DRC AI Displacement Forecasting
DRC AI Displacement Forecasting: Foresight, AHEAD & Predictive Models for Anticipatory Action
The Danish Refugee Council (DRC) has developed four machine learning models — Foresight, AHEAD, SODRD, and SPIN — to predict forced displacement driven by conflict, governance fragility, climate stress, and economic shocks. Covering 27+ countries (over 90% of global displacement), the models achieve remarkable accuracy: more than half of Foresight's annual forecasts are within 10% of actual displacement. These tools enable humanitarian actors to shift from reactive crisis response to anticipatory action, informing strategic planning, resource allocation, and early warning systems.
The Global Displacement Challenge: From Reactive to Predictive
Forced displacement has reached unprecedented levels, with conflict, climate stress, governance fragility, and economic shocks driving millions from their homes. The current humanitarian architecture rests on a fragile assumption: that neighboring low- and middle-income countries will continue absorbing the majority of displacement despite persistent violence and contracting external financing. In 2025, large-scale returns reduced aggregate figures without addressing structural drivers, creating renewed displacement risk.
The core problem DRC set out to solve:
Humanitarian planning has historically relied on reactive assessments and expert judgment. DRC recognized that machine learning could analyze hundreds of displacement drivers to forecast forced movement 1–3 years ahead — enabling anticipatory action, strategic resource allocation, and evidence-based advocacy.
Beginning in 2020 in collaboration with IBM and with funding from the Danish Ministry of Foreign Affairs, DRC developed a suite of four predictive models: Foresight (national-level, 1–3 years), AHEAD (sub-district, 3 months), SODRD (drought-related pastoralist displacement), and SPIN (pastoralist risk corridors in the Sahel).
Foresight: National-Level Displacement Forecasting (1–3 Years)
Foresight is a machine learning model that predicts forced displacement (IDPs, refugees, and asylum seekers) at the national level 1–3 years into the future. It currently covers 27 countries representing over 90% of global displacement.
violence, governance, economy, environment, socio-demographics
annual forecasts accuracy
including Syria, Afghanistan, CAR, Somalia
Data sources include the World Bank, UN agencies (UNHCR, IOM, OCHA), NGOs, and academic institutions. Foresight outperforms standard humanitarian planning figures in 12 out of 18 countries where comparisons are possible. It is used for DRC's annual strategic planning, HNO/HRP processes, and scenario analysis (e.g., Taliban takeover in Afghanistan, election violence in CAR).
📌 Key finding from 2026 Global Displacement Forecast:
Three structural drivers converge — ongoing conflict/institutional fragility, climate stress interacting with fragility, and constrained financing environments — pointing to continued displacement growth, concentrated in fragile and climate-exposed contexts.
AHEAD: Operational Early Warning at Sub-District Level
AHEAD (Anticipatory Humanitarian Action for Displacement) was developed to provide more operational, granular forecasts. It predicts displacement at the sub-district level three months into the future, enabling field staff to prepare for specific locations and timeframes.
Currently active in the Liptako-Gourma region (Burkina Faso, Mali, Niger), South Sudan, and Somalia. The model uses open-source data on conflict, health, environment, food insecurity, IDP numbers, and income. A significant portion of the initial AHEAD model was built by DRC's West Africa regional office to ensure local contextualization.
"The Foresight model is valuable for long-term planning, but we needed operational tools that provide precise details on when and where displacement will actually occur — helping staff on the ground prepare." — DRC Global Advisory
Climate-Induced Displacement: SODRD & SPIN Models
SODRD (Slow-Onset Drought-Related Displacement) is applied in Somalia and the Somali region of Ethiopia. It analyses interdependencies between rainfall, livestock, land structure, population movements, and socioeconomic parameters, simulating potential displacement among pastoralists based on seasonal weather forecasts. The model has been used for proactive response planning in Somalia.
SPIN was developed when replicating SODRD in the Sahel proved too challenging due to data gaps. SPIN predicts risk levels in Sahelian pastoralist corridors based on historical security incidents, leveraging a regional pastoralist network's early warning system. It predicts future alerts and maps safe corridors for pastoralists to mitigate climate risks.
Somalia & Ethiopian Somali region — drought-related pastoralist displacement, seasonal weather forecasts.
Sahel countries — risk levels in pastoralist corridors using historical security incidents and alert messages.
Results: Remarkable Accuracy & Real-World Impact
All four models were rigorously tested against historical data and real-time scenarios. The models have shown "remarkable accuracy" based on comparison between historical forecasts and actual displacement. The 2026 Global Displacement Forecast report highlights that the current global asylum architecture rests on increasingly fragile assumptions, and DRC's models provide crucial foresight for policymakers, donors, and humanitarian actors.
