A generalized simulation development approach for predicting refugee destinations

Abstract

In recent years, global forced displacement has reached record levels, with 22.5 million refugees worldwide. Forecasting refugee movements is important, as accurate predictions can help save refugee lives by allowing governments and NGOs to conduct a better informed allocation of humanitarian resources. Here, we propose a generalized simulation development approach to predict the destinations of refugee movements in conflict regions. In this approach, we synthesize data from UNHCR, ACLED and Bing Maps to construct agent-based simulations of refugee movements. We apply our approach to develop, run and validate refugee movement simulations set in three major African conflicts, estimating the distribution of incoming refugees across destination camps, given the expected total number of refugees in the conflict. Our simulations consistently predict more than 75% of the refugee destinations correctly after the first 12 days, and consistently outperform alternative naive forecasting techniques. Using our approach, we are also able to reproduce key trends in refugee arrival rates found in the UNHCR data.

Figure 1

Simulation development approach for predicting the distribution of refugee arrivals across camps.

Figure 2

Overview of geographic network models for (a) Burundi, (b) Central African Republic and (c) Mali. Models contain conflict zones (red circles), refugee camps (dark green circles), forwarding hubs (light green circles) and other major settlements (yellow circles). Interconnecting roads are given in a simplified straight-line representation, with adjacent blue numbers used to indicate their length in kilometres. Background maps are courtesy of https://carto.comcarto.com created using OpenStreetMap data that is further modified with the use of https://inkscape.org/en/release/0.91Inkscape0.91.

Figure 3

Comparison of number of refugees in camps between the simulation and the data (left column), and overview of the averaged relative difference between simulation and data (right column). The averaged relative difference across camps between simulation and data is given by the red line. We provide these comparisons respectively for (a,b) the Burundi simulations (top row), (c,d) the CAR simulations (middle row) and (e,f) the Mali simulations (bottom row).

Figure 4

Number of refugees as predicted by our simulation and obtained from the UNHCR data for the Burundi conflict. (a–e) Graphs are ordered by camp population size, with the most populous camp on the top to the smallest one on the bottom.

Figure 5

Number of refugees as predicted by our simulation and obtained from the UNHCR data for the CAR conflict. (a–h) Graphs are ordered by camp population size, with the most populous camp on the top to the smallest one on the bottom (see remaining six camps in Fig. S1).

Figure 6

Number of refugees as predicted by our simulation and obtained from the UNHCR data for the Mali conflict. (a–g) Graphs are ordered by camp population size, with the most populous camp on the top to the smallest one on the bottom.