Topics: multispectral satellite imagery, machine learning, informal economies, war.
+
In Syria, over a decade of war has ravaged one of its most important industries. Oil is a basic necessity for local residents who need to heat their homes and keep the lights on. It’s also an important source of income for armed groups who control production; by some estimates the Islamic State was making over $40 million per month in oil revenues, and the Syrian Democratic Forces were making $120 million per year selling oil to their adversaries, the Syrian Government. This, in turn, has made oil production facilies a frequent target of airstrikes, leading to catastrophic environmental consequences.
+
The destruction of Syria’s oil infrastructure and its importance as a source of revenue for armed groups has led to a massive rise in makeshift oil extraction and refining. These makeshift refineries are often constructed by digging a large pit, lining it with a tarp, and filling it with polluted water. A furnace heats crude oil, which is run through a pipe cooled by the basin and collected in drums:
+
Wim Zwijnenburg wrote an excellent Bellingcat article on the subject, which you should read before going any further in this tutorial. In the article, Wim notes that these facilities “can be spotted by the ditch and the black spot at the end with oil waste residues, which blacken the soil around the furnace.” These refineries also frequently leak, blackening large swaths of land around them.
Looking around Northwestern Syria, we can see agricultural fields pockmarked by these makeshift refineries (you can pan around and zoom in):
+
+
+
+
+
+
Previous efforts to quantify informal oil production have involved manually sifting through satellite imagery and counting them. This is a painstaking process that leaves a number of important questions unanswered. Even if we were to count all of the individual refineries, could we get an estimate of polluted area? What if we wanted to count the refineries in a new part of Syria? Or get annual or even monthly estimates of new refineries?
+
Below is an Earth Engine application that automates the detection of makeshift refineries in Northeastern Syria, using mutlispectral satellite imagery and machine learning. Blue dots represent the locations of predicted makeshift oil refineries and general oil pollution, while red areas indicate areas predicted to be contaminated by oil.
+
+
You can draw an Area of Interest (AOI) and get the total number of contaminated points as well as the total number of contaminated square meters within the AOI. drawing multiple AOIs will show a running total of these statistics. It’s not perfect– it misses some refineries and falsely identifies some others– but it is generally quite accurate; you can visually verify the results of the prediction by zooming in using the “+” button. You can toggle different layers using the “layers” tab as well. This tool could be used to get an estimate of oil production in a user-defined area, and eventually to direct cleanup efforts. The fullscreen version of the application can be found here, and the source code here. This tutorial will first cover the basics of multispectral remote sensing, before moving into a step-by-step guide in the construction of this model.
+
+
Multispectral Remote Sensing
+
Different materials reflect light differently. An apple absorbs shorter wavelengths (e.g. blue and green), and reflects longer wavelengths (red). Our eyes use that information– the color– to distinguish between different objects. But our eyes can only see a relatively small sliver of the electromagnetic spectrum covering blue, yellow, and red; we can’t see UV or infrared wavelengths, for example, though the extent to which different materials reflect or absorb these wavelengths is just as useful for distinguishing between them. For example, Astroturf (fake plastic grass) and real grass will both look green to us, espeically from a satellite image. But living plants absorb radiation from the sun in a part of the light spectrum that we can’t see. There’s a spectral index called the Normalized Difference Vegetation Index (NDVI) which exploits this fact to isolate vegetation in multispectral satellite imagery. So if we look at Gilette Stadium near Boston, we can tell that the three training fields south of the stadium are real grass (they generate high NDVI values, showing up red), while the pitch in the stadium itself is astroturf (generating low NDVI values, showing up blue).
+
+
+
+VHR image of Gilette Stadium with Sentinel-2 derived NDVI overlay
+
+
+
In other words, even though these fields are all green and indistinguishable to the human eye, their spectral profiles beyond the visible light spectrum differ, and we can use this information to distinguish between them. Below is a plot of the spectral profiles of different materials, including oil.
+
+
+
The European Space Agency’s Sentinel-2 satellite collects spectral information well beyond the visible light spectrum, enabling this sort of analysis. It chops the electromagnetic spectrum up into “bands”, and measures how strongly wavelengths in each of those bands is reflected:
+
+
We’ll be using this satellite to distinguish between oil and other materials, similar to the way we were able to distinguish between real and fake grass at Gilette Stadium. First, we’ll have to do a bit of pre-processing on the Sentinel-2 imagery after which we’ll train a machine learning model to identify oil.
+
+
+
Machine Learning Workflow
+
+
Pre-Processing
+
As always, the first step in our project will be to load and pre-process satellite imagery. For this project, we’ll be using Sentinel-2 imagery. Let’s load imagery from 2020-2021, filter out cloudy images, and define visualization parameters:
When loading the Sentinel-2 imagery, I’ve also onlyh selected the bands that we will ultimately use in our analysis. There are a number of other bands included in the data that we don’t need. I’ve omitted a few bands (B1, B9, B10) because they’re collected at a much lower spatial resolution (60 meters) compared to the other bands.
+
A couple types of landcover are so readily identifiable that we can remove them with thresholds. Water and vegetation both have spectral indices; we looked at NDVI above, but there’s a similar one for water called NDWI. These can be calculated from Sentinel-2 imagery as follows:
+
var ndvi=sentinel.normalizedDifference(['B8','B4'])
+.select(['nd'],['ndvi'])
+
+var ndwi=sentinel.normalizedDifference(['B3','B8'])
+.select(['nd'],['ndwi'])
+
We use the normalizedDifference function and specify which bands we want to use for each index. NDVI uses the red and near infrared bands (B4 and B8), while NDWI uses bands 3 and 8. Finally, we want to rename the resulting band from ‘nd’ to the name of the spectral index.
