Our Address

7518 SOUTHPOINTE PL
Pensacola, United States,
Florida, 32514

Contact Information

A longitudinal ecological investigation tracking settlement patterns across coastal Kenya has established that integrating multiple satellite indicators can systematically identify peri-urban transitions at the local health system level, while exposing substantial differences in data stability. Reporting in the International Journal of Health Geographics, researchers evaluated ten Community Health Units within the Kaloleni Rabai Health and Demographic Surveillance System in Kilifi County from 2017 to 2024. The study examined whether publicly accessible Earth observation products and gridded demographic layers could characterize settlement transitions where administrative rural and urban labels fail to capture ongoing community growth. By harmonizing annual measurements of artificial radiance, physical building expansion, population distribution, and standardized urbanization degrees, the research team demonstrated that a multi-proxy model successfully isolates a persistent core of developing peri-urban localities. However, the findings simultaneously revealed that individual geospatial proxies frequently diverge over short intervals, warning health planners against relying upon single satellite indicators without local ground validation.

The operational setting encompasses the Kaloleni Rabai Health and Demographic Surveillance System, centred on Mariakani township across Kaloleni and Rabai sub-counties. Although the broader surveillance catchment is predominantly rural and characterized by subsistence agriculture, high poverty, informal commerce, and seasonal flooding, it contains an expanding peri-urban transport corridor linked directly to Mariakani. Within Kenya’s national community health strategy, the Community Health Unit serves as the primary operational platform for delivering preventive and promotive services, with each unit typically covering 500 to 1,000 households. The ten administratively defined units analyzed comprise Buni, Kibwabwani, Kombeni, Kwale, Mnazimwenga, Mutsengo, Mwele Kisurutini, Tsangatsini, Viragoni, and Vishakani. Because census frameworks and administrative boundaries are updated infrequently, local health managers face difficulties identifying where demographic expansion alters service demand, travel distances, and community health worker workloads, creating an operational need for objective, continuously updated geospatial stratification methods.

To capture settlement intensity across multiple dimensions, the investigators analyzed five continuous geospatial proxies summarized annually within unit boundaries using Quantum Geographic Information System software version 3.28.3. Human activity and electrification infrastructure were measured through annual mean radiance extracted from the Visible Infrared Imaging Radiometer Suite nighttime lights instrument. Physical structural expansion was tracked using two indicators: the percentage of surface area classified as built up in the high-resolution Sentinel-2 land cover product, and estimated built-up percentage provided by WorldPop. Demographic concentration was captured using WorldPop mean annual population density expressed in persons per square kilometre. Broader settlement character was represented by the Degree of Urbanization urban percentage metric, which applies population density and cluster size thresholds to gridded population data. Tabular panel harmonization verified complete geographic matching and temporal coverage across all ten units, yielding an analytic dataset of eighty unit-year observations per proxy without missing raw values.

The mathematical workflow established explicit formulas to evaluate settlement dynamics without imposing predefined categorical cut-offs during initial data processing. For each unit and indicator, the team computed relative annual percent change as the difference between the current year and the preceding year divided by the preceding year value, multiplied by one hundred. Cumulative percent change between the 2017 baseline and the 2024 endline was calculated as the net eight-year difference divided by the baseline value, multiplied by one hundred, alongside absolute arithmetic differences. Crucially, calculations involving zero denominators were treated strictly as undefined rather than being imputed. This methodological constraint affected nine annual transitions and one cumulative calculation for Sentinel-2 built-up area, alongside thirty-three annual transitions and five cumulative comparisons for the Degree of Urbanization metric. These structural zeros reflected the specific classification mechanics of underlying products rather than missing observations, preventing artificial distortion in subsequent trend estimations.

Summary metrics across the surveillance area demonstrated that baseline and endline distributions expanded across all five geospatial indicators between 2017 and 2024, confirming continuous overall regional development. Median nighttime lights radiance increased from 0.346 with an interquartile range of 0.257 to 0.523 at baseline to 0.575 with an interquartile range of 0.421 to 0.806 at endline. Modeled population density increased steadily from a median of 284.6 persons per square kilometre to 332.1 persons per square kilometre, with the interquartile range shifting upward from 110.7 to 700.6 in 2017 to 125.0 to 807.7 in 2024. WorldPop built-up area extent rose from a median of 1.322 percent to 2.002 percent across the surveillance units. Sentinel-2 built-up area expanded from an initial median of 3.162 percent to 6.419 percent, while the Degree of Urbanization urban percentage rose from a median of 26.34 percent to 30.40 percent across the community health units.

Despite widespread system-level growth, individual indicators exhibited divergence in cumulative trajectories and annual stability. Cumulative percent change from 2017 to 2024 was highest for Sentinel-2 built-up area, which posted a median expansion of 110.74 percent with an interquartile range of 39.37 percent to 218.86 percent, though this wide interval was driven partly by low starting baseline denominators in rural units. Nighttime lights radiance followed with a median cumulative rise of 62.25 percent, while WorldPop built-up extent grew by 34.14 percent and population density rose by 14.67 percent. In sharp contrast to its net growth, Sentinel-2 built-up area showed severe short-term instability: twenty-five of sixty-one valid annual transitions, representing 41.0 percent of recorded year-to-year shifts, were negative. The researchers attributed this optical volatility to tropical atmospheric interference, cloud cover, and seasonal vegetation dynamics rather than structural demolition. Meanwhile, the Degree of Urbanization metric exhibited a pronounced floor effect, with thirty-seven of eighty unit-year records remaining at zero.

