To improve transparency, the Centre has created a catalogue of predictive models with information on ‘who is doing what, where and when’. We hope that this effort will make it easier for partners to get a quick overview of the models that are available and their current state of development. If your organization would like to be listed in the catalogue, please fill in this form.

Model name Partners involved * Topic * Geographical scope * Description Update frequency Link to model code (if available) Main contacts Additional documentation
Project Jetson UNHCR Innovation Service, UN Global Pulse, Essex University Big Data & Technology Centre, Omdena Foundation Affected people Somalia, Ethiopia Jetson is experimenting with two methods: gravity model and multivariate TSA, machine learning-based (open source, Python and R applications). The target variable is forced displacement, and the nexus between displacement, climate/weather anomalies and changes and violent conflict. The model predicts at least one month of forced displacement movement (population flow, influx-only). This is the arrivals of IDPs and Refugees in the different regions of Somalia (admin level 1) and the southern border region of Ethiopia (Dollo Ado). Every 3 months since 2017
Unaccompanied and separated children projection model UNHCR, University of East Anglia Affected people South Sudan, Democratic Republic of the Congo, Somalia Arrivals of unaccompanied and separated children in refugee settings. The outcome is a prediction of the number of arrivals of Unaccompanied and Separated Children (UASC), usually by month but sometimes by week. Under development - models created for several countries and updated each year.
The Joint Analysis of Disaster Exposure (JADE) OCHA (Asia-Pacific), WFP (Asia-Pacific), Pacific Disaster Centre Affected people 20 countries supported in the Asia Pacific region Estimates the number of people likely to be in need after a sudden onset natural hazard. The JADE provides the number of people living in affected areas, the number of people living in the worst affected areas, and the number of people living in the worst affected area who were already considered "vulnerable" before the crisis. There are a number other values and figures provided. The spatial resolution is 1 x 1 km. The initiative was launched in 2018. The actual analysis is launched within 24 hours of a sudden onset emergency. It is not expected to be updated during the course of an emergency.
Flee Brunel University London (contributions from HiDALGO consortium and VECMA consortium. Also received guidance and input from IOM and UNHCR) Affected people The results cover case studies of three African countries at the moment but can be generalized to more contexts as per the authors. Flee aims to establish a tool that can (a) forecast the destinations of forcibly displaced persons, (b) estimate the effects on migration of major policy decisions such as border closures and camp relocations, (c) provide an estimation of the number of displaced persons in areas where the data collection is incomplete. The first major version of Flee was established in 2017. It is updated continuously and is funded until May 2023. An updated and more realistic rule set for agent movement is expected in Q1 or Q2 2020.
INFORM Risk Index Joint Research Centre of European Commission with contributions from multiple partners Coordination and context Global coverage (subnational in some countries) Risk for humanitarian crisis and disasters is evaluated along three dimensions coping capacity, vulnerability and hazard. The model has global coverage, and the output is national scale. It was developed in the 2014, and it is updated twice a year
Foresight Project Danish Refugee Council and IBM Refugees and persons of concern Afghanistan, Myanmar The project develops a forecasting model for total forced displacement from a country 1-3 years ahead. Total forced displacement is a combination of refugees and asylum seekers outside a given country and the internally displaced in the same country. The project was updated in 2019-2020 and it will be continuously updated with more countries.
The Inundation Model - Google flood forecasting model Google AI for Social Good Affected areas Pilot in India (Bihar) with extended coverage in in 2019 in the Indian subcontinent. Simulation based real time flood forecasting model. Accurate predictions of severity and locations of flooding based on AI and physics simulation. Integrated with local partners to put an active warning system in place. Project started mid 2018 and latest 2019 iteration available online.
African Risk Capacity African Union, Africa Risk Capacity Affected people 33 African Union member states Large umbrella effort to have better responses towards extreme weather events and natural disasters. The frameworks uses predictive analytics and innovative financing (insurance to redistribute risk and risk pooling) to respond. Framework covers multiple disasters types: droughts, floods, diseases, and cyclones. The pure modelling efforts are ARV (Africa riskview). ARV model updated every year (sometimes for every harvest season) Proprietary software package - access can be requested. Proprietary software. Model details available here
Famine Early Warning Systems Network (FEWSNET) USAID (also NASA, NOAA, USDA, USGS) Food security 28 countries with a focus on West, Southern and East Africa Early warning and analysis of food insecurity - livelihood based model. Efforts are made to understand impacts on livelihoods, markets/sources, nutrition. A key component of monitoring is satellite based remote sensing. Mixed methods analysis (quantitative and qualitative analysis) enables FEWSNET to build reliable scenarios and Integrated Food Classification (IPC). Updates may be issued every month. Consistent new datasets for all monitored countries are released every ~6 months (every harvest season)
Artemis Model World Bank Food security Pilot in 5 countries to be expanded to 21 additional countries The model combines qualitative, remote sensing, and data insights from multiple sources to generate a subnational signal of food insecurity at least 6 months to 1 year in advance. No planned updates of the model after publication in scientific journal. Scientific paper to be published in 2020
Hunger Map LIVE WFP and Alibaba Food security Global coverage (90 countries) Advanced statistical/ML system with a focus on large scale nowcasting. The hunger map is a platform that uses streams of public data from multiple sources to create a complex dynamic subnational portrait of world hunger. In places where the data is missing ML algorithms are used to make extrapolations. Daily
Crisis Computing/ Artificial Intelligence for Digital Response Qatar Computing Research Institute Coordination and context Global coverage Natural Language processing model which focuses on extracting disaster and emergency related information from social media messages. The objective of the model is to gain actionable information from troves of social media to inform better humanitarian response In real time Part of code used in the crisis computing project is available online at the official github
Asset Impact Monitoring System (AIMS) WFP Geography and infrastructure Pilots in Niger, Aghanistan, Sudan, South Sudan, Tajikistan The objective of this system is to monitor the rehabilitation of degraded landscapes using satellite imagery and remote sensing. The AIMS model mainly monitors the degraded land which is rehabilitated with with the Food Assistance for Assets (FFA) program. This approach gives evidence based quantifiable impacts of policy and landscape changes. AIMS Model in development/pilot phase. Last public update was given in Oct 2018 for the 5 pilot countries.