+
Now we can use these indices to filter out water and vegetation. We do this using the updateMask function, and specify that we want to remove areas that have an NDVI value higher than 0.2 and and NDWI value higher than 0.3. You can play around with these thesholds until you achieve the desired results.
We also want to only select bands that are relevant to our analysis; Sentinel
+
Finally, let’s clip the image to our Area of Interest (AOI) and add it to the map using the visualization parameters we defined earlier.
+
Map.addLayer(image.clip(AOI), s_rgb,'Sentinel');
+
+
+
+water and vegetation have been removed from this Sentinel-2 image. What remains is largely fallow agricultural land, urban areas, and oil spills.
+
+
+
Now that we’ve loaded and preporcessed our satellite imagery, we can proceed with the rest of our task. Ultimately, we want to create a map of the study area which shows us different “landcovers” (materials). This can broadly be achieved in three steps:
+
+
Generate labeled landcover data
+
Train a model using labeled data
+
Validate the model
+
+
+
+
1. Generating Labeled Data
+
A vital step in any machine learning workflow is the generation of labeled data, which we will use to train a model to differentiated between different types of land cover and later to test the model’s accuracy. By looking around the study area, we can get a sense of the different land cover classes that we might encounter:
+
+
Agricultural Land
+
Urban Areas
+
Oil Contamination
+
+
Naturally we could subdivide each of these into sub-categories, and there are probably other classes we haven’t included that may be present in the study area. The choice of classes is partly informed by the nature of the task at hand. In theory, the most efficient number of classes for this task would be two: oil, and everything else. The problem is that the “everything else” category would be pretty noisy since it would include a wide range of materials, making it harder to distinguish this from oil. In practice, a visual inspection of major landcover classes in the study area is a quick-and-dirty way of getting at roughly the right number of classes. This is also an iterative process: you can start with a set of labeled data, look at the model results, and adjust your sampling accordingly. More on this later.
+
The main landcover class we’re interested in is, of course, oil. Some oil contamination is readily visible from the high resolution satellite basemap; rivers of oil flow from the leaking Ger Zero refinery. We can draw polygons around the oil contamination like so:
+
+
The same process is applied to agricultural land and urban areas. In general, you want to make sure that you’re sampling from all across the study area. I’ve generated between 4-10 polygons per landcover class in different places. We’re now left with a featureCollection composed of polygons for each class. I’ve named them oil, agriculture, and urban.
+
However, I don’t just want to use all of the pixels contained in these polygons for training. There are several reasons for this. First, it would likely lead to overfitting. Second, there are probably over a million pixels between all of the polygons, which would slow things down unnecessarily. Third, I haven’t drawn the polygons to be equal sizes across classes, so I could end up with way more points from one class compared to another. It’s OK to have some imbalance between classes, but you don’t want it to be extreme.
+
As such, the next step involves taking random samples of points from within these polygons. I do so using the randomPoints function:
In the first line, I create a new featureCollection called oil_points which contains 3000 points sampled from the polygons in the oil featureCollection. I then map through each of these points, and set a property called “class” equal to 0. I do the same for the urban and agricultural areas, setting the “class” property of these featureCollections to 1 and 2, respectively. Ultimately, our model will output a raster in which each pixel will contain one of these three values. A value of 0 in the output will represent the model predicting that that pixel is oil, based on the training data; a value of 1 would indicate predicted urban land cover, and 2 predicted agricultural landcover.
+
Now we want to create one feature collection called “sample”, which will contain all three sets of points.
+
var sample=ee.FeatureCollection([oil_points,
+ urban_points,
+ agriculture_points
+ ])
+.flatten()
+.randomColumn();
+
We’ve also assigned a property called “random” using the randomColumn function. This lets us split our featureCollection into two: one used for training the model, and one used for validation. We’ll use a 70-30 split.
Having generated labeled training and testing data, we now want to teach an algorithm to associate the pixels in those areas (in particular, their spectral profiles) with a specific landcover class.
+
The list of points we generated in the previous step contain a label (0: oil, 1: urban, 2: agriculture). However, they do not yet contain any information about the spectral profile of the Sentinel-2 image. The sampleRegions function lets us assign a the band values from an image as properties to our feature collection. We do this for both training sample and the validation sample.
Each point in the featureCollections above will contain a property denoting each Sentinel-2 band’s value at that location, as well as the property denoting the class label.
+
Now we’re ready to train the model. We’ll be using a Random Forest classifier, which basically works by trying to separate your data into the specified classes by setting lots of thresholds in your input properties (in our case, Sentinel-2 band values). It’s a versatile and widely-used model.
+
We first call a random forest classifier with 500 trees. More trees usually yields higher accuracy, though there are diminishing returns. Too many trees will result in your computation timing out. We then train the model using the train function, which we supply with the training data as well as the name of the property that contains our class labels (“class”).
+
var model = ee.Classifier.smileRandomForest(500)
+.train(training,'class');
+
The trained model now associates Sentinel-2 band values with one of three landcover classes. We can now feed the model pixels it has never seen before, and it will use what it now knows about the spectral profiles of the differnt classes to predict the class of the new pixel.
+
var prediction = image.classify(model)
+
prediction is now a raster which contains one of three values (0: oil, 1: urban, 2: agriculture). We’re only interested in oil, so let’s isolate the regions in this raster that have a value of 0, and add them in red to the map:
+
var oil_prediction=prediction.updateMask(prediction.eq(0))
+
+Map.addLayer(oil_prediction, {palette:'red'},'Predicted Oil Conamination')
+
+
+
+
3. Validation
+
The image above should look somewhat familiar. It’s Ger Zero, where we trained part of our model. We can see in red the areas which the model predicts to be oil pollution. These largley align with the areas that we can see as being contaminated based on the high resolution basemap. It’s not perfect, but it’s pretty good.
