Harel Dan
5 min readOct 22, 2018

X Marks The Spot: Identifying MIM-104 Patriot Batteries From Sentinel-1 SAR Multi-temporal Imagery

There are two main types of remote sensing satellites: optical and radar. Each type can be subdivided further into sub-categories based on aperture, orbit, and bands. One of the most used is ESA’s Copernicus program Sentinel 1 pair of satellites, S1A and S1B, giving a combined average revisit time of 1.5 days in a best case scenario.

Such a high-resolution and high revisit time, as well as the open access approach, has made the data these satellites provide essential in many fields of study, from emergency response, marine monitoring, vegetation analysis, wildfire quantification, and urban planning.

The data can be freely downloaded and analysed on many platforms, including Copernicus Open Data Hub, Sentinel EO Browser and Google Earth Engine,

Whereas optical imagery can have optical interferences like clouds and dust, radar images can for the most part “see” through water vapor and other fine particulate matter. Conversely, radars can be susceptible to interferences from other radiation sources on the ground and in the air transmitting in similar wavelengths.

Free roaming with Sentinel-1 data, you encounter many types of interferences, blips, bloops, speckles, and waves, so aggregating several images can create a smoother image, and remove some if not all of the interferences.

“Blips, bloops, speckles, and waves” — Screenshots from Sentinel-Hub EO Browser
ESA, 2016

These image artifacts are the result of higher return signals. They come in a variety of polarizations, dimensions and locations, but will always have a main angle perpendicular the the satellite flight direction, thus will have two distinct tilt angles based on the orbit type.

Two paragraphs ago I mentioned that most of the noise can be removed by some forms of image aggregation, or multi-temporal analysis, where for each image pixel the lowest value is selected. When I attempted such a feat in Google Earth Engine, I accidentally selected the maximum value, and the results were staggering.

Displaying a combination of VH and VV polarizations, these lines, the result of overlapping ascending and descending orbit interferences, consistently converge.

X marks a spot, what is it?

What is it?

Long story short, some of these are AN/MPQ-53/65 phased array radars that form a Patriot missile battery C². Looking at official documentation, the military G-band is the civilian C-band. Sentinel-1 central frequency is 5.405 Ghz, well within this range, hence my working hypothesis is that there is some sort of ground based interference with the Sentinel-1 signal.

So anywhere in the world these artifacts appear, they may point to a location of a patriot battery, or other early warning system, as I shall show.

Here are several locations as seen in the GEE results viewer, and a zoom in of the convergence using Google satellite imagery.

Ascending and Descending orbits converge over Al-Udeid Base in Qatar
Al-Udeid MIM-104 Patriot battery
Ascending and Descending orbits converge over Isa Airbase in Bahrain
Isa Airbase MIM-104 Patriot battery

Corroboration for my analysis comes from other GEOINT analysts, whom using the recent Strava Faux Pas, pinpointied Emirati Patriot deployment in Aden. This exact location is once a gain the exact center of the X

Ascending and Descending orbits converge over Aden, Yemen
Ascending and Descending orbits converge over Al-Azraq Airbase, Jordan
Al-Azraq Airbase MIM-104 Patriot battery
Ali Al-Salem Airbase, Kuwait
Israeli Patriot deployment
White Sands Missile Range

These example are with high confidence the result of the AN/MPQ-53/65 Radar. However, there are other sources for interferences, such as the Swedish STRIL array

STRIL Array around Sweden
STRIL location in Sweden

The above screenshots are from my GEE script.

If you don’t have a GEE account, the results are available via GEE Apps.

The script collates a defined time span of images, performs the necessary filters, and displays the result. The longer the time span, the more “noise” gets added to the result, and the convergence lines become more robust. Theoretically, If you limit the time span, one could infer deployments of forces based on the time the convergence appears and disappears, give or take a few days.

A final note: EO data is becoming more democratized and accessible, similar to how Google Earth has democratized GIS in a way, and made aerial imagery publicly available some 15 years ago. That said, it is crucial that companies, organizations and nations adapt to the new reality, especially those who deal with perceived classified information they do not want to have arm-chair analysts expose.