UNICEF
Mapping Children Displaced in a Changing Climate
Vision: Unlock new insights on climate and weather-related displacement of children to inform UNICEF’s global advocacy work to prepare for and minimize the risks of child displacement
Project Overview
Data Challenge
Improve disaster risk model resolution, Combining data layers
Data needs
Filling data gaps, Machine Learning, Population estimates
Data practices
Predictive modeling, vulnerability mapping, forecasting
target users
Governments, Policy Makers, UN agencies, Civil Society
tech team
UNICEF Project TEAM
A UNICEF statistician and demographer led the project and partnership with IDMC, and contracted a geographer with a geospatial data education background to complete the analysis.
hypothesis
New data can be used to target response efforts and resources to the most vulnerable children – to protect their futures from the impacts of displacement as the climate continues to change.
“We had thought about doing something to better understand this issue for quite a long time! Early estimates of child displacements were based on an assumption that internal displacement was somewhat random, and national level age structures were used to form our estimates. We needed better resolution. We had aspirations to accomplish this, but lacked the resources to work on the issue. Climate change had been a priority from the UNICEF executive director since the start of their tenure, and that helped elevate the work to a priority.”
Introduction
Census data in the global south can be unreliable, leaving governments and other international development organizations in the dark about where individuals live and how many people there are in households or communities. This data is especially challenging to collect in informal settlements that grow on the outskirts of cities. Delivering services to an unknown quantity becomes a guessing game. Talented mathematicians and geospatial data scientists have developed methods to address this challenge, employing advanced statistical techniques coupled with high quality satellite data that helps locate and quantify building footprints. These methods can locate buildings and predict how many people live in them over a massive land area, even entire countries. Knowing the answers to ‘who’, ‘how many’ and ‘where’ are crucial for local and national planning and service delivery efforts, but they are not sufficient to understand the risks those people face from natural disasters such as wildfires, floods, heat waves, and drought. Layering geographic risk estimates based on historical data of the built and natural environment with high confidence population estimates would provide the basis for a framework to forecast where children may be the most vulnerable to disasters. This would allow governments to better target their limited resources and prepare to respond with crucial aid where and when it is acutely needed.
READ More:
HOW IT Started:
This project started as discussions between UNICEF and the Internal Displacement Monitoring Centre (IDMC) as an attempt to better understand the effects of climate change on children. The two groups have been collaborating since 2018, comparing notes and methods when IDMC’s displacement estimates are released on an annual basis.
IDMC’s 2021 report was their first global report with a special focus on children. This work illuminated the lack of data on what happens to children in the immediate aftermath of disasters and an incomplete understanding of how many of them are on the move.
“We had thought about doing something to better understand this issue for quite a long time! Early estimates of child displacements were based on an assumption that internal displacement was somewhat random, and national level age structures were used to form our estimates. We needed better resolution. We had aspirations to accomplish this, but lacked the resources to work on the issue. Climate change had been a priority from the UNICEF executive director since the start of their tenure, and that helped elevate the work to a priority.”
UNICEF’s Policy and Program groups helped stir interest in the project. UNICEF and IDMC worked together on the proposal to the PJMF’s Data Practice Accelerator to secure the resources, and contracted a GIS consultant with the time and expertise to carry out the analysis.
Approach
1. Study IDMC displacement data
“There were some revisions necessary, and each data entry was located in a specific area. For older entries this was not as complete and we helped to identify and fill in the gaps. The initial work served to help IDMC to improve their previous entries and helped support future work. The dataset is in much better shape now.”
IDMC’s Global Internal Displacement Database is composed of almost 39,000 data points on displacement related to weather-related events. In addition to data wrangling and a revision of the oldest facts’ locations, filters needed to be applied to retrieve only the best and latest estimate for each disaster. In total, almost 135 million internal displacements triggered by weather-related disasters were recorded between 2016 and 2021. In absolute numbers, most of them occurred in Asia – with China, the Philippines and India topping the list. When viewing displacement figures relative to the countries’ population, small island states—such as the Dominican Republic or Pacific islands (including Vanuatu and Fiji) – were the most affected by internal displacements.
2. Map the different age structures for each administrative level
“The idea was to find the smallest admin level for the percentage of children (under 18). Pointing the displacement data on the map, we calculated the inference of this. Previous models used children estimates at national level. In France there are few young people in rural areas, but cities are more dynamic, and these trends can differ across the globe.”
