I used approximately 1 TB of satellite imagery to capture how nighttime lights have changed since 2012 for the entire world. A fellow lab member Brandon Prehn (former light infantryman in the US Army) spotted some bombed villages in Syria on my map.  We decided to explore a bit more.  Just by looking at the map alone with no labels or administrative boundaries, he recognized more destroyed cities.  Then we looked at northern Iraq, Puerto Rico, Ukraine, South Sudan, Bangladesh, and the western Rakhine state of Myanmar...

We were shocked.

At this very moment, we still have people on this planet who are switching the lights off in their homes, and not because they need to leave for work, or because they want to save the environment. Rather, they are displaced and have to flee their homeland. In 2017, one in every four of the world’s school-age children were living in a place affected by a current humanitarian crisis. There are nearly 500 million of them who live among horrors caused by natural disasters, local conflicts and ongoing hardships. In 2016, there were 31 million new instances of internal population displacements, equivalent to one person forced to flee every single second. Hurricane Maria swept through the Caribbean in late September 2017, causing half of the population of Puerto Rico, presumably a community with limitless disaster relief and funding as a U.S. Territory, to remain without power and running water for what has now been 4 months. Against the backdrop of brightly illuminated regions in the rest of the world, I shamefully cannot even name the places where clusters of lights have dimmed towards darkness. Pictures of bombed buildings, rubble-strewn streets, and flooded villages cannot alone summarize the scale of tragedy and destruction that the effects of an artillery bombardment have on a square city block dense with residential housing. It is hard for many of us to recognize the horror that people are living through every minute of every day. Knowledge of the horrific conditions experienced during warfare within our homes is long gone, or nearly beyond living memory. In this time of near-limitless access to information, we cannot let ourselves become desensitized to suffering. So I decided to map the dark.

How did I do it:

Step 1: Data and pre-processing. I used the latest nighttime light imagery from the Visible Infrared Imaging Radiometer Suite (VIIRS) Day/Night Band (DNB). The time series is a monthly “vcm” series for the most recent 6 years (i.e. January 2012-December 2017). The “vcm” version excluded any data impacted by stray light and therefore was used for all subsequent processes. All DNB bands have been filtered to remove data impacted by lightning, lunar illumination, and cloud cover. Before compositing, all images have been re-projected to a Sinusoidal world projection with a 900-meter cell size for processing. All map layers have been resampled for display purposes from the 900-m observation scale to a geographic grid. A total of nearly 1 TB of data were processed to cover the entire globe.

Step 2: Temporal trend identification. The temporal depth of DNB monthly composites allowed me to chronicle changes in brightness on a monthly basis. Rather than simply comparing two images between the initial and end points of the period, I calculated a pixel-wise temporal trend utilizing every single cell possible in the entire image archive. The advantage of such a process was to take into account variations of DNB temporal trajectories therefore ultimately quantifying the rate of DNB changes. A total of 6 annual averaged composites were created for each year from 2012 to 2017. To demonstrate the most recent changes in nighttime lights, a monthly stacked image composite was also generated to capture temporal light source variations for 2017. A Mann-Kendall non-parametric test was used to determine the significance of the monotonic trend in the DNB image stacks. The Theil-sen trend slope was calculated using only pixels identified by Mann-Kendall test as statistically significant (p < 0.05). Instead of using the temporal average value, the Theil-sen slope computed the median of all pairwise temporal periods therefore largely minimizing the negative impact caused by potential outliers. A steeper temporal DNB trend slope represents faster, more dramatic changes in illumination while a flatter trend slope indicates a much stable status of nighttime light conditions.

Step 3: Visualization. I followed one of John Nelson’s ESRI blogs on firefly cartography using ArcGIS Pro with all NASA imagery packed in one ArcGIS Pro project file.

I generated a global light trend map that is capable of not only showing where the lights are on and off in a binary fashion but also statistically quantifying the rate of such changes. The results are striking yet tragic, particularly in places where there were instances of massive population displacements throughout the past 6 years. Below, you will see an overview of the temporal trend of nighttime lights over the entire world from 2012 to 2017. I have also highlighted 3 events – 3 stories of what happened in the dark.