7

Advancing an Earth Systems Science Approach to Modeling Climate Migration

The final workshop session built on the discussion of opportunities and challenges in advancing an Earth systems science approach to modeling climate migration. Speakers examined key gaps in data and modeling along with opportunities to overcome these gaps and advance the study of climate change and human migration.

HARMONIZING SPATIAL AND TEMPORAL DATA

Chris Funk (University of California, Santa Barbara) spoke about ways to harmonize spatial and temporal data and models to advance an Earth systems science approach to modeling climate migration. At the Climate Hazard Center, he has worked to co-develop datasets and strategies for monitoring extremes by collaborating with Earth scientists and data analysts from Africa, South Asia, and Central America to support the Famine Early Warning Systems Network (FEWS NET).¹ With funding from the U.S. Agency for International Development, FEWS NET monitors global food insecurity, which has increased from 35 million people who were extremely food insecure in 2015 to around 129 million in 2023.

Researchers at the Climate Hazard Center have developed strategies and datasets to support FEWS NET for about 25 years, and Funk said that much of this work is relevant to issues of migration. In particular, a staged early warning system framework for humanitarian crises related to drought illustrates how environmental factors can be approached at different levels

___________________

¹ See https://fews.net (accessed July 1, 2024).

of granularity and lead times to understand and anticipate pressures that people experience (Funk et al. 2023). Starting at a global scale and long lead times, climate forecasts are used to assess features such as El Niños and La Niñas. Then, at shorter lead times, interoperable weather forecasts are integrated to provide information at regional scales, high-resolution weather observations are used to track rainfall and temperature more locally, and finally, very high-resolution satellite imagery of vegetation can provide farm- and household-level information.

Funk pointed to several high-resolution datasets that are relevant for understanding climate impacts in the Global South. These include a dataset from thermal infrared geostationary satellite observations for 1983 to the present, which provides information about precipitation and temperature at local scales (Verdin et al. 2020). High-resolution heat projections are also available (Williams et al. 2024). In addition, the Climate Hazard Center’s researchers perform high-resolution climatology studies, which are important for understanding where people are exposed to extreme precipitation or temperatures, as well as climate model reanalyses and projections. While each of these data sources has limitations, Funk said that they can be blended to create high-quality products that are widely used. For FEWS NET, researchers take information from global El Niños and La Niñas down to local weather, extreme precipitation, extreme temperatures, and precipitation deficits and translate this information into assumptions about whether people in particular locations are going to have enough food. This is an example of effective harmonization where Earth scientists from multiple agencies work closely with food security analysts and dynamic modelers to use Earth science information to anticipate future challenges, according to Funk.

Funk described some of the challenges researchers have encountered in complex modeling to be able to consider the dynamics of climate change impacts and migration from an Earth systems framework. For example, there are a number of variables to consider when looking at the impacts of temperature-related shocks in a warming world. Increasing climate volatility, warmer temperatures that can increase saturation vapor pressure (the amount of water air can hold at 100 percent humidity), and rainfall deficits that when combined with warmer air can draw more water from the stomates of leaves, can then cause plants to experience stress and irreversible damage. Impact on vegetation may then lead to downstream consequences related to food insecurity. Although this has received attention in data-rich areas such as the western United States, Funk said that relatively little focus has been given to it in the Global South. This makes it difficult to translate air temperature anomalies that occur, for example, over southern Africa, into quantitative impacts on food production and income. Overall, Funk said, the early warning community is much better at tracking the impacts of precipitation deficits than temperature extremes.

Another gap includes how to integrate evaluation of the impact of humid heat shocks on labor and health into warning systems. Funk said that there are research studies linking extreme humid heat to health impacts, reductions in labor productivity, and increased energy costs associated with air conditioners. However, there is little capacity to integrate this information into early warning systems. Furthermore, his new experimental data on wet-bulb globe temperature monitoring data show extreme wet-bulb globe temperatures over much of the planet, but there is not currently a way to rapidly incorporate the impacts of extremely humid and hot conditions on food security and livelihoods, Funk noted. On the opposite end of the spectrum, researchers are also working to understand risks associated with dry heat, which can be measured with vapor pressure deficits. Funk said that projections indicate a large increase in areas with either extreme humid heat risks or high vapor pressure deficits across the Sahel in Africa, parts of eastern Africa and, surprisingly, even in places in India by 2050 (Williams et al. 2024). This is predicted to lead to adverse health impacts, rapid plant senescence, longer dry seasons, and decreased crop production.

