“We all live under the same sun,” Moussa thought as he loaded the last bundles onto the cart. “That is what my father used to say, and his father, and his father before him; but the sun has changed.” Moussa cast one last disconsolate glance at the land that was his and had belonged to his ancestors, then absent-mindedly helped his wife and his daughter, the beautiful Fadoul, with the loading. […] He was doing it above all for them; had it been up to him, he would have stayed. Yes, because he had done everything he could […]. Yet, however carefully one chooses the best land and tends it as if it were one’s own child, if the rain does not come… it does not come. And if the rain then decides to fall all at once in a single day and sweeps away the stacks of millet, the blessing turns into misfortune. […] Where he had sown a garden, he thought, he was harvesting only uncertainty. […] It was no longer possible to plan a future there.
Moussa’s story, told in the book Effetto serra, effetto guerra (Greenhouse effect, war effect) by Grammenos Mastrojeni and Antonello Pasini, is not only an individual story. His experience of drought reflects a broader phenomenon: every year, millions of people are forced to leave their homes because of environmental disasters of different kinds. Between 2008 and 2023, more than 403 million internal displacements1 dcaused by disasters were recorded, almost half of them triggered by floods2, which are the leading cause of disaster displacement worldwide. As climate change intensifies extreme events, the number of displaced people appears set to increase.
Behind these figures, however, lie decisions, vulnerabilities and individual trajectories that global models cannot describe. Although increasingly accurate hydrological models, early warning systems and maps now allow us to estimate the extent of an event and the number of people potentially exposed, understanding how water behaves is only part of the equation when working with national and local actors involved in territorial risk management or humanitarian action.
Why do some households evacuate while others remain? Why are some able to return within a few months, while others experience prolonged displacement? How much do housing quality, income, crop losses, trust in warnings or attachment to place influence these outcomes?
These questions directly affect the ability of institutions, local authorities and humanitarian organisations to prevent, plan for and respond to crises. It is within this nexus between scientific research and humanitarian action that the work of Eleonora Panizza on flood displacement risk in Sudan is situated. Panizza is now a researcher at CIMA Research Foundation and carried out the study while pursuing her PhD at the University of Genoa, hosted by the Foundation. The work was conducted with contributions from Programme Director Roberto Rudari, researcher Mirko D’Andrea and Yared Abayneh Abebe, a researcher at TUDelft and, at the time, a project consultant. It formed part of the project Addressing Drivers and Facilitating Safe, Orderly and Regular Migration in the Contexts of Disasters and Climate Change in the IGAD Region, coordinated by the United Nations Office for Project Services (UNOPS) and funded by the Multi-Partner Trust Fund (MPTF).
By collecting field data and developing an Agent-Based Model3 calibrated to the local context and the specific characteristics of the sample, the study helps to improve understanding of displacement dynamics and provides evidence relevant to planning and risk reduction. Humanitarian action needs science not only to understand how floods behave, but also how people behave.
Observing people, not only floods
“Understanding how people respond to a flood requires a shift in perspective. When risk models are developed, information on hazard, exposure and vulnerability is combined. The challenge, particularly in models developed at global or national scale, is to represent vulnerability in sufficient detail, because it varies from one community to another and even from one household to another. This is where field data become essential,” explains Eleonora Panizza.
“The agent-based model allows us to represent each household as an ‘agent’: an autonomous unit with its own characteristics – such as income, housing quality, livelihoods or access to information – which influence decisions during a flood. Rather than describing an average population response, we can simulate interactions among different households, each with its own vulnerabilities and response capacities.”
This was the approach adopted for the research conducted in Sudan. Between June and July 2022, 300 households were interviewed across seven rural communities in Khartoum State. The sites were selected to represent different contexts sharing high flood exposure: long-established villages along the Nile, rural settlements and a camp hosting communities that had already been displaced.
The survey also collected information rarely captured in large international databases: household composition, socioeconomic conditions, housing characteristics, access to essential services, livelihoods, flood experience, displacement dynamics, risk perception and adaptation strategies. These data were combined with geospatial information and flood scenarios, creating a knowledge base that links the physical and social dimensions of the event.
The result is a high-resolution dataset that makes it possible to examine disaster displacement from a different perspective: not only how many people move, but who moves, when, for how long, why, and which factors influence that decision. This knowledge provides the foundation for developing models that represent community behaviour and support more effective humanitarian planning.
Risk begins before the emergency
Understanding the effects of a disaster on people and the dynamics of displacement requires attention to the conditions that precede the emergency. This means going beyond the characterisation of exposure and vulnerability typically used in risk assessment and moving closer to the local realities of individual households.