+
Let’s scroll to another area, far from where the model was trained. This image shows two clusters of makeshift refineries which were identified by the model. This is good, though we can only get so far by visually inspecting the output from our model. To get a better sense of our model’s performance, we can use the validation data that we generated previously. Remember, these are labeled points which our model was not trained on, and has never seen before.
+
We’ll take the validation featureCollection containing our labeled points, and have our model classify it.
+
var validated = validation.classify(model);
+
Now the validated variable is a featureCollection which contains both manual labels and predicted labels from our model. We can compare the manual labels to the predicted output to get a sense of how well our model is performing. This is called a Confusion Matrix (or an Error Matrix):
+
var testAccuracy = validated.errorMatrix('class','classification');
+
+print('Confusion Matrix ', testAccuracy);
+
+
+
+
+
+
+
+
+
+
+
+
+
+
Labels
+
+
+
+
+
+
+
+
Oil
+
Urban
+
Agriculture
+
+
+
+
Oil
+
876
+
1
+
5
+
+
+
Prediction
+
Urban
+
0
+
168
+
8
+
+
+
+
Agriculture
+
1
+
4
+
514
+
+
+
+
Now, we can see that of the 877 points that were labeled “oil”, only one was falsely predicted to be agicultural land. The model also falsely predicted as oil one point that was labeled urban, and five points that were labeled agriculture. Not bad. We can get a sense of the model’s overall accuracy using the accuracy function on the confusion matrix:
This tells us that the overall accuracy of our model is around 98%. However, we shouldn’t take this estimate at face value. There are a number of complicated reasons (spatial autocorrelation in the training data, for example) why this figure is probably inflatred. If we were submitting this analysis to a peer-reviewed journal, we’d take great care in addressing this, but for our purposes we can use the accuracy statistics to guide our analysis and get a rough sense of how well the model is performing.
+
This model isn’t perfect; it often misclassifies the shorelines of lakes as oil, or certain parts of urban areas. As previously mentioned, training a model is often an iterative process. At this stage, if your accuracy is not as high as you’d like it to be, you can use the output to figure out how to tweak the model. For example, you may observe that your model is confusing urban areas with oil spills. You can draw a polygon over the erroneous area, label it urban landcover and retrain the model thereby hopefully improving accuracy. We could further refine our model in this way.
+
+
+
+
Communicating the Results
+
Now that we’ve got a model that can identify oil from multispectral satellite imagery fairly well, we can set about making our results accessible.
+
One of the things we’re particularly interested in is the distribution of small refineries. The way we’re currently visualizing the prediction (the raster output from the model where predicted oil is shown in red and everything else is transparent) makes it hard to see these small refineries when we zoom out:
+
+
We can convert our raster into a series of points using the reduceToVectors function. In essence, this takes homogenous regions of an image (e.g., an area predicted to be oil surrounded by an area not predicted to be oil) and converts it into a point:
Now the distribution of small refineries is much more easily visible as blue dots:
+
+
If we zoom out even further, we can see clusters of points that correspond to areas of high oil production. Using geolocated photographs, we can roughly ground-truth the model output:
+
+
+
+
+
+
+
+
+
+
+
+
+
+
\ No newline at end of file
diff --git a/_book/SyriaNTL.html b/_book/SyriaNTL.html
new file mode 100644
index 0000000..74e020e
--- /dev/null
+++ b/_book/SyriaNTL.html
@@ -0,0 +1,736 @@
+
+
+
+
+
+
+
+
+
+Google Earth Engine for OSINT - War at Night
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
War at Night
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
Data
+
Satellite images of Syria taken at night capture a subtle trace left by human civilization: lights. Apartment buildings, street lights, highways, powerplants– all are illuminated at night and can be seen from space. Researchers often use these nighttime lights signatures to track development; as cities grow, villages recieve power, and infrastructure is built, areas emit more light. But this works both ways. As cities are demolished, villages burned, and highways cutoff, they stop emitting lights.
+
The timelapse below uses imagery from the Defense Meteorological Satellite Program (DMSP), a joint program run by the U.S. Department of Defense and the National Oceanographic and Atmospheric Agency. One image is taken per year between 2005 and 2013:
+
+
+
Ukraine
+
+
Pre-Processing
+
+
+
Analysis
+
+
+
+
Iraq
+
A link to the GEE code for this section can be found here.
+
+
Pre-Processing
+
First, let’s start by importing a few useful packages written by Gennadii Donchyts. We’ll use utils and text to annotate the date of each image on the timelapse. We’ll also define an Area of Interest (AOI), which is just a rectangle. You can do this manually by clicking the drawing tools in the top left. I’ve drawn an AOI over the area covering Mosul, Irbil, and Kirkuk in Northern Iraq.
+
var utils =require("users/gena/packages:utils");
+var text =require("users/gena/packages:text");
+
+// define the Area of Interest (AOI)
+var AOI = ee.Geometry.Polygon(
+ [[[42.555362833405326,36.62010778397765],
+ [42.555362833405326,35.18296243288332],
+ [44.681217325592826,35.18296243288332],
+ [44.681217325592826,36.62010778397765]]])
+
+// start and end dates for our gif
+var startDate ='2013-01-01';
+var endDate ='2018-01-01';
+
+// a filename for when we export the gif
+var export_name='qayyarah_viirs'
+
+// A palette to visualize the VIIRS imagery. This one is similar to Matplotlib's "Magma" palette.