Operationalizing AI: Dashboards, Training & In-House Management
Primary users are DRC program staff and humanitarian actors involved in displacement response. They access user-friendly dashboards displaying displacement numbers and trends. For non-data-scientist staff, DRC provides snapshots, reports, and regular training sessions. The models are managed in-house by DRC data scientists who handle modeling, updates, and information extraction.
⭐ Scenario-based forecasting: To address the "black box" problem, DRC developed functionality that allows users to tweak underlying parameters and generate their own forecasts, building trust and understanding of model assumptions.
Implications for Monitoring & Evaluation (M&E) & Anticipatory Action
From reactive to predictive M&E
Traditional M&E measures outcomes after displacement occurs. Foresight enables prospective evaluation — comparing predicted vs. actual displacement to calibrate models and inform anticipatory action plans before crises escalate.
Evidence-based resource allocation
Donors and humanitarian coordinators can use 1–3 year forecasts to pre-position funding, shift resources to high-risk areas, and advocate for preventive action — moving beyond the annual appeal cycle.
Tracking structural drivers over time
By monitoring 120+ indicators, M&E systems can identify which drivers (governance, climate, violence) contribute most to displacement in specific contexts, informing program design and policy advocacy.
Key Learnings & Future Roadmap
📌 Critical learning: Structured data collection from the start
DRC initially aimed to leverage internal data but found much of it lacked historical depth and systematic collection. Alexander Kjaerum, Global Advisory at DRC, reflected: "If we had the opportunity to start over, we would prioritize a more structured approach to data collection from the beginning."
Future plans include: expanding AHEAD to 12+ more countries via a three-year grant; integrating social media data (X, Facebook) and remote sensing; enhancing scenario-based forecasting; developing a specialized model for predicting surges in displacement (as current models struggle with unprecedented events); and expanding the data science team. Long-term sustainability relies on partnerships with IOM, tech companies, and donors like Danish MFA, SIDA, and ECHO.
Prioritize structured, historically deep data collection from the outset.
Build interpretability features (scenario-based tweaks) to overcome "black box" mistrust.
Localize model development (West Africa regional office built AHEAD) for contextual accuracy.
Frequently Asked Questions
What is the Foresight model and how accurate is it?
Foresight is a machine learning model that predicts forced displacement (IDPs, refugees, asylum seekers) at the national level 1–3 years ahead using 120+ indicators. More than half of its annual forecasts are within 10% of actual displacement, and it outperforms standard humanitarian planning figures in 12 out of 18 countries.
How does AHEAD differ from Foresight?
Foresight provides national-level forecasts 1–3 years ahead for strategic planning. AHEAD operates at the sub-district level with a 3-month horizon, designed for operational early warning and field-level preparedness. AHEAD is active in Liptako-Gourma (Burkina Faso, Mali, Niger), South Sudan, and Somalia.
What data sources does DRC use for its models?
Open-source data from World Bank, UN agencies (UNHCR, IOM, OCHA), NGOs, academic institutions, and regional networks (e.g., pastoralist early warning systems). Indicators cover violence, governance, economy, environment, health, food security, and socio-demographics.
How can other organizations access or learn from DRC's models?
DRC publishes annual Global Displacement Forecast reports and a Foresight technical note on their website. They actively partner with IOM, tech companies, and donors. Contact Alexander Kjaerum (alexander.kjaerum@drc.ngo) for collaboration inquiries. The models are managed in-house but DRC shares methodologies and findings openly.
A Paradigm Shift: Anticipatory Humanitarian Action Powered by AI
The Danish Refugee Council's suite of predictive models — Foresight, AHEAD, SODRD, and SPIN — demonstrates that machine learning can fundamentally transform how the humanitarian sector anticipates and responds to forced displacement. By shifting from reactive crisis response to evidence-based foresight, DRC enables donors, policymakers, and field staff to allocate resources proactively, design anticipatory action protocols, and advocate for structural prevention.
For Monitoring & Evaluation professionals, these models offer a new frontier: comparing forecasted vs. actual displacement as a core performance metric, tracking the predictive power of structural drivers, and embedding early warning into program cycles. As DRC expands to 12+ more countries and integrates social media and remote sensing data, the potential for AI-driven humanitarian foresight will only grow.
"The real attraction is the ability to forecast displacement anywhere on earth, enabling targeted interventions and better resource allocation in the ongoing efforts to protect vulnerable populations." — DRC Global Advisory Team
