Climate induced migration in Africa

The climate crisis, in the form of more extremes like heatwaves, droughts and floods, impacts not only the biosphere but also the anthroposphere (that space where humans live). The agenda of the United Nations Framework Convention on Climate Change (UNFCC) contains an item dedicated to migration, displacement and human mobility. The sustainable development goal number 10 is specifically covering how to reduce inequalities. Recently, the AI community has made decisive steps in trying to help (AI4Good and other fora). Climate-induced displacement is a huge problem: 25 million displacements per year, 95% from vulnerable regions. Use Case 2 is focused on uncovering the underlying mechanisms of how changing climate conditions cause migration through modern observational causal inference models to understand, quantify and predict migration flow effects from socioeconomic contextual information as well as from environmental variables extracted from Earth Observation data.

There is a growing number of media reports assuming the link of climate change, conflicts, and forced migration. The issue is that this kind of cause-effect chain is not assessed empirically from reliable data and employing rigorous advanced statistical tools. At present, there is no theoretical approach which adequately represents the causal mechanisms through which climate change induces human displacement and migration flows. This climate induced displacement problem needs to be tackled with advanced AI methods of regression, interpretability and causality.

The overarching goal is to model, anticipate and understand migration flows from reliable data. This leads to this Use Case’s main objectives:

  1. collect and identify a large set of potential explanatory variables
  2. develop and apply causal discovery methods to understand variable relations
  3. infer narratives and prediction models from the causal graphs.

The causal inference techniques implemented in the wildfire and drought modelling developed for the climate-induced drought Use Case, as well as the eXplainable AI (XAI) for Deep Learning (both GANs, U-Nets and LSTMs) implemented in the wildfire and drought modelling, will be part of an open source suite. Causal inference methodologies and the developed modules will be part of the “CauseMe” causality platform (https://causeme.uv.es/) hosted by the University of Valencia, to benchmark causal discovery methods.

This Use Case will approach the problem in 3 main steps, yielding a set of useful products: 1) Harmonize explanatory data from diverse EO & non-EO data sources, 2) Derive causal graphs by the systematic characterization of data dependences and deployment of causal inference methods, and 3) Translate causal graphs into explanatory narratives, storylines, and prediction models taking into account only causally relevant features.

Use Case Leader
University of Valencia

Interested in learning more? Contact us!
Dr. María Piles, Maria.Piles@uv.es