+var viirs_palette = [
+"#000004",
+"#320a5a",
+"#781b6c",
+"#bb3654",
+"#ec6824",
+"#fbb41a",
+"#fcffa4",
+];
+
+// Visualisation parameters for the VIIRS imagery, defining a minimum and maximum value, and referencing the palette we just created
+var VIIRSvis = { min:-0.1,max:1.6,palette: viirs_palette };
+
Next, we’ll load the VIIRS nighttime lights imagery. We want to select the avg_rad band of the image collection, and filter blank images. Sometimes, we get blank images over an area in VIIRS if our AOI is on the edge of the satellite’s imaging swath. We can filter these images, similarly to how we filter for cloudy images in Sentinel-2:
+
var VIIRS= ee.ImageCollection("NOAA/VIIRS/DNB/MONTHLY_V1/VCMCFG")
+.select('avg_rad')
+// Calculate the sum of the 'avg_rad' band within the AOI
+.map(function(image) {
+var blank=image.reduceRegions({
+collection: AOI,
+reducer: ee.Reducer.sum(),
+scale:10})
+.first()
+.get('sum')
+// For each image, define a property 'blank' that stores the sum of the 'avg_rad' band within the AOI.
+// We're also going to take a base 10 log of the image-- this will help us visualize the data by dampening extreme values
+return image.set('blank', blank).log10().unmask(0)
+ })
+// Now, we can filter images which are fully or partially blank over our AOI
+.filter(ee.Filter.gt('blank',10))
+// Finally, we filter the collection to the specified date range
+.filterDate(startDate, endDate)
+
+
Let’s have a look at the first image in the collection to make sure everything’s looking right. We’ll set the basemap to satellite and center our AOI:
If we decrease the opacity of the VIIRS layer, we can see the cities of Mosul, Erbil, and Kirkuk shining brightly at night. We can also see a string of bright lights between Kirkuk and Erbil– these are methane flares from oil wells.
+
+
+
Analysis
+
Having pre-processed the VIIRS imagery, we can now define a function gif that will take:
+
+
An image collection (col, in this case the nighttime lights imagery VIIRS)
+
Visualization parameters (col_vis, in this case VIIRSvis)
+
An Area of Interest AOI
+
+
The function will then return a timelapse.
+
var gif =function (col, col_vis, AOI) {
+
+// Define the date annotations to be printed in the top left of the gif in white
+var annotations = [
+ {
+textColor:"white",
+position:"left",
+offset:"1%",
+margin:"1%",
+property:"label",
+// Dynamically size the annotations according to the size of the AOI
+scale: AOI.area(100).sqrt().divide(200),
+ },
+ ];
+
+// Next, we want to map over the image collection,
+var rgbVis = col.map(function (image) {
+// Get the date of the image and format it
+var start = ee.Date(image.get("system:time_start"));
+var label = start.format("YYYY-MM-dd");
+// And visualize the image using the visualization parameters defined earlier.
+// We also want to set a property called "label" that stores the formatted date
+return image.visualize(col_vis).set({ label: label });
+ });
+
+// Now we use the label proprty and the annotateImage function from @gena_d to annotate each image with the date.
+ rgbVis = rgbVis.map(function (image) {
+return text.annotateImage(image, {}, AOI, annotations);
+ });
+
+// Define GIF visualization parameters.
+var gifParams = {
+maxPixels:27017280,
+region: AOI,
+crs:"EPSG:3857",
+dimensions:640,
+framesPerSecond:5,
+ };
+
+// Export the gif to Google Drive
+ Export.video.toDrive({
+collection: rgbVis,
+description: export_name,
+dimensions:1080,
+framesPerSecond:5,
+region: AOI,
+ });
+// Print the GIF URL to the console.
+print(rgbVis.getVideoThumbURL(gifParams));
+
+// Render the GIF animation in the console.
+print(ui.Thumbnail(rgbVis, gifParams));
+};
+
Ok that was a pretty big chunk of code. But the good news is that we basically never have to touch it again, since we can just feed it different inputs. For example, if I want to generate a gif of nighttime lights over a different area, it’s as simple as dragging the AOI. If I want to look at a different time period, I can just edit the startDate and endDate variables. And if I want to visualize an entirely different type of satellite imagery– Sentinel-1, Sentinel-2, or anything else, all I have to do is change the image collection (col) and visualization parameters (col_vis) variables. Now, let’s look at some timelapses.
+
+
The Fall of Mosul
+
The function returns a timelapse of nighttime lights over Northern Iraq:
+
gif(VIIRS, VIIRSvis, AOI);
+
+
+
+I’ve done a bit of post-processing to this gif, adding more annotations and blending between frames to make it a bit smoother. I typically use ffmpeg and ezgif for the finishing touches.
+
+
+
This timelapse gives a play-by-play of one of the most important campaigns in the war against the Islamic State. In the first few frames, Mosul is under the control of the Kurdistan Regional Government (KRG). In the summer of 2014, ISIS captures the city, and power is cut off. Mosul and many villages along the Tigris river are plunged into darkness. In 2015, the front line in the campaign to retake the city emerges around Mosul, advancing in 2016 and 2017. Mosul is eventually retaken by the KRG in 2017, after which it brightens once again as electricity is restored.
+
+
+
The Qayyarah Fires
+
Farther south, there is an interesting detail. Above the “h” in “Qayyarah”, a bright set of lights emerges just before Mosul is recaptured, around December 2016. Fleeing Islamic State fighters set fire to the Qayyarah oilfields, which burned for months.
