives).

The eastern side of New York City was also imaged with an adjacent track of COSMO-SkyMed satellites as is summarized on Table III, and their swath footprint is indicated in Figure 1. However, we noticed that one of the SAR images was contaminated with radio frequency interference (RFI). Figure 10a is SAR image acquired on 2012-10-26 showing the RFI effect, whereas Figure 10b was acquired on 2012-11-04 and does not have as much RFI effect. We have requested raw data from ASI and are currently working on removing the RFI from the raw data.

Building footprint information is critical for refining the DPM, because both vegetation change and human activities can easily be ruled out by applying the building footprint mask. New York City has provided a layer of building footprint of the southern part of the city, and we are currently doing analysis with the building footprint layer. Figure 9 shows the building footprints, where about one million buildings are outlined with the blue polygons. Taking the intersection of the building footprint layer with the DPM will dramatically reduce the number of red pixels in the DPM, which will substantially reduce the number of pixels to validate.

Figure 8. Damage proxy map and Google Earth images of the docks in the city of Bayonne, New Jersey (a littler upper left from the center of Figure 3). The changes on the dock by human activities produced the false positives.

Figure 9. Building footprint of southern part of New York City. About one million building outlines are displayed.

Figure 10. (a) SAR amplitude image contaminated with the radio frequency interference (RFI) (2012-10-26), with a number of horizontal white lines and overall degradation with blurring. (b) a SAR amplitude image with no significant RFI (2012-11-04).


Following the major devastation of Hurricane Sandy, the ARIA team at JPL/Caltech, ASI, and GISCorps teamed up to produce a damage detection map of New York City and its validation product, using radar data from COSMO-SkyMed satellites, NOAA’s aerial photographs, and Google Earth images. We used ROI_PAC and damage proxy map algorithm, both developed at JPL/Caltech, Google Earth, and ArcGIS for data processing and analyses. With no dedicated system or personnel prepared for this major disaster, we were able to produce the damage proxy map on Day 11 and validation products on Day 15. This involves manual data discovery, manual data request (via Emails), manual data downloading (via ftp file transfer), and manual coordination of work (via teleconference and Emails). The results were encouraging and demonstrated great potential of future automated system. Given the data acquisition latency was 5 days for strip-map mode data of New York City, the automated system ARIA projects are building would have produced the products within a week.


This research was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration and funded through Advanced Rapid Imaging and Analysis for Monitoring Hazards (ARIA-MH) project, under the NASA Advanced Information Systems Technology (AIST) Program. We acknowledge the Italian Space Agency for its unreserved support for this response activity by providing the COSMO-SkyMed data, which became the critical part of this effort (COSMO-SkyMed Product – ASI – Agenzia Spaziale Italiana – (2012). All Rights Reserved). We appreciate Michael Evanoff at Google Inc. for providing service of overlaying the large WorldView image on Google Earth through the Earthbuilder. We also thank Sean Maday, Noel Gorelick and other Google Earth team members for demonstrating the usage of Google Earth and Maps Engines. We thank Brenda Jones, the Disaster Response Coordinator at USGS EROS Center, for providing access to the USGS HDDS archive and providing us with the bulk downloading instructions. We also thank the NOAA for timely data acquisition and release.


[1] The National Hurricane Center, “Post-Tropical Cyclone SANDY,” October 2012. [Online]. Available:

[2] The CBS Boston, “Hurricane Sandy Grows To Largest Atlantic Tropical Storm Ever,” October 2012. [Online]. Available:

[3] The Washington Post, “Sandy Brings Hurricane-Force Gusts After New Jersey Landfall,” October 2012.

[4] Wikipedia, “Hurricane Sandy,” January 2013. [Online]. Available: Sandy

[5] Reuters, “Factbox: Storm Sandy blamed for at least 132 deaths in U.S., Canada,” November 2012. [Online]. Available:

[6] The Washington Post, “Superstorm Sandy packed more total energy than Hurricane Katrina at landfall,” November 2012.

[Online]. Available: blog.html

[7] The Huffington Post, “Hurricane Sandy’s Storm Surge Mapped … Before It Hit,” November 2012. [Online]. Available: n 2094939.html

[8] The New York Times, “Staten Island Was Tragic Epicenter of Storms Casualties,” November 2012. [Online]. Available: r=0

[9] S.-H. Yun, E. Fielding, M. Simons, P. A. Rosen, S. Owen, and F. Webb, “Damage proxy map and coherence change time series, IEEE Transactions on Geoscience and Remote Sensing,” in review.

[10] NOAA, “Aerial Photos of Hurricane Sandy Damage from NOAA’s National Geodetic Survey,” November 2012. [Online]. Available:

[11] H. Zebker, P. A. Rosen, S. Hensley, and P. J. Mouginis-Mark, “Analysis of active lava flows on Kilauea volcano, Hawaii, using SIR-C radar correlation measurements,” Geology, vol. 24, no. 6, pp. 495–498, 1996.

[12] M. Simons, Y. Fialko, and L. Rivera, “Coseismic deformation from the 1999 mw 7.1 hector mine, california, earthquake as inferred from insar and gps observations,” Bulletin of the Seismological Society of America, vol. 92, no. 4, pp. 1390–1402, 2002.

[13] E. J. Fielding, M. Talebian, P. A. Rosen, H. Nazari, J. A. Jackson, M. Ghorashi, and R. Walker, “Surface ruptures and building damage of the 2003 Bam, Iran, earthquake mapped by satellite synthetic aperture radar interferometric correlation,” J. Geophys. Res., vol. 110, no. B3, p. B03302, 2005.

[14] J. Hoffmann, “Mapping damage during the Bam ( Iran ) earthquake using interferometric coherence,” International Journal of Remote Sensing, vol. 28, no. 6, pp. 1199–1216, 2007.

[15] P. A. Rosen, S. Hensley, G. Peltzer, and M. Simons, “Updated Repeat Orbit Interferometry package released,” Eos Trans. AGU, vol. 85, no. 5, p. 35, 2004

Back To Top