Correlation analyses revealed marked discrepancies between broad geographic patterns and localized temporal movements across the observation period. Pooled raw Pearson correlation coefficients across all eighty unit-year records were high, ranging between 0.710 and 0.955. However, the researchers emphasized that these pooled estimates were descriptive summaries of an overarching spatial gradient rather than independent statistical proof, because observations were clustered repeatedly across just ten geographic units. Unit-level median correlations similarly demonstrated strong between-unit agreement, ranging from 0.736 to 0.967. Yet when annual proxy measurements were centered by subtracting each unit’s mean to isolate temporal co-movement, correlations weakened substantially, scattering from 0.076 to 0.914. Furthermore, among annual percentage change trajectories, only WorldPop built-up area and population density maintained a statistically detectable moderate correlation of 0.484, with a 95 percent confidence interval spanning 0.281 to 0.646. This statistical decoupling demonstrated that while different satellite instruments agree on where settlement intensity is concentrated, they disagree on the exact timing and short-term magnitude of local growth.

To translate these multidimensional measurements into an actionable stratification, the researchers implemented an exploratory consensus classification framework based on unsupervised machine learning. For each community health unit, median values across the 2017 to 2024 window were converted to standardized z-scores across all five indicators. Independent k-means clustering with two clusters was then applied to each standardized indicator separately, grouping units into lower-intensity and higher-intensity categories. Each unit received an overall vote score ranging from zero to five, representing the total number of proxies designating it as higher intensity. Units securing three or more votes under this prespecified majority-rule threshold were designated as consensus peri-urban, while units scoring two or fewer were categorized as consensus rural. Under this operational scheme, exactly three of the ten units, or 30.0 percent, with an exact binomial 95 percent confidence interval from 6.7 percent to 65.2 percent, achieved consensus peri-urban status. Buni and Mwele Kisurutini attained unanimous scores of five, while Vishakani achieved a score of four.

The classification framework identified distinct spatial groupings while delineating intermediate zones that complicate simple binary categorizations. Five units—Mnazimwenga, Mutsengo, Viragoni, Kibwabwani, and Tsangatsini—consistently registered zero votes across all metrics, reflecting rural contexts. Conversely, Kombeni and Kwale each registered intermediate vote scores of two, positioning them directly on the rural-urban boundary. Extensive sensitivity analyses verified the core peri-urban classification: Buni, Mwele Kisurutini, and Vishakani remained classified as peri-urban when substituting multi-year medians with longitudinal means or 2017 baseline values, and their status persisted across leave-one-proxy-out exclusion tests. Temporal resampling simulations assigned peri-urban probabilities of 1.00 to Buni, Mwele Kisurutini, and Vishakani, compared to 0.255 for Kombeni and 0.00 for the rural group. However, when evaluating 2024 endline measurements alone, executing joint k-means clustering, or applying principal component analysis where the first component accounted for 92.48 percent of standardized variance, both Kombeni and Kwale grouped with the higher-intensity cluster, indicating they represent rapidly emerging transitional zones.

The study team outlined several critical methodological caveats, emphasizing that the resulting classification represents an exploratory spatial stratification rather than an externally validated clinical planning tool. The analytic sample was constrained to ten units within a single demographic surveillance platform, and several indicators, specifically WorldPop population density, WorldPop built-up area, and Degree of Urbanization, share underlying demographic models and covariate surfaces that create structural correlation rather than fully independent verification. Furthermore, satellite indices capture physical surface properties rather than housing quality, road connectivity, electricity reliability, or healthcare service readiness. Most importantly, the authors highlighted that no health worker workload data, disease burden statistics, healthcare utilization figures, or service coverage metrics were incorporated into the model. Consequently, public health leaders were cautioned that while multi-proxy geospatial clustering offers an objective, reproducible framework for identifying settlement transitions, it cannot replace structured field verification, local healthcare expert consultation, or empirical validation against frontline community health outcomes.

Subject of Research: Geospatial classification of peri-urban settlement transitions in coastal Kenya

Article Title: A longitudinal multi‑proxy geospatial classification of peri‑urban transitions across community health units in coastal Kenya

Article References: Makena, P., Oluoch, F., Ondiek, R. I., Gudda, F., Keter, A., Das, J., Bhutta, Z., & Ngugi, A. (2026). A longitudinal multi‑proxy geospatial classification of peri‑urban transitions across community health units in coastal Kenya. International Journal of Health Geographics, 25(1), Article 52. https://doi.org/10.1186/s12942-026-00483-5

Image Credits: AI Generated

DOI: 10.1186/s12942-026-00483-5

Keywords: peri-urban classification, community health units, Kilifi County, KRHDSS, nighttime lights, Sentinel-2, WorldPop, Degree of Urbanization, k-means clustering, longitudinal ecological analysis, settlement intensity, Earth observation, VIIRS radiance, coastal Kenya, population density, spatial stratification


BHFN Editorial Team covers breaking news, culture, and global developments impacting Black America, Africa, Kenya, and the African diaspora. Focused on timely reporting and community-driven perspectives, the team delivers news, analysis, and stories that inform, connect, and amplify diverse voices.