Are diseases acquired in your bed, at night? In your ZIP code or neighborhood? At the coordinates of your house?
Much of spatial epidemiology begins with some version of that assumption. We map people to their homes, aggregate cases by neighborhood or administrative area, or attach environmental exposures to residential locations. Often this is entirely reasonable. But it quietly makes a strong assumption: that where someone lives is a useful proxy for where the processes affecting their health actually occur.
For infectious diseases, that assumption can be especially problematic. People leave home. They work, farm, attend school, visit markets and places of worship, cross borders, sleep in other locations, and move through environments where exposure may be very different from that around their residence. Mosquitoes, pathogens, pollution, and other hazards do not restrict themselves to the places where our datasets happen to locate people.
This repository is about that mismatch: the difference between where we locate people in our data and the places they actually occupy.
I began thinking about activity spaces when I was in graduate school learning GIS and spatial analytic approaches, and how to think spatially under the mentorship of Stephen Matthews. In particular, his work on spatial polygamy shaped how I conceptualize people's relationship to place:
Matthews, S.A. (2011). Spatial Polygamy and the Heterogeneity of Place: Studying People and Place via Egocentric Methods. In: Burton, L., Matthews, S., Leung, M., Kemp, S., Takeuchi, D. (eds) Communities, Neighborhoods, and Health. Social Disparities in Health and Health Care, vol 1. Springer, New York, NY. https://doi.org/10.1007/978-1-4419-7482-2_3
Stephen’s main focus was on U.S. social science, where individuals in data are often linked to census tracts or blocks. In my own work, I sometimes have much more spatial detail — for example GPS coordinates for a participant’s house. In a census or survey, we might have individuals linked to houses, many houses in a community, several communities in a region, and so on. With such data, we can map disease cases to households and create incidence or prevalence maps.
Inherent in such maps is the assumption that the mapped location is important for the process being studied — in infectious disease epidemiology, this often means transmission. For example, a map of incidence by household might reveal clustering patterns, suggesting local transmission at or near those homes. We don't always state this assumption, but otherwise why show cases per home?
However, few people spend all of their time at home. This is Stephen’s point: we are “married” to multiple places. We might spend significant time at school, work, or places of worship — all of which could be important for transmission — but these sites are almost never in our datasets or maps.
This line of thinking has inspired several branches of my research program:
In Southeast Asia, many farmers have farm huts near their fields, where they stay during peak agricultural labor periods (often easier than returning home daily). These periods often coincide with seasonal peaks in disease transmission — for example, malaria season overlaps with rice farming season.
Figure 1. Agricultural activities by season in malarious rural areas on the Thailand–Myanmar border. Malaria tends to peak in July–August each year, and occasionally there is a second peak in November.
One approach I’ve taken is to map farm huts in study villages (from our tMDA work), link them to their respective households, and look for spatial and temporal patterns in malaria infections that incorporate both home and farm hut locations.
- Parker, D.M., Landier, J., von Seidlein, L. et al. (2016). Limitations of malaria reactive case detection in an area of low and unstable transmission on the Myanmar–Thailand border. Malar J 15, 571. https://doi.org/10.1186/s12936-016-1631-9
Figure 2. Example of mapped farm huts linked to households for spatial epidemiological analysis.
Earth observation datasets (often rasters) are usually linked to individuals via their home location, sometimes using a buffer around the home to capture environmental conditions (here’s a tool to do this yourself). The buffer size is important — too small, and you miss relevant exposures; too large, and you dilute the signal. Movement ranges of residents should inform these choices.
- Rattanavong, S., Dubot-Pérès, A., Mayxay, M., Vongsouvath, M., Lee, S.J., et al. (2020). Spatial epidemiology of Japanese encephalitis virus and other infections of the central nervous system in Lao PDR (2003–2011): A retrospective analysis. PLOS Negl Trop Dis 14(5): e0008333. https://doi.org/10.1371/journal.pntd.0008333
Figure 3. Environmental indices for villages with study patient homes for the duration of the study period (January 2003 through August 2011) for all study patient villages, non-study patient villages in the study area, and for major diagnoses (LP = lumbar puncture; JEV = Japanese Encephalitis virus; Crypto = cryptococcal infection; ST = scrub typhus; MT = murine typhus; dengue = Dengue virus; lepto = Leptospira spp. infection). The buffer size used influences the summary measures of the environmental measure (A: normalized flooding index, NFI; B: normalized difference vegetation index, NDVI; C. enhanced vegetation index, EVI).