+
Using the VIIRS data we’ve already loaded, we can further analyze the effect of the conflict using a chart. First, let’s define two rectangles (again, you can draw these) over Mosul and Qayyarah:
+
var mosul = ee.Feature(
+ ee.Geometry.Polygon(
+ [[[43.054977780266675,36.438274276521234],
+ [43.054977780266675,36.290642221212416],
+ [43.24792516796199,36.290642221212416],
+ [43.24792516796199,36.438274276521234]]],null,false),
+ {
+"label":"Mosul",
+"system:index":"0"
+ }),
+
+ qayyarah = ee.Feature(
+ ee.Geometry.Polygon(
+ [[[43.08240275545117,35.8925587996721],
+ [43.08240275545117,35.77899970860588],
+ [43.26642375154492,35.77899970860588],
+ [43.26642375154492,35.8925587996721]]],null,false),
+ {
+"label":"Qayyarah",
+"system:index":"0"
+ })
+
+// Let's put these together in a list
+var regions=[qayyarah, mosul]
+
Once we’ve got the rectangles, we can make a chart that will take the mean value of the VIIRS images in each rectangle over time:
We can clearly see Mosul (the red line) darkening in 2014 as the city is taken by ISIS. During this period the Qayyarah oilfileds are, as we might expect, quite dark. All of a sudden in 2016 Qayyarah becomes brighter at night than the city of Mosul ever was, as the oilfields are set on fire. Then, almost exactly when the blaze in Qayyarah is extinguished and the area darkens (i.e. when the blue line falls back to near zero), Mosul brightens once again (i.e. the red line rises) as the city is liberated.
+
+
+
+
The Battle for Aleppo
+
The images below were taken between 2012 and 2014. Vast swaths of the city darken as neighbourhoods are razed by fighting.
+
+
Though this is a trend that can be observed across the country, nowhere is the decline in nightlights more visible than in Aleppo. Below is a comparison of longitudinal trends in nighlights signatures between several cities:
+
+
The most salient trend is Aleppo plummeting over the course of 2012, and becoming steadily darker over the course of the next four years. Raqqa drops in 2012 as well, but remains in flux until 2017, when the battle to reclaim the city pluges it into near total darkness. Damascus also experiences a dip in 2012, but stabilizes relatively quickly. The Turkish city of Gaziantep– less than 100km from Aleppo and roughly 1/5th the size– stands in stark contrast to the Syrian cities, becoming progressively brighter over the entire period.
+
Another interesting pattern here is the difference in seasonal trends in nightlights. Under normal circumstances in this part of the world, cities become brighter at night during the summer months. Restaurants, bars, and markets stay open later and conduct business outdoors. Gaziantep, which still attracts scores of tourists every year, displays pronounced seasonality. Damascus, the most stable of the three Syrian cities, also maintains a seasonal trend throughout the war. In contrast, both Raqqa and Aleppo maintain extremely low and roughly constant levels of nightlights year-round during the periods following intense fighting.
+
Reliable economic data for Syria haven’t been available for nearly a decade, and assessing the country’s recovery is consequently difficult. But subtle indications of economic growth are visible above: all three Syrian cities have been on a steady upward trend since 2017, and beginning to display seasonal variation once again.
+
+
+
Fighting for Oil
+
Throughout the war, sudden massive spikes in nightlights signatures can be observed throughout the country. In the center of the map just west of Palmyra, some particularly large spikes occur in 2017:
+
These flashes of light show gas wells being set on fire, a common form of sabotage carried out by retreating Islamic State fighters. Modified Sentinel-2 imagery of the Hayyan gas field (indicated by the green box above) shows this in greater detail. Substituing the Red band in an RGB image with Near Infrared (NIR) highlights thermal signatures, showing fires burning brightly even during the day.
+
The large complex on the right is the Hayyan Gas Plant, which produced nearly 1/3 of Syria’s electricity. The plant and its associated wells changed hands several times throughout the war, but were under Islamic State control until February 2017. In the video below, Islamic State fighters can be seen rigging the plant with explosives and destroying it on January 8th:
+
In February, three Russian oil and gas companies (Zarubij Naft, Lukoil and Gazprom Neft) were given restoration, exploration, and production rights to the hydrocarbon deposits West of Palmyra. On January 12th, 2017, the Syrian Army’s 5th Legion and Russian special forces launched a counterattack known as the “Palmyra offensive”, with the aim of retaking several important hydrocarbon deposits including Hayyan.
+
The timing of well fires aligns closely with a detailed timeline of the campaign.The Near Infrared Sentinel-2 image below shows the layout of the Hayyan Gas Plant and the wells in the Hayyan gas field:
+
The Syrian Army took the Hayyan gas field on February 4th, and retreating ISIS fighters set fire to wells 1, and 3. However, ISIS managed to briefly retake the Hayyan field on February 7th, setting fire to wells 2 and 4. These moments in the Palmyra Offensive are captured in NIR signatures
+
Interestingly, despite the massive explosion caused by the bombing of the Hayyan Gas Plant, no prolonged thermal anomalies were detected over the area of the plant itself. The well fires, on the other hand, lasted for months. Below is an image of well fire at the Hayyan field taken from this video; based on the nearby infrastructure and date (04/02/2017) of posting, it is likely Well-3.
One of the main advantages of GEE is that it hosts several Petabytes of satellite imagery and other spatial data sets, all in one place. Among these are a many that could prove useful to those investigating illegal mining and logging, estimating conflict-induced damage, monitoring pollution from extractive industries, conducting maritime surveillance without relying on ship transponders, verifying the locations of artillery strikes, tracking missile defense systems, and many other topics. This section highlights ten categories of geospatial data available natively in the GEE catalogue ranging from optical satellite imagery, to atmospheric data, to building footprints. Each sub-section provides an overview of the given data type, suggests potential applications, and lists the corresponding datasets in the GEE catalogue. The datasets listed under each heading are not an exhaustive list– there are over 500 in the whole catalogue, and the ones listed in this section are simply the ones with the most immediate relevance to open source investigations. If a particular geospatial dataset you want to work with isn’t hosted in the GEE catalog, you can upload your own data. We’ll cover that in the next section.