The original dataset was imported to GIS software, mapping all displacement events recorded by IDMC across the globe. The project team ran different studies such as visual and statistical hotspot analyses to highlight areas with high levels of displacement where efforts should be targeted to prevent and respond to displacement.
3. Iterate and Revise
An iterative process of revision to see if results were in line with what we had imagined. For some displacement events, they already had data. “We tried to think through the limitations of our methodology. Sometimes there was evidence of men staying in a place to protect the house – we had to take these into consideration”.
Data cleaning steps were interspersed along the way!
4. Import the data into the Cloudera Platform
Importing the data into the Cloudera Platform to support faster analysis and developing deeper insights through machine learning. This online solution enables replication of the IDMC data model, automatization of the analysis and will improve the efficiency to support new learning.
5. Deploy Disaster displacement risk model
IDMC’s disaster displacement risk model launched in in 2017, based on the UN Office for Disaster Risk Reduction (UNISDR) model that analyzes the risk of economic losses due to disasters.
“The risk model uses information on hazards (e.g. cyclones), exposure (people and buildings), and vulnerability (fragility of buildings) to estimate displacement risks in the future. The model takes into account the risk of medium to large-scale events, that is disasters which occur more frequently but lead to relatively few displacements, as well as “once in a lifetime” disasters (or even once in a century or once in one thousand years), which can lead to large- scale displacements.”
Using this information, they can estimate how many people could be displaced on average in every given year in the future, the average annual displacement (AAD).
“The AAD should not be understood as the number of displacements which can be expected to occur every year. It is really an average of displacements as a consequence of all kinds of events which could occur over a long timeframe. A single large disaster event – such as a once in a thousand-year flood – will result in far more displacements than the AAD. By expanding the time horizon, the AAD metric can become more concrete and tangible: It can tell us how many displacements we could expect in the next 10, 20, or even 50 years.”
The risk model uses long term climatological and other environmental data to identify areas at risk of hazards, and data on physical vulnerability (such as location and quality of buildings) to estimate the number of houses destroyed. This, taken together with the average household size, makes it possible to estimate the number of displacements. But both the climatological and environmental data, and the information on location and quality of buildings refer to current and historic situations: They do not take into account the influence of climate change on the frequency and severity of future hazard events. Nor do they account for demographic changes (such as the size or age structure of populations,
“Crucially, we connected the gridded population with worldpop data and that produced the age structure from different grids, allowing us to calculate the number of children displaced.”
6. Verify and complement the analysis
Verify and complement the analysis by looking deeper into the impacts of child displacement at the local level. Given that the project will generate a global model at a small-scale, this was crucial. To achieve this, UNICEF presented its findings to internal stakeholders, country offices and external partners working on climate mobility.
NOTE: Model Caveats
“An important difference to the analysis of historic displacements is that the hazards considered in the risk model are narrower – a consequence of how the model is computing the risks. For example, while the historic displacement data usually do not differentiate between coastal flooding and riverine flooding (and include other forms of flooding such as flash floods), the displacement risk model considers only riverine floods. As a consequence, it is not possible to directly compare the displacement numbers between the two types of analyses.“
Another difference is that the risk model does not include pre-emptive evacuations. This is because the model estimates future displacements based on the extent of damage and destruction that hazards of different intensities are likely to cause. Pre-emptive evacuations can – depending on hazards and the country’s level of preparedness – contribute a large proportion of the recorded displacements (e.g., wildfires in the United States, cyclones in the Philippines).
Most of the evacuees may be able to return to their houses after the hazard subsides and their displacement may last for only a short while, while others may find their homes, schools, health facilities and other infrastructure destroyed and become trapped in protracted displacement. Since the risk model does not account for pre-emptive evacuations the resulting numbers are a significant underestimate of the actual figure of children likely to be displaced in the future.
Similar to the analysis of the historic displacement, the model developed for this study allows us to identify areas that carry a high risk of being affected by specific hazards in the future and having a large child population, thus allowing to identify expected hotspots of child displacements in the future. Identifying the geographic areas of potential future disasters and the scale of the expected child population affected can help countries to prepare for disaster-related displacement and mitigate the risks and impact on children and their communities.