510 Typhoon impact model Netherlands Red Cross - 510 Coordination and context The Philippines Predictive model to identify high priority areas in the wake of a natural disaster. The model uses a large numbers of factors to assess this including but not limited to socio-economic factors, housing types, impact data and administrative boundaries. One fully validated model for Philippines available. Pilots in development to test model scale ability in Mozambique.
Mongolia Dzud model National Agency for Meteorology and Environmental Monitoring (NAMEM) Coordination and context Mongolia Each year, National Agency for Meteorology and Environmental Monitoring (NAMEM) publishes Dzud risk map which uses summer condition, pasture carrying capacity, livestock number, anomalous precipitation and temperature, snow depth, biomass, drought index, temperature forecast etc. to predict which regions may experience severe winter condition. Yearly
REACH flood susceptibility model REACH Affected areas Yemen A model is build using high resolution data sets to account for topographical features, normalized vegetation index (NDVI), rainfall, drainage, land cover and soil type. These factors are combined in a weighted linear fashion to compute an area's flood susceptibility (not risk). Static model updated last in June 2019
Global Disaster Displacement Risk Model IDMC, UNDRR, ETH Zurich Affected people Global Coverage The model combines probabilistic analysis of events recorded in national loss databases (retrospective risk) with probabilistic risk estimates for low-frequency, high-impact events (prospective risk). Prospective disaster displacement risk is expressed as a function of hazard, exposure and vulnerability: Risk = Exposure x Hazard x Vulnerability The prospective risk profile is calculated using hazard scenarios provided by a wide range of partners. These events represent all of the possible disasters that could affect a country over different return periods. For each grid cell in the exposure map, which has a resolution of between 1km x 1km in coastal areas to 5km x 5km inland, the expected impact of a hazard is calculated using vulnerability curves to determine an expected level of damage to homes and other structures based on the hazard’s intensity. Whenever the simulated damage to a home is more than 55 per cent, it is considered uninhabitable and the household displaced. Model developed in 2018 using from the 2015 Global Assessment Report ( )
Pastoralist Livelihood and Displacement Simulator IDMC, Climate Interactive Affected people Kenya, Ethiopia, Somalia System dynamics (a modelling technique often used to analyse population dynamics and the behaviour of complex systems) is used to understand displacement dynamics of pastoralists in the Horn of Africa. The model formalises the causal relationships and drivers of pastoralist displacement, providing a decision support tool to explore the impact of specific policies and interventions. The model user interface is designed to answer a decision-maker’s “what if?” questions. Developed in 2015
Global Cholera Risk Model (GCRM) University of Florida and University of Maryland Affected people Global coverage The algorithm uses precipitation, temperature, and population data from different data sources to project the risk of a cholera outbreak 4 weeks later. The risk values represent the conditions under which Vibrio’s are more likely to grow. Cholera risk increases if a region experiences a two month above average air temperature, followed by one month of above average precipitation. These two conditions must be accompanied by damage to WASH infrastructure, further facilitating interaction of cholera bacteria with humans and thereby leading to an outbreak. Daily, weekly and monthly updates available
PRIO Conflict Prediction Peace Research Institute Oslo (PRIO) Conflict Global Modelling effort that predicts internal armed conflict in a country. The scope of the study ranges from 1970 to 2050. The model uses a wide range of predictors from conflict history, development indicators of the country, neighborhood stability. The model subsequently goes through extensive testing and validation exercises. Last update in Jan 2013
ViEWS: Violence Early-Warning System Upsalla University - Department of Peace and Conflict Research Conflict Africa An early warning system to track and predict three classes of political violence. The model can predict armed conflict involving states and rebel groups, armed conflict between non-state actors, and violence against civilians. More importantly the model can disaggregate results into three different levels of analysis. We can analyse violence at national level, subnational level and actor level. The model takes into account a large array of input variable including but not limited to history of conflict, political institute, economic development, temporal factors, demography, natural resources and proximity of conflict. Every month Main Website - Current forecasts - Peer reviewed publication -
Conflict Early Warning and Response Mechanism (CEWARN) Intergovernmental Authority on Development (IGAD) Conflict Migration Djibouti, Eritrea, Ethiopia, Kenya, Somalia, South Sudan, Sudan, Uganda Information of conflicts, migrants and natural disasters is shared between participating countries in a transparent manner. Thus information about potentially violent situations are jointly analysed to develop case scenarios and optimal responses. The analysis, data and response options are also communicated to all IGAD member states
CrisisWatch The International Crisis Group Conflict Global coverage Crisis watch is a "nowcasting" tool that is updated every month to get an overview of security and political developments across 80 countries in almost real time. The tool provides a bird eye view of ongoing conflict trends, escalations and developmental impact for all decision makers. 1 month Main Website -
The Early Warning Project Simon-Skjodt Center and Dartmouth College Conflict Global coverage The statistical model of this project computes risk of mass killing and genocide by studying previous episodes (1945- present) to identify predictors and risk factors. Various models are cross validated and the best performing ones are then used to make predictions for the next two years. The output of a model is an estimated risk (of an onset of a mass killing) and the country's vulnerability rank. New predictions made every 2 years Main Website -