+
+
1.1 Optical Imagery
+
+
+
+Sentinel-2 timelapse showing the ancient city of Hasankeyf being flooded following the construction of a dam by the Turkish government.
+
+
+
Optical satellite imagery is the bread and butter of many open source investiagtions. It would be tough to list off all of the possible use cases, so here’s a handy flowchart:
This is, of course, a bit of an exaggeration. But if you’re interested in a visible phenomenon that happens outdoors and that isn’t very tiny, chances are an earth-observing satellite has taken a picture of it. What that picture can tell you naturally depends on what you’re interested in learning.
+Detected fires over Ukraine since 27/02/2022 showing the frontline of the war
+
+
+
Earth-observing satellites can detect “thermal anomalies” (fires) from space. NASA’s Fire Information for Resource Management System (FIRMS) provides daily data on active fires in near real time, going back to the year 2000. Carlos Gonzales wrote a comprehensive Bellingcat article on the use of FIRMS to monitor war zones from Ukraine to Ethiopia. The map above shows that FIRMS detected fires over Eastern Ukraine trace the frontline of the war.
+
FIRMS data are derived from the MODIS satellite, but only show the central location and intensity of a detected fire. Another MODIS product (linked in the table below) generates a monthly map of burned areas, which can be used to assess the spatial extent of fires.
+
+
Applications
+
+
Identification of possible artillery strikes/fighting in places like Ukraine
+
Environmental warfare and “scorched earth” policies
+Population density estimates around Pyongyang, North Korea
+
+
+
Sometimes, we may want to get an estimate the population in a specific area to ballpark how many people might be affected by a natural disaster, a counteroffensive, or a missile strike. You can’t really google “what is the population in this rectangle i’ve drawn in Northeastern Syria?” and get a good answer. Luckily, there are several spatial population datasets hosted in GEE that let you do just that. Some, such as WorldPop, provide estimated breakdowns by age and sex as well. However, it is extremely important to bear in mind that these are estimates, and will not take into account things like conflict-induced displacement. For example, Oak Ridge National Laboratory’s LandScan program has released high-resolution population data for Ukraine, but this pertains to the pre-war population distribution. The war has radically changed this distribution, so these estimates no longer reflect where people are. Still, this dataset could be used to roughly estimate displacement or the number of people who will need new housing.
+
+
Applications:
+
+
Rough estimates of civilians at risk from conflict or disaster, provided at a high spatial resolution
+Building footprints in Mariupol, Ukraine colored by whether the building is damaged
+
+
+
A building footprint dataset contains the two dimensional outlines of buildings in a given area. Currently, GEE hosts one building footprint dataset which covers all of Africa. In 2022, Microsoft released a free global building footprint dataset, though to use it in Earth Engine you’ll have to download it from their GitHub page and upload it manually to GEE. The same goes for OpenStreetMap (OSM), a public database of building footprints, roads, and other features that also contains useful annotations for many buildings indicating their use. Benjamin Strick has a great youtube video on conducting investigations using OSM data.
+
+
Applications:
+
+
Joining damage estimate data with the number of buildings in an area
+Second-level administrative boundaries in Yemen
+
+
+
Spatial analysis often have to aggregate information over a defined area; we may want to assess the total burned area by province in Ukraine, or count the number of Saudi airstrikes by district in Yemen. For that, we need data on these administrative boundaries. GEE hosts several such datasets at the country, province, and district (or equivalent) level.
+
+
Applications
+
+
Quick spatial calculations for different provinces/districts in a country
+
+
e.g. counts of conflict events by district over time
The Global Power Plant Database is a comprehensive, open source database of power plants around the world. It centralizes power plant data to make it easier to navigate, compare and draw insights. Each power plant is geolocated and entries contain information on plant capacity, generation, ownership, and fuel type. As of June 2018, the database includes around 28,500 power plants from 164 countries. The database is curated by the World Resources Institude (WRI).
+
+
Applications:
+
+
Analyzing the impact of conflict on critical infrastructure.
+
+
e.g. fighting in Ukraine taking place around nuclear power facilities.
+
+
Could be combined with the atmospheric measurements of different pollutants and the population estimates data to assess the impact of various forms of energy generation on air quality and public health.
Spatial resolution governs how “sharp” an image looks. The Google Maps satellite basemap, for example, is really sharp Most of the optical imagery that is freely available has relatively low spatial resolution (it looks more grainy than, for example, the Google satellite basemap),
+
+
+
+
+
3.1.2 Spectral Resolution
+
What open source imagery lacks in spatial resolution it often makes up for with spectral resolution. Really sharp imagery from MAXAR, for example, collects
+
Different materials reflect light differently. An apple absorbs shorter wavelengths (e.g. blue and green), and reflects longer wavelengths (red). Our eyes use that information– the color– to distinguish between different objects. But our eyes can only see a relatively small sliver of the electromagnetic spectrum covering blue, yellow, and red; we can’t see UV or infrared wavelengths, for example, though the extent to which different materials reflect or absorb these wavelengths is just as useful for distinguishing between them. For example, Astroturf (fake plastic grass) and real grass will both look green to us, espeically from a satellite image. But living plants absorb radiation from the sun in a part of the light spectrum that we can’t see. There’s a spectral index called the Normalized Difference Vegetation Index (NDVI) which exploits this fact to isolate vegetation in multispectral satellite imagery. So if we look at Gilette Stadium near Boston, we can tell that the three training fields south of the stadium are real grass (they generate high NDVI values, showing up red), while the pitch in the stadium itself is astroturf (generating low NDVI values, showing up blue).
