Fig 1.
Forced migration process, drivers, and data source: Authors’ adaption based on previous studies [32–34].
Fig 2.
Asylum seekers (per 100 origin population).
A: Asylum-seeking rate by origin–destination dyad flows. B: Asylum-seeking rate by countries of origin.
Fig 3.
Conflict, economic, and climate data.
A: Trends in conflict–induced mortality rates. B: Trends in GDP per capita. C: Trends in risks of extreme dry and wet weather condition derived from the Standardised Precipitation–Evapotranspiration Index (SPEI).
Fig 4.
Data splits for cross-validation.
Fig 5.
Coefficient estimates by FE and FTG models.
Note: The solid bands in FE model represent the 95% confidence intervals of the point estimates, and that in FTG model contain 95% of the estimated flow-specific coefficients.
Fig 6.
Each point corresponds to the observed and predicted log ASR for a given flow and in a given year. Points’ colors differentiate the length of training data (as shown in Fig 4). The black solid lines represent 1:1 relationships.
Fig 7.
Effect of panel data length on model performance.
Changes in training and testing errors (RMSE) with respect to the length of training data (as shown in Fig 4).
Fig 8.
Observed and predicted temporal trends of forced migration.
The observed data are represented by dots. The selected flows represent the top 10 asylum-related migration flows to Europe during 2008-2019. The temporal trends are predicted by the model trained on the data 2008-2018 (i.e., the longest data split in Fig 4).