How we found migrants affected by climate-driven disasters
To understand how climate change may be influencing irregular migration to the United States, Columbia Journalism Investigations (CJI) and Documented spent nearly a year analyzing more than nine million records of people apprehended by the U.S. Border Patrol from 2010 to 2024 under U.S. Code Title 8, a classification of U.S.-Mexico border encounters. The federal Customs and Border Protection (CBP) dataset, obtained through Freedom of Information Act requests by researchers at the University of Virginia and CJI, included the reported birthplace of each person apprehended, and represents the most detailed source available on where border crossers were born down to the locality, including city and town names as well as states, departments and municipalities...
The data has gaps. The location names are recorded at the border by CBP agents — who generally do not speak foreign languages — and during times of high crossings are not always reliably filled out. Across 190 countries, CBP recorded more than four million unique city names. Many were misspelled, abbreviated or inaccurate, or had provinces listed instead of cities. In the dataset, about 35% of the entries were left blank and less than 1% were listed as “Unknown.”
We chose to analyze the CBP dataset because of its city- and state-level information, which is not available in nationwide border statistics. While it reflects only apprehensions under U.S. Code Title 8, it is also the only government dataset that consistently documents the birthplace of people stopped at the border. The dataset provides the most granular and verifiable geographic information about migrants’ origins. This means our totals are lower than figures that include all southwest border encounters, which encompass people deemed “inadmissible” at ports of entry or processed under Title 42 during the pandemic.
It’s important to note that migrants who qualify for Temporary Protected Status (TPS) due to natural disasters generally fall into the inadmissible category. But we did not include TPS recipients in our analysis because that status is granted to migrants who are already in the United States and the three countries at the center of our analysis — Guatemala, Bangladesh and Senegal — have not received TPS designations between 2019 and 2024.
Despite these limitations, we spoke to nine climate and migration experts who agreed that this is the best government data currently available. As part of a forthcoming study on how climate and environmental changes drive migration dynamics in Central America, Fabien Cottier, a political scientist at Columbia University and the University of Geneva, cleaned the CBP dataset for Mexico, Guatemala, El Salvador, Honduras, Nicaragua, Belize, Costa Rica and Panama, which accounts for two thirds of the data from 2010 to 2022. Cottier developed an automated system to standardize birthplace data by correcting misspellings, resolving phonetic errors and matching each location to the correct city or state level. He shared the clean dataset with us and helped vet our geo-coding methodology. He also collaborated with us on the analysis underpinning the story.
Building on that foundation, we expanded the data-cleaning process across 181 countries. We cleaned roughly 45% of the dataset using automated tools, and reviewed the results by hand. The automation involved comparing place names in a flexible way that could catch spelling differences — for example, recognizing that the words “Quetzaltenango” and “Quetzaltengo” refer to the same city. We caught spelling variants and used custom dictionaries to identify Indigenous and local language names. Then, we geocoded checks to validate coordinates.
For ambiguous cases, such as towns near international borders or names with multiple possible matches, we manually reviewed entries. Country by country, we merged variants – there were 40 versions of the place “Huehuetenango” in Guatemala, for example, and dozens of French and Arabic influenced spellings in Senegal. We verified each entry against trusted geographic sources, like the geographical database GeoNames.
Once cleaned, we analyzed the dataset to identify localities where more migrants had left to cross the U.S. southern border than their countries’ national average. Rather than compare raw counts of migrants between cities, which would favor more populated areas, we looked to measure the growth rate — or the percentage change in border apprehensions from each city over time, relative to that country’s overall trend. This enabled us to control for population size because it highlights how much migration has increased (or decreased) within each locality. Outliers could point to places where climate shocks or other pressures may be disproportionately driving migration. To place these migration flows into context, we compared the number of people apprehended at the U.S.-Mexico border who had these localities listed in the data as their City of Birth and Country of Birth in 2019 versus 2024, spanning the end of Donald Trump’s first presidency, when border crossings first surged, and Joe Biden’s term, when they rose to record highs.
The broader 2010–2024 dataset was used to identify longer-term patterns and to link localities to repeated climate disasters over time, while the 2019–2024 comparison isolates more recent migration dynamics. Since the CBP data only records place of birth, not last residence, it does not capture internal moves within countries before crossing the border.
Our list of outlier localities included places like Quetzaltenango in Guatemala, Feni in Bangladesh and Diourbel in Senegal. In total, our final list highlighted 15 Bangladeshi, 96 Senegalese and 683 Guatemalan cities of origin.
To understand the climate pressures in these same locations, we turned to the Emergency Events Database, or EM-DAT, a widely cited international disaster database created in Belgium by the Centre for Research on the Epidemiology of Disasters (CRED) and the World Health Organization (WHO). The database compiles information on major disasters from United Nations agencies, aid groups, government reports and other official sources. We filtered for weather-related events that climate scientists agree are accelerating and intensifying because of the warming planet, such as floods, storms, drought and extreme heat, while excluding categories that were not related like earthquakes, volcanic activity, infestations and epidemics.
To avoid exaggerating one-off events, we compared the 2019-2024 range for 3 plus climate catastrophes with the larger 2010-2024 range and did calculations for both. We extended the analysis to earlier years, since experts noted that migration decisions often unfold slowly, and people may first move internally before crossing borders. Also, the CBP data only captures place of birth, not intermediate and internal movements.
By merging the two datasets, we created a first-of-its-kind map of more than 80 cities where migration and climate disasters overlap. The results show a striking pattern: many of the same towns and cities driving higher-than-average migration flows to the U.S. have also faced repeated hurricanes, cyclones, storms, floods and droughts over the past decade. We used this list as a reporting tool to find recent migrants living in New York City whose experiences with climate disasters might have influenced their own migration decisions.
Cottier and other researchers have found mixed results on climate change’s influence on migration in Central America because of the different reasons why people decide to leave their countries. It’s important to note that our analysis does not control for social or economic differences between localities. Yet it responds to a common concern about selection bias by using disaster data to explore whether places repeatedly affected by climate shocks were also the ones sending more migrants. Rather than establishing causality, the analysis serves to identify potential overlaps between areas hard hit by climate-driven disasters and those showing above-average growth in migration, helping guide where in person interviews might be most relevant and revealing.
This analysis can’t prove that hurricanes, floods or droughts forced someone to leave; migration is always shaped by individual characteristics and a range of economic, demographic and other factors. But it reveals how climate disruption is quietly and consistently pushing vulnerable communities closer to the breaking point around the world.
— Carla Mandiola