A caveat needs to be mentioned here as this analysis does not take into account the area’s vulnerability nor the potential mitigation measures in place to prevent the hazard to affect the population, such as houses resistant to disasters. In the following, we give examples of country-level future risk analysis: Riverine floods in Nigeria, storm surge in the Philippines, and cyclonic winds in Bangladesh.
Data Fluency
“In the process of working with the IDMC model we had to learn about the different concepts and what they express in order to effectively translate them for a lay audience in an advocacy report to bring attention to the cause. There is a fine balance between making statements that are correct while not getting lost in the conditions and caveats of the models.”
UNICEF had to work closely with IDMC to confirm language and they co-edited many passages of the final report.
“There are so many assumptions built into the risk models. Very rare events can lead to traumatic displacements. People have difficulty thinking about probability. Once-in-a-hundred-year events can happen more often. A famous example from Hurricane Katrina. An event of that scale happened 3 years in a row. It is difficult to explain that big events can happen every year but with low probability.”
“It was great to apply this knowledge! It felt like the perfect time to refine my skills and learn more programming languages that were not in my curriculum. Discussions with the McGovern team were great to speed up the work. The software has thousands of add-ons, and is a never ending learning process! I now feel much more advanced than before”
Lessons Learned
Cleaning large data sets requires manual effort and patience.
At the beginning of the project the consultant spent a lot of time cleaning the data, and during a mapping step, realized that data from a different country was ending up in Denmark.
“When we realized that a lot of administrative data was not labeled with numbers, we had to manually resolve it. Another time, we noticed that some Middle Eastern countries were mislabeled. The McGovern data team helped us with a method to mash up different names with alphanumeric characters that were not exactly the same, and that helped immensely with the global estimates.”
There will be challenges working with data that you cannot foresee
You can have a perfect plan for what to do with your data and then all of sudden you realize that something in a map is not where it’s supposed to be. For quality control, sometimes you have to go back through and check all other data points you’ve been working on.
Displacement due to slow onset disasters like droughts are likely radically underreported.
“Displacement linked to slow-onset disasters is difficult to measure as it is often driven by an interplay of aspirations, socio-economic and climate-related factors. And in countries affected by fragility and conflict, and poor data collection and monitoring capacities, it can be even more difficult to measure.”
Displacement data is not systematically available.
“Displacement data is most readily available for large events and in countries with a regional or national disaster agency, support from international agencies and a strong media presence. Further, there is significantly more information available on Internally Displaced People (IDPs) who take refuge at official shelters compared with those living in host communities or cities. As the vast majority of IDPs move to cities or urban areas, available figures are likely to be substantial underestimates.”
There are challenges combining demographic characteristics with displacement data.
“Experience at the country level shows that the proportion of children among the internally displaced population is often higher than the proportion among the overall national population. This means it is likely that using national age distribution data to estimate the number of internally displaced children is a significant underestimate, overlooking thousands, or even hundreds of thousands of children in some countries.”
impact metrics
UNICEF is publishing this data in a global report which highlights the policy implications, and advocacy for building resilience at schools and other infrastructure that children depend on. Its ultimate aim is to push countries to take the responsibility to reduce their emissions.
“Almost 135 million internal displacements linked to 7,382 weather-related displacement events were recorded by IDMC between 2016 and 2021. Based on the analysis, there were 43 million displacements of children globally over the last six years – of which 22.4 million were of boys and 20.8 million of girls. This is the equivalent to approximately 20,000 child displacements per day.”
project
SustainabilitY
Future Direction:
The team is thinking about updating estimates with new data from IDMC that were originally released.
UNICEF is internally socializing the findings from the report, and is using it as a centerpiece to prepare key messages at the upcoming climate events before the COP 28.
The publishing of a global report is not meant as the end point for the work. One project idea the team is contemplating is to create a regionally-focused product building on our global report in South Asia. There is a potential to do a reanalysis for 9 countries accounting for special hazards in the region.
In addition, IDMC is developing a new risk model. Their current model is based on past climatological and methodological data looking back on historical data, but it does not take into account future climate change scenarios. The new model will include forecasts based on various global temperatures and small metals concentrations. UNICEF plans to bring together the age-disaggregated geolocated population data to further refine different child displacement scenarios.
“The numbers we generate are a means to grapple with the scale of the problem, and human interest stories put faces to the problem.”