Overall, Funk said that carefully stacking multiple sources of information can help to overcome data scarcity, pointing to FEWS NET as a tangible illustration of the benefits of effective cross-disciplinary, cross-scale harmonization. However, more work may be needed in monitoring, anticipating, understanding, and mitigating temperature-related shocks. During the discussion, Funk suggested that the datasets compiled by FEWS NET could be valuable for informing policy making in the context of migration projections, noting that these datasets could be particularly useful in identifying emerging hazards, analyzing trends, and anticipating future risks. He also provided more details on stacking approaches to address the problem of poor high-resolution data or a lack of longitudinal data, describing as an example researchers’ approach to assessing extreme wet-bulb globe temperatures in Guatemala, a country with significant variability at small scales. In this case, researchers combined historical station data with satellite observations and elevation data to create a high-quality climatology analysis and long-term average temperature for the region. Using satellite observations, climate model reanalyses, and temperature data, they then estimated anomalies around this average, providing a reasonable approximation of high-resolution variations despite not having as many observation stations as desired.

LEVERAGING LOCATION-BASED DATA

Ali Mostafavi (Texas A&M University) discussed how using location-based data can shed light on population movement and risk exposure (Hsu et al. 2024a). Location-based data are drawn from individual cell

phones, which record the user’s positions and movements as they move throughout the day (Fan et al. 2021). Over the past few years, this type of data has received a lot of attention from the scientific community because it can provide insights into behavior, especially during a crisis. During the COVID-19 pandemic, for example, this type of data was used to create empirical epidemiological models to evaluate viral spread and adherence to interventions such as shelter-at-home orders (Coleman et al. 2022; Fan et al. 2021, 2022). By revealing deviations and fluctuations in visitation and mobility patterns, location-based data can also be used to evaluate the impacts of climate-related hazards such as hurricanes and wildfires, in both the short term and the long term (Dargin et al. 2021).

For example, Mostafavi’s group used location-based data to examine evacuation patterns in the area of Florida affected by Hurricane Ian in 2022. Using anonymized location-based data aggregated at the block group level to ensure privacy, the researchers were able to track where and when people evacuated ahead of the storm. The study showed how the near real-time availability of this type of data can produce timely insights during a crisis, as well as insights into longer-term changes in population as people relocate, either temporarily or permanently, in response to damage to their homes or neighborhoods. These data can also be associated with sociodemographic characteristics of communities such as income and racial characteristics to assess disparities in the ability to evacuate or relocate (Esmalian et al. 2022; Li and Mostafavi 2022). Mostafavi noted that empirical observational data on human activities do not always align with assimilation and model-based patterns. For example, research has shown that actual patterns of evacuation behavior are not consistent with any of the agent-based models reported in the literature, suggesting that incorporating location-based data might be useful to help validate or improve other types of models.

Mostafavi also described how location-based data can be used to understand populations’ exposure to hazards between crises (Hsu et al. 2024b). For example, researchers used location-based data to evaluate where people go during their daily activities and how much of their time is spent in flood-prone areas (Lee et al. 2022). Defining a metric called life activity flood exposure, which is the proportion of minutes per week that a person spends in flood-prone areas, the researchers assessed flood exposure in different places across the United States (Rajput et al. 2024). They found that in San Francisco, for example, people spend only 6 percent of their time in flood-prone areas, while in Miami-Dade County, this proportion is 86 percent (Rajput et al. 2024). By accounting for where people spend their time and not just where they live, this approach provides more precise measurements of the degree to which populations are exposed to flood risks. The same approach can also be used to quantify exposure to other

environmental hazards, such as air pollution, urban heat (Huang et al. 2023), and toxic sites (Liu and Mostafavi 2023), Mostafavi said, as well as capture changes over time as climate hazards and migration patterns evolve.

