More than seven million Venezuelans have fled economic and political turmoil in their country since 2015, with some 400,000 living in Brazil as of 2023. Many Venezuelans regularly cross the border between these countries for economic or family reasons. However, border crossings came to an abrupt halt in early 2020 when COVID restrictions closed the border. Thousands of people had already camped out on the Venezuelan side, waiting to cross, and no one knew how long the border would remain closed – or what might happen when the situation changed. 

In anticipation of the border reopening, UN Global Pulse teamed up with UNHCR, the UN Refugee Agency, to forecast how many people would enter Brazil from Venezuela and how this would affect the need for UNHCR’s shelter system.

Our impact

We capitalised on the data analytics that UN Global Pulse had refined over more than a decade to estimate how many people might intend to leave their homes and head for the border, and to develop a queue modelling tool that could simulate border crossings under various conditions. Since UNHCR could not collect first-hand information on crossings while the border was closed, we used social media activity, internet search queries,  and other public information on Venezuela’s political and economic environment to monitor the situation on the ground, an approach which is referred to as “now-casting”.

The project helped UNHCR evaluate the likelihood of different scenarios they were considering – each of which called for different preparations.

Our lessons

  • Now that COVID-related entry restrictions have been lifted, we’re better able to understand normal movement patterns between the two countries and adapt our predictive models to support ongoing operations. In dynamic border settings like this one, predictive models might need to be updated constantly to keep pace with changing realities on the ground – for example, we had to revise our models as the borders reopened, and due to changes in the variables used to measure border crossings.
  • Over the course of the project, we have continued to refine our models and simulations to make them more understandable and easier for policymakers to use. These tools can be helpful for contingency planning and advocacy purposes, so it is important that they be interpretable for a wide variety of users, and co-developed with operational teams on the ground.
  • We also want to extend the approach developed here to other forced displacement and disaster contexts, emphasising the need for flexibility. In the face of uncertainty, we’ve found that incorporating different models and perspectives avoids overreliance on one set of forecasts and provides insight into many possible outcomes. For example, we incorporated simulations and nowcasting to supplement predictions, and we included uncertainty intervals with our predictions to provide a range of possible estimates.