In the seven rural communities in Khartoum State included in the study, vulnerability is part of everyday life. Thirty-eight per cent of households live in extreme poverty, one in five lives in temporary or inadequate housing, and only 51% has stable access to healthcare. Many households also depend on agriculture and livestock, livelihoods that are particularly exposed to flood impacts.
These data show that risk does not depend solely on the intensity of the natural event. It also emerges from the interaction between exposure, vulnerability and response capacity, at both community and household level. Two households may face the same flood yet experience profoundly different consequences. Changing the scale of analysis makes it possible to capture displacement in greater detail.
Leaving and returning: what shapes the decision?
One of the central questions for humanitarian actors is who will move and who will remain in the affected area. The study findings show that there is no single answer.
The 65% of the households interviewed had been forced to leave their homes at least once because of flooding. Of these, only 7% evacuated in advance; in almost all other cases, displacement occurred during or immediately after the event. Severe damage to housing was the main driver of displacement, followed by the loss of livelihoods, including crops and livestock.
Within the surveyed sample, 21% of households did not leave their homes at any stage despite being exposed to and/or affected by flooding, while the remaining 14% did not experience conditions severe enough to trigger displacement.
Perhaps the most significant finding concerns the 21% of households that did not move despite being at risk or experiencing flood impacts. Remaining does not necessarily mean being safe. Some people do not move because they lack the financial or material resources required; others choose to remain to protect their homes, land or animals, on which their livelihoods depend. Distinguishing between these situations is essential, because they call for very different institutional and humanitarian responses.
Moreover, the emergency phase is only the beginning of what is often a much longer process.
More than half of the households included in the study were able to return within six months, but for 38% displacement lasted more than a year. During this period, the main difficulties concerned access to safe housing, the recovery of income and livelihoods, and the availability of water and essential services.
This means that humanitarian response cannot be limited to immediate assistance. Understanding the duration of displacement and the difficulties associated with return is equally important for planning effective interventions and supporting the long-term recovery of communities.
From data to models: when research becomes a decision-support tool
Collecting field data was not the endpoint of the research, but its starting point.
The Agent-Based Model developed from the collected data acts as a virtual laboratory in which different scenarios can be explored and changes in displacement dynamics can be observed as conditions vary. It incorporates both the evolution of the flood and the characteristics of people, their resources, risk perceptions and possible decisions.
“A model is not intended to predict with certainty what each individual person will do. It helps us understand how behaviours change as conditions change, and which patterns emerge. We start from the ‘micro’ level – the data collected from individual households – and move towards the ‘macro’ level, observing how interactions among specific data generate collective dynamics and trends that help assess the effectiveness of different risk reduction strategies and support more informed decisions,” Panizza continues.
“Using the agent-based model developed for Sudan, we can also simulate different policy scenarios – such as strengthening Early Warning Systems, providing financial support to households or promoting resilient housing reconstruction – and assess how each may influence people’s behaviour, generating evidence to guide risk reduction interventions and strategies.”
“Scientific research is never an end in itself. Its value lies not only in producing new knowledge, but also in its potential to turn that knowledge into a means of identifying critical issues, detecting correlations and informing political, strategic and humanitarian decisions. This is also what we are taking forward in Mozambique,” she concludes.
Building on the work carried out in Sudan, a new phase of the research is now under way in Mozambique. Here, the project combines the scientific component with a strong operational and cooperation dimension, including activities to share findings and engage with institutions, local authorities and humanitarian actors in support of planning and risk reduction.
It is within this range of possibilities that scientific research, humanitarian action and institutional decision-making meet. Field data are not merely the output of a single study: they become a resource for strengthening planning and early warning systems, supporting risk reduction policies and evidence-based decisions and, where possible, anticipating community needs before a crisis unfolds.
Studying people’s behaviour therefore contributes to more informed risk management throughout the disaster risk management cycle, from prevention and mitigation to preparedness and emergency response. In this sense, scientific research becomes a means of informing action.
- See https://www.internal-displacement.org/internal-displacement/ ↩︎
- Source: IDMC (2025). Global Internal Displacement Database (GIDD), accessed 9 April 2025, https://www.internal-displacement.org/database/displacement-data/ ↩︎
- An Agent-Based Model (ABM) is a simulation model that represents individuals, households or other entities as autonomous “agents”, each characterised by specific attributes and behavioural rules. By simulating interactions among these agents and with their surrounding environment, the model makes it possible to analyse how decisions made by individual agents can generate collective dynamics and to assess the effects of different scenarios or intervention strategies. ↩︎