Mostafavi noted several challenges with using location-based data for longitudinal studies, including privacy protections, data limitations, policy variations, and bias and representativeness. Because of the anonymization of user data and changing user identities, Mostafavi said, it is difficult to track relocation patterns beyond about 6 months. He noted that overcoming this challenge would require agreements with data providers to ensure continued access to conduct longitudinal studies necessary to capture population migration after climate-related events. Additionally, he suggested that establishing a global framework for collecting longitudinal data could enhance understanding of migration patterns and their interaction with climate hazards on a global scale. Finally, to address issues of bias and representativeness (e.g., the fact that rural areas are likely to be underrepresented compared with urban areas), Mostafavi suggested that generative adversarial network models could be leveraged to create synthetic data and improve the utility of location-based data for analysis. “I think we have an opportunity for better, more near-time observational evaluation of human interplay with climate hazards and the impacts that climate change ha[s] on people,” said Mostafavi. “But we need to have discussions, a framework, and collaborations between researchers, public organizations, and also private entities that collect and provide these datasets so that we can have the opportunity to conduct longitudinal studies in a privacy-protected and, of course, ethical manner.”

During the discussion, Mostafavi clarified that the researchers compared location-based evacuation data with agent-based models (ABMs) in the context of evacuation but not migration. The results showed that existing behavior-based ABMs for evacuation did not accurately capture actual evacuation patterns; in light of this, Mostafavi suggested that combining location-based data with survey-based methods could improve ABMs, noting that each event is unique and requires precise measurements of behavior. He highlighted the importance of learning from errors in ABMs to better understand their limitations and improve future iterations.

DATA SOURCES FOR MIGRATION

Deborah Balk (City University of New York) discussed key data sources available for studying migration in relation to climate-related drivers, along with some considerations for using them effectively. She noted that although there are many uncertainties about the future, there are a few things that are known. One is that a greater share of the world’s population will live in urban areas, and that this urbanization trend will be most acute in

Asia, Africa, and Latin America. In addition to increasing urbanization, she said that it is clear that in the future a greater percentage of the population will be older and that the world will be hotter, have more frequent and more intense storms and flooding, see more drought-prone areas, and experience higher sea levels.

The primary sources of demographic data available to understand current trends and project future ones include censuses, surveys, vital registration systems, tax records, basic demography (e.g., age, fertility, marriage, mortality) and health surveys. Though rich in many ways, there are limitations when used with climate and disaster data. Balk said that, in general, the spatial resolution of demographic data is somewhat coarse and often regional, which can pose problems for integrating demographic data with climate data. Differences in data formatting is another limitation in data integration. Finally, Balk stated that while datasets are often designed to be nationally or subnationally representative, they are usually not representative of environmentally delineated subgroupings that are most relevant to climate hazards, such as flood zones.

Climate data from satellites, Earth observations, and weather stations, also limit how they can be combined with other sources of information. She added that the format of satellite-derived information may be unfamiliar to social scientists working with demographic data. The spatial and temporal resolution of climate models varies from fine to coarse, resulting in a wide range of resolution.

As other speakers noted, Balk stated that internal migration within countries is generally more common than international migration and that more data are needed for internal measures. She suggested that improving assessment of internal migration may improve understanding of urbanization, and where people are moving to and from, particularly if this migration increases in response to climate-related stressors. Currently, there are relatively few studies comparing the prevalence of internal migration or urban migration across a large number of countries, making it difficult to know how countries compare in terms of internal migration. She noted that the IPUMS² data collection at the University of Minnesota allows for the harmonization of census and survey data but eliminates information that is not uniformly available, including origin destination information or reasons for migration. There are also varied sources of data for migration between and within countries, with different survey questions asked across data sources and over time.