+
+
+
+VHR image of Gilette Stadium with Sentinel-2 derived NDVI overlay
+
+
+
In other words, even though these fields are all green and indistinguishable to the human eye, their spectral profiles beyond the visible light spectrum differ, and we can use this information to distinguish between them. Below is a plot of the spectral profiles of different materials, including oil.
+
+
+
The European Space Agency’s Sentinel-2 satellite collects spectral information well beyond the visible light spectrum, enabling this sort of analysis. It chops the electromagnetic spectrum up into “bands”, and measures how strongly wavelengths in each of those bands is reflected:
+
+
We’ll be using this satellite to distinguish between oil and other materials, similar to the way we were able to distinguish between real and fake grass at Gilette Stadium. First, we’ll have to do a bit of pre-processing on the Sentinel-2 imagery after which we’ll train a machine learning model to identify oil.
+
+
+
3.1.3 Temporal Resolution
+
Finally, time There is often a tradeoff between spatial and temporal resolution.
+
The Google Maps basemap is very high resolution, available globally, and is freely available. But it has no temporal dimension: it’s a snapshot from one particular point in time. If the thing we’re interested in involves changes over time, this basemap will be of limited use.
+
The “revisit rate” is the time it takes a satellite to image the same point on earth
Unable to display output for mime type(s): application/vnd.colab-display-data+json
+
+
+
+
6.1 Import geemap library
+
The geemap Python package has two plotting backends: ipyleaflet and folium. A key difference between folium and ipyleaflet is that ipyleaflet is built upon ipywidgets and allows bidirectional communication between the front-end and the backend enabling the use of the map to capture user input, while folium is meant for displaying static data only (source). Note that Google Colab currently does not support ipyleaflet (source). Therefore, if you are using geemap with Google Colab, you should use import geemap.eefolium as geemap.
+
+
import ee
+import geemap.eefolium as geemap
+
+
+
+
6.2 Create an interactive map
+
+
Map = geemap.Map()
+# Map
+
+
+
+
6.3 Add Earth Engine data
+
+
# Add Earth Engine dataset
+image = ee.Image('USGS/SRTMGL1_003')
+
+# Set visualization parameters.
+vis_params = {
+'min': 0,
+'max': 4000,
+'palette': ['006633', 'E5FFCC', '662A00', 'D8D8D8', 'F5F5F5']}
+
+# Print the elevation of Mount Everest.
+xy = ee.Geometry.Point([86.9250, 27.9881])
+elev = image.sample(xy, 30).first().get('elevation').getInfo()
+print('Mount Everest elevation (m):', elev)
+
+# Add Earth Engine layers to Map
+Map.addLayer(image, vis_params, 'DEM')
+Map.addLayer(xy, {'color': 'red'}, 'Mount Everest')
+
+# Center the map based on an Earth Engine object or coordinates (longitude, latitude)
+# Map.centerObject(xy, 4)
+Map.setCenter(86.9250, 27.9881, 4)
+
+
+
+
6.4 Display the map
+
+
Map.addLayerControl()
+Map
+
+
+
+
+
+
+
+
+
+
+
+
+
\ No newline at end of file
diff --git a/_book/images/Figure_1.gif b/_book/images/Figure_1.gif
new file mode 100644
index 0000000..045107f
Binary files /dev/null and b/_book/images/Figure_1.gif differ
diff --git a/_book/images/Landsat.png b/_book/images/Landsat.png
new file mode 100644
index 0000000..087290c
Binary files /dev/null and b/_book/images/Landsat.png differ
diff --git a/_book/images/Maxar.png b/_book/images/Maxar.png
new file mode 100644
index 0000000..9df3286
Binary files /dev/null and b/_book/images/Maxar.png differ
diff --git a/_book/images/NDVI.jpg b/_book/images/NDVI.jpg
new file mode 100644
index 0000000..3cfb24a
Binary files /dev/null and b/_book/images/NDVI.jpg differ
diff --git a/_book/images/S2_bands.png b/_book/images/S2_bands.png
new file mode 100644
index 0000000..914c016
Binary files /dev/null and b/_book/images/S2_bands.png differ
diff --git a/_book/images/Sentinel2.png b/_book/images/Sentinel2.png
new file mode 100644
index 0000000..f68e081
Binary files /dev/null and b/_book/images/Sentinel2.png differ
diff --git a/_book/images/UNEP.PNG b/_book/images/UNEP.PNG
new file mode 100644
index 0000000..2ac302d
Binary files /dev/null and b/_book/images/UNEP.PNG differ
diff --git a/_book/images/WoS Articles.png b/_book/images/WoS Articles.png
new file mode 100644
index 0000000..3c88d8e
Binary files /dev/null and b/_book/images/WoS Articles.png differ
diff --git a/_book/images/big_red.png b/_book/images/big_red.png
new file mode 100644
index 0000000..d1502b2
Binary files /dev/null and b/_book/images/big_red.png differ
diff --git a/_book/images/fao_gaul.jpg b/_book/images/fao_gaul.jpg
new file mode 100644
index 0000000..c9dda77
Binary files /dev/null and b/_book/images/fao_gaul.jpg differ
diff --git a/_book/images/fires.jpg b/_book/images/fires.jpg
new file mode 100644
index 0000000..7100602
Binary files /dev/null and b/_book/images/fires.jpg differ
diff --git a/_book/images/footprints.png b/_book/images/footprints.png
new file mode 100644
index 0000000..fe43f82
Binary files /dev/null and b/_book/images/footprints.png differ
diff --git a/_book/images/ger_zero.png b/_book/images/ger_zero.png
new file mode 100644
index 0000000..b31a4e0
Binary files /dev/null and b/_book/images/ger_zero.png differ
diff --git a/_book/images/ger_zero_pred.png b/_book/images/ger_zero_pred.png
new file mode 100644
index 0000000..7590cf4
Binary files /dev/null and b/_book/images/ger_zero_pred.png differ
diff --git a/_book/images/hasankeyf.gif b/_book/images/hasankeyf.gif
new file mode 100644
index 0000000..5b5794c
Binary files /dev/null and b/_book/images/hasankeyf.gif differ
