A longitudinal multi‑proxy geospatial classification of peri‑urban transitions across Community Health Units in coastal Kenya

Background: Urbanization is spatially heterogeneous in many sub-Saharan African settings. This complicates community health planning when rural and peri-urban labels are applied informally or updated inconsistently. We assessed whether multiple publicly available geospatial proxies could characterize settlement dynamics and support Community Health Unit (CHU) level classification within the Kaloleni Rabai Health and Demographic Surveillance System (KRHDSS) in coastal Kenya. Methods: We conducted a longitudinal ecological analysis across 10 CHUs in Kilifi County with annual observations from 2017 to 2024. Five geospatial proxy indicators of settlement intensity were summarized within CHU boundaries: Visible Infrared Imaging Radiometer Suite (VIIRS) nighttime lights radiance, Sentinel-2 built-up area percentage, WorldPop built-up area percentage, WorldPop population density, and Degree of Urbanization urban percentage. The built-up proxies quantify the extent of built surfaces and do not distinguish construction materials or housing quality. We quantified year-to-year and cumulative percent change, assessed concordance using Pearson correlation coefficients, and derived a consensus classification using k-means clustering applied separately to standardized CHU-level median proxy values, followed by cross-proxy vote scoring. Results: The analytic dataset comprised 80 CHU-year observations per proxy with no missing raw proxy values. Degree of Urbanization showed a pronounced floor effect with 37 zero observations. Median cumulative percent change from 2017 to 2024 was 62.25% for nighttime lights, 110.74% for Sentinel-2 built-up area, 34.14% for WorldPop built-up area, and 14.67% for population density. Sentinel-2 built-up area showed substantial interannual variability, with 41.0% of valid year-to-year transitions negative. Pearson correlation coefficients between pairs of proxy indicators, using pooled raw values across all 80 CHU-year observations, ranged from 0.710 to 0.955. Cross-proxy vote scoring classified 3 of the 10 CHUs as consensus peri-urban. Conclusions: A multi-proxy geospatial approach provides a reproducible complementary framework for delineating CHU settlement contexts in transitional settings. It may support periodic reassessment alongside established administrative classifications where settlement patterns are changing. Divergent proxy patterns, including volatility and floor effects, highlight the value of combining complementary products when distinguishing peri-urban from rural contexts for routine health system planning.