Balk compared two approaches to obtaining migration data: censuses and demographic and health Surveys (Acosta et al. 2020). Censuses are fairly reliable for population projections because they fully enumerate the

___________________

² See https://ipums.org (accessed July 1, 2024).

population, occur at fixed intervals, and capture longer periods of data. Demographic and Health Surveys capture population dynamics at a more granular level because they record local moves, can provide insights into flows between rural and urban areas, and are available for African countries, where historical censuses are largely absent.

Obtaining a global picture of subnational urban in-migration involves patching different types of data together. When doing so, Balk said it is important to be aware of differences between data sources in terms of what types of moves count as migration, potential sources of selection bias, the role of nonresponders, and differences in the bounding unit size. To improve migration statistics within national borders, Balk suggested focusing on migration between cities and towns, citing an example from Mexico showing that moves between major administrative areas miss a lot of moves within and to urban areas. This is important in the context of climate-related migration because evidence suggests that people tend to stay close, often within the same labor market, when relocating in response to disasters, Balk said, adding that such patterns can amplify inequities and have implications for who is in the path of the next possible storm.

Balk also highlighted emerging sources of data from cell phones, social media, and administrative data, such as Internal Revenue Service data relevant for moves within the United States, and vital registration data. For all of these sources, she said that it is important to think about who these data represent—only those with cell phones, only those who use Facebook, or only tax filers, for example. She suggested that combining such data sources in before/after studies or with other anchor data could enhance their utility. She added that partnerships are needed to aid in understanding these data and also stressed the importance of ensuring privacy protections.

Closing, Balk emphasized that measuring migration is critical but often more complex than measuring other demographic features. Even apart from climate change, future demographic shifts will be oriented around migration, so countries have an incentive to think hard about strategies to improve these measurements in their national statistical infrastructures.

DISCUSSION ON ADVANCING AN EARTH SYSTEMS APPROACH

During the panel discussion, speakers further examined challenges and opportunities related to data integration. They noted that data integration is important to advancing an Earth systems approach for understanding climate-related migration. Speaking about opportunities to bridge the gap between global, high-resolution climate data and location-based cell phone data, Funk noted that there are studies underway using cell phone data from Somalia and Zambia to examine where people are moving, which can be useful for food security monitoring. He added that Climate Hazards

group Infrared Precipitation with Stations (CHIRPS) data have also been bundled with Demographic and Health Survey data to assess health impacts and suggested that there are many other opportunities for collaboration going forward.

Mostafavi noted that studies combining different datasets are bounded by the common denominator in terms of resolution. For example, if climate hazards or climate projections are at a certain scale, then movement data are typically aggregated at that same scale. This limits the ability to examine variations in climate hazards or compare different levels of exposure among populations. Another limitation is the lack of demographic information on individuals. Although this protection is necessary for individual privacy, it makes it difficult to directly analyze demographic characteristics, and researchers must rely on block group data as a proxy, which may not be as precise. He noted that having demographic information associated with the data would improve the resolution of findings.

Mostafavi also responded to a question about whether new sources of data could be used to penalize communities that fail to respond to government announcements regarding hazards, leading to increased insurance premiums or other penalties. He noted that this is a complex issue and there is a need to consider both risks and opportunities with the increasing availability and volume of different types of data. In addition to understanding potential risk, he said that such data could also be used to inform evacuation planning for future events.

Responding to a question about data sources relevant to understanding adaptation to specific hazards, Balk suggested that improving data collection methods to include both place of origin and destination could be useful. She noted that migration is just one possible response to hazards; other factors such as income or access to cooling resources may also play a role. She also emphasized the importance of studying individuals who do not migrate and suggested using longitudinal datasets to track their experiences over time, underscoring the need for systematic collection of data on adaptation policies and initiatives. Funk added that studying food insecurity can offer insights into understanding adaptation to hazards. Rather than directly modeling food insecurity, he said that researchers can focus on proximate drivers such as reductions in crop production and labor impacts. Similarly, when examining risks related to extreme heat exposure, it is important to investigate local labor opportunities, barriers, and income changes. To improve estimates of household adaptive capacity, he also noted that researchers are experimenting with using remote sensing to map poverty and adaptive capacity components at a high resolution. He suggested that advancements in all of these areas could inform future migration studies.