diff --git a/_book/images/iraq_check.png b/_book/images/iraq_check.png
new file mode 100644
index 0000000..5c03f97
Binary files /dev/null and b/_book/images/iraq_check.png differ
diff --git a/_book/images/kh11.png b/_book/images/kh11.png
new file mode 100644
index 0000000..6e93fef
Binary files /dev/null and b/_book/images/kh11.png differ
diff --git a/_book/images/makeshift-refining.png b/_book/images/makeshift-refining.png
new file mode 100644
index 0000000..42d5729
Binary files /dev/null and b/_book/images/makeshift-refining.png differ
diff --git a/_book/images/mining.jpg b/_book/images/mining.jpg
new file mode 100644
index 0000000..6b2d4de
Binary files /dev/null and b/_book/images/mining.jpg differ
diff --git a/_book/images/mishraq.gif b/_book/images/mishraq.gif
new file mode 100644
index 0000000..5968374
Binary files /dev/null and b/_book/images/mishraq.gif differ
diff --git a/_book/images/mishraq_small.gif b/_book/images/mishraq_small.gif
new file mode 100644
index 0000000..3d39b5f
Binary files /dev/null and b/_book/images/mishraq_small.gif differ
diff --git a/_book/images/points.png b/_book/images/points.png
new file mode 100644
index 0000000..08c43f9
Binary files /dev/null and b/_book/images/points.png differ
diff --git a/_book/images/pop.jpg b/_book/images/pop.jpg
new file mode 100644
index 0000000..b4f3eca
Binary files /dev/null and b/_book/images/pop.jpg differ
diff --git a/_book/images/power.jpg b/_book/images/power.jpg
new file mode 100644
index 0000000..bf3acff
Binary files /dev/null and b/_book/images/power.jpg differ
diff --git a/_book/images/qayyarah_chart.png b/_book/images/qayyarah_chart.png
new file mode 100644
index 0000000..98d60fe
Binary files /dev/null and b/_book/images/qayyarah_chart.png differ
diff --git a/_book/images/radar ships.jpg b/_book/images/radar ships.jpg
new file mode 100644
index 0000000..2194f95
Binary files /dev/null and b/_book/images/radar ships.jpg differ
diff --git a/_book/images/refinery.png b/_book/images/refinery.png
new file mode 100644
index 0000000..9ecd09f
Binary files /dev/null and b/_book/images/refinery.png differ
diff --git a/_book/images/rojava_preprocessed.png b/_book/images/rojava_preprocessed.png
new file mode 100644
index 0000000..ded65d9
Binary files /dev/null and b/_book/images/rojava_preprocessed.png differ
diff --git a/_book/images/small_refinery.png b/_book/images/small_refinery.png
new file mode 100644
index 0000000..1e0fec5
Binary files /dev/null and b/_book/images/small_refinery.png differ
diff --git a/_book/index.html b/_book/index.html
new file mode 100644
index 0000000..b870c74
--- /dev/null
+++ b/_book/index.html
@@ -0,0 +1,475 @@
+
+
+
+
+
+
+
+
+
+
+
+Google Earth Engine for OSINT
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
The analysis of satellite imagery is a foundational element of open source investigations. Though no-code platforms such as Sentinelhub have been invaluable in allowing the OSINT community to access and process satellite imagery, the analytical capabilities of these platforms are limited. Google Earth Engine (GEE) is a cloud-based platform that stores petabytes of satellite imagery from a variety of sources and allows users to perform advanced analyses on Google servers for free using a browser-based interface. This textbook is designed for investigators who want to perform more sophisticated analysis using geospatial data, and assumes no prior knowledge of coding or remote sensing (satellite imagery analysis). It is organized into four chapters:
+
+
Data Acquisition
+
Getting Started
+
Algorithms
+
Application Development
+
Case Studies
+
+
+
What is Google Earth Engine?
+
As geospatial datasets—particularly satellite imagery collections—increase in size, researchers are increasingly relying on cloud computing platforms such as Google Earth Engine (GEE) to analyze vast quantities of data.
+
GEE is free and allows users to write open-source code that can be run by others in one click, thereby yielding fully reproducible results. These features have put GEE on the cutting edge of scientific research. The following plot visualizes the number of journal articles conducted using different geospatial analysis software platforms:
+
+
Despite only being released in 2015, the number of geospatial journal articles using Google Earth Engine (shown in red above) has outpaced every other major geospatial analysis software, including ArcGIS, Python, and R in just five years. By storing and running computations on google servers, GEE is far more accessible to those who don’t have significant local computational resources; all you need is an internet connection.
+
GEE applications have been developed and used to present interactive geospatial data visualizations by NGOs, Universities, the United Nations, and the European Commission.
Wim Zwijnenburg wrote an excellent Bellingcat article on the subject, which you should read before going any further in this tutorial. In the article, Wim notes that these facilities “can be spotted by the ditch and the black spot at the end with oil waste residues, which blacken the soil around the furnace.” These refineries also frequently leak, blackening large swaths of land around them.
This image shows two clusters of makeshift refineries which were identified by the model. This is good, though we can only get so far by visually inspecting the output from our model. To get a better sense of our model’s performance, we can use the validation data that we generated previously. Remember, these are labeled points which our model was not trained on, and has never seen before.
View source on GitHub
Notebook Viewer
Run in Google Colab