The Story

Between 2002 and 2020, the United States conducted over five thousand airstrikes across Afghanistan, Pakistan, Yemen, and Somalia. The Bureau of Investigative Journalism documented each one — the dates, the locations, the casualties. Their record is the most comprehensive public accounting of a campaign that three presidents escalated, none ended, and no administration fully disclosed.

The Escalation

Each administration expanded what the previous one started. Under Bush, strikes were rare — 59 total across eight years, almost all in Pakistan. Obama industrialized the program: one strike every four days in Pakistan, a parallel campaign in Yemen, the first drone strikes in Somalia. By his second term, 970 strikes had killed between 6,067 and 8,755 people.

Then Trump removed the constraints. In 2017 alone, Somalia saw more strikes than the previous decade combined. Afghanistan, where the US had barely used drones before 2015, was hit with 2,761 strikes in a single year — 2019. The total under Trump: 4,155 strikes in four years.

Each president inherited a drone program. Each left it larger than they found it.

The Civilian Cost

The official framing was precision. The record says otherwise. Across all four countries, between 977 and 2,300 civilians were killed, including 326 to 480 children. In Pakistan, only 4% of those killed were ever identified as al-Qaeda members. In Afghanistan, 77% of strikes reported zero casualties — not because they missed, but because no one was counting.

9,418–17,800 Total killed
977–2,300 Civilians killed
326–480 Children killed

The deadliest single strike for children hit a madrassa in Bajaur, Pakistan, on October 30, 2006. Between 68 and 70 children were killed. The youngest was seven. In Yemen, a cruise missile strike on al-Majala killed 44 civilians, including 22 children and five pregnant women. The youngest was one year old.

The Accountability Vacuum

In June 2011, John Brennan claimed there had not been "a single collateral death" from drone strikes in nearly a year. The Bureau documented 45 to 56 civilian deaths during the same period. When Afghanistan's data reporting stopped entirely in October 2017, no explanation was offered. Of 2,494 people killed in Pakistan, 1,765 remain nameless.

Of 28 identified militant locations targeted in Pakistan, 27 were incorrect.
— Retired Pakistani Brigadier

The pattern repeated in every theater: official claims of precision, followed by documented evidence of civilian harm, followed by silence. In Yemen, a UN investigator called the CIA's involvement "an almost insurmountable obstacle to transparency." In Somalia, only a third of reported strikes were independently verified.

The Paradox

The strikes were supposed to defeat militant organizations. In practice, they often strengthened them. In Pakistan, child casualties served as "recruitment agents for militants." In Somalia, strikes gave al-Shabaab "almost a legitimacy." In Yemen, a journalist told the US Senate: "Drone strikes are the face of America to many Yemenis."

Despite years of sustained bombing, the Taliban continued territorial gains in Afghanistan. A "constant flow of people" kept joining IS-Khorasan despite military pledges to eliminate them by 2017. The pattern persisted across every theater and every administration.

Four Countries, Four Arcs

Afghanistan

4,238 strikes

The largest and least documented campaign. Excellent data at the province level, near-zero at district or location. 77% of strikes report zero casualties. The US military reported at province level — beneath that, a void.

Pakistan

430 strikes

The longest-running and most scrutinized program. Best detail at all levels. The campaign peaked under Obama's first term — 311 strikes — and virtually ended under Trump. 92.6% of strikes hit North and South Waziristan.

Yemen

326 strikes

Good province coverage, patchy below that. The Arab Spring surge of 2011-2013 gave way to Trump's unprecedented escalation — 100+ strikes in 2017, more than the previous four years combined. Twenty strikes in a single morning.

Somalia

190 strikes

The weakest overall documentation, but the most improvable. 77% of strikes occurred under Trump. The lowest confirmed civilian casualty rate — which may say more about reporting than restraint.

The Data

View

Individual strikes and the limits of the record. Regions show "somewhere in here."

Resolution

How precisely do we know where each strike landed?

Showing

Which casualty type drives the size and color of markers.

Countries
Estimate

TBIJ records a range for every casualty figure. Choose which end to display.

Nov 2002 — Mar 2020
Filter by presidential administration

The Gap

The data doesn't just show what happened. It shows the shape of what we can't see — the strikes with no casualties reported, the deadliest attacks with no journalism, the ranges so wide they tell two different stories.

The Zero-Casualty Paradox

77.4% of Afghanistan strikes report zero people killed. That's 3,281 out of 4,238 strikes. These aren't misses — they're the reporting gap. The US military was the primary source, and they simply didn't count.

The Strikes Nobody Wrote About

The five deadliest strikes for children. The top two have TBIJ articles investigating what happened. Ranks 3 through 5 — all Afghanistan, 2016 to 2018 — have none. The data records the deaths. No journalism investigated them.

B6 — Pakistan
October 30, 2006
68–70 children killed
Bajaur madrassa. Youngest victim: age 7.
Article exists
YEM002 — Yemen
December 17, 2009
21–22 children killed
Al-Majala cruise missile. 5 pregnant women among dead.
Article exists
AFG0284 — Afghanistan
November 3, 2016
20 children killed
No article
AFG0623 — Afghanistan
September 23, 2018
11 children killed
No article
AFG0389 — Afghanistan
August 30, 2017
10 children killed
No article

The Range Problem

Every casualty figure in the TBIJ database is a range — a minimum and a maximum. The gap between them is the uncertainty. Choosing one end or the other tells a very different story.

About

The Data

All strike data comes from The Bureau of Investigative Journalism (TBIJ), which maintained the most comprehensive public database of US drone and air strikes from 2004 to 2020. Their methodology combined media reports, legal documents, leaked cables, and field reporting to document each strike's date, location, and casualty toll.

This project presents their data interactively. The journalism is theirs; the visualization and analysis of what's missing is ours.

What This Shows

5,184 strikes across Afghanistan (4,238), Pakistan (430), Yemen (326), and Somalia (190). Geographic boundaries come from geoBoundaries and the NGA GeoNames database, matched to TBIJ location data through fuzzy string matching.

What This Can't Show

  • Afghanistan district-level data is 89% unknown — the US military reported at province level, so only 465 of 4,238 strikes have district locations.
  • All casualty figures are ranges — the gap between minimum and maximum can be enormous (Trump-era Afghanistan: 1,759 to 7,069 killed).
  • 77% of Afghanistan strikes report zero casualties — this reflects reporting, not reality.
  • Geographic resolution varies sharply by country — Pakistan has excellent location data; Afghanistan has almost none below province level.

Geographic Confidence

Geographic confidence: percentage of strikes resolved at each administrative level
LevelAFPKYESO
Country100%100%100%100%
Province99.5%99.5%92.0%75.3%
District8.8%99.5%39.3%51.1%
Location4.2%65.6%26.7%34.7%

Percentage of strikes resolved to each geographic level. The pipeline's match rates are 78–100% wherever source data exists — the constraint is what TBIJ's sources recorded, not the matching process.

Articles

159 TBIJ articles are cross-referenced to the strike database. 57 directly match to specific strikes; 7 cover related periods; 24 provide broader context. 71 are non-strike pieces covering legal, policy, and methodology topics. Articles surface as annotations on the map when viewing matched strikes.

Data: The Bureau of Investigative Journalism
Geographic boundaries: geoBoundaries & NGA GeoNames