- Roberts, T., Parker, D.M., Bulterys, P.L., Rattanavong, S., Elliott, I., et al. (2021). A spatio-temporal analysis of scrub typhus and murine typhus in Laos: implications from changing landscapes and climate. PLOS Negl Trop Dis 15(8): e0009685. https://doi.org/10.1371/journal.pntd.0009685
Another approach is to measure actual human movement directly using GPS loggers in cohort studies. This is logistically complex but provides rich movement data.
Analysis for one such study — Human movement patterns of farmers and forest workers from the Thailand–Myanmar border — is documented in a repository built and maintained by my student and collaborator (S.T.T Tun): HumMovPatt.
- Tun, S.T.T., Min, M.C., Aguas, R. et al. (2023). Human movement patterns of farmers and forest workers from the Thailand–Myanmar border [version 2; peer review: 2 approved, 2 approved with reservations]. Wellcome Open Res 6:148. https://doi.org/10.12688/wellcomeopenres.16784.2
Figure 4. GPS tracks from 3 cohort study participants (indicated by different colors). Successive GPS logs are linked with lines to indicate relative movement.
GPS loggers are great for detailed studies, but they cover few people. To scale up, we’ve used mobile phone handover data, which can capture large portions of the population. With colleagues at Addis Ababa University and EThiotelecom, we analyzed when people move by time of day and compared this to the biting times of local mosquito vectors. We found that many people are moving during peak biting hours — meaning interventions like bednets, which only protect when you’re home and under them, can be “leaky.”
- Haileselassie, W., Getnet, A., Solomon, H. et al. (2022). Mobile phone handover data for measuring and analysing human population mobility in Western Ethiopia: implication for malaria disease epidemiology and elimination efforts. Malar J 21, 323. https://doi.org/10.1186/s12936-022-04337-w
Figure 5. Human mobility patterns in relation to mosquito biting times (from human landing catches) in Gambella Region, Ethiopia. Human mobility derived from mobile phone handover data, indicating plenty of movement during times when mosquito vectors are active.
More recently, we applied the activity-space concept to tuberculosis transmission in Botswana by combining geographic information on places people routinely occupied with whole-genome sequencing of Mycobacterium tuberculosis. Rather than representing individuals only by their residential locations, the analysis incorporated workplaces, schools, markets, places of worship, social venues, and other locations frequented during the potential infectious period.
The resulting geographic patterns differed from those obtained using residential locations alone. For several genomic outbreak groups, incorporating activity spaces helped identify localized areas that may have been important for transmission. This provides another example of why the locations recorded in conventional health datasets may not adequately represent the geography of infectious-disease exposure.
- Baker CR, Barilar I, de Araujo LS, Parker DM, et al. (2026). Using genomic epidemiology and geographic activity spaces to investigate tuberculosis outbreaks in Botswana. International Journal of Health Geographics 25:30.
This repository will soon-ish include:
- Code examples for incorporating multiple activity spaces into spatial epidemiology analyses
These repositories connect different parts of my spatial epidemiology research:
- research-atlas — Research atlas connecting projects across questions, places, and approaches.
- earth-observation-hub — How Earth Observation methods became central to my work, with curated papers and case studies.
- activity-spaces — Research on multi-place exposure (farm huts, GPS, mobile phone data) and its health relevance.
- METF-mapping — Mapping malaria post placement & community engagement.
- tMDA-program — Targeted mass drug administration trials & modeling.
- early-dx-tx — Early access to malaria diagnosis & treatment.
- tm-border-mch — Maternal and child health research on the Thailand–Myanmar border.
Unless otherwise noted, materials in this repository are licensed under the MIT License.