Bayelsa State is one of the states in Nigeria and one of the states in the Niger Delta area of the country. Administratively, it has eight LGAs and 105 local wards (Fig. 1). The LGAs are divided into two categories: four upland and four riverine. The dominant vegetation comprises mangroves and freshwater swamp forests. Data on the annual incidence of confirmed uncomplicated malaria from 2017–2024 were obtained from Nigeria’s National Malaria Elimination Program through the District Health Information System 2 (DHIS2) platform (20). The DHIS2 serves as Nigeria's official national health management information system, capturing comprehensive malaria surveillance data from health facilities across all 36 states and the Federal Capital Territory. The platform provides standardized, real-time reporting of malaria case management, prevention activities, and epidemiological indicators, making it a highly credible and authoritative source for malaria data used by the Nigerian Ministry of Health, WHO, and international research organizations.

Data on the potential environmental predictors of confirmed uncomplicated malaria were downloaded from the Google Earth Engine (GEE). The GEE is a cloud-based platform for planetary-scale geospatial analysis that enables Google's massive computational capabilities to address a variety of high-impact societal issues, including deforestation, drought, disaster, disease, food security, water management, climate monitoring and environmental protection (21) Data obtained from the GEE include the land surface temperature (LST), enhanced vegetation index (EVI), normalized difference vegetation index (NDVI), soil adjusted vegetation index (SAVI), normalized difference built-up index (NDBI), normalized difference water index (NDWI), shuttle radar topographic mapping (SRTM) data, soil moisture, soil pH, monthly precipitation, and VIIRS nighttime lights.

Data on built-up areas, water coverage, and cropland coverage were extracted from the Environmental Systems Research Institute (ESRI) Living Atlas database (22). These land use/land cover data were processed from Sentinel-2 and are available at a 10-meter resolution worldwide (23). Administrative boundaries (states and LGAs) were obtained from Grid3.org (Fig. 1). All raster datasets were projected to a common coordinate system (WGS84 UTM32N). Zonal statistics were used to extract the average of each of the potential parameters for each LGA. The extracted values were combined with the administrative map of the LGAs in Bayelsa State for geospatial analysis and statistical modelling.

Descriptive analysis was conducted on the confirmed uncomplicated malaria data to understand the spatiotemporal distribution of malaria incidents across LGAs and over time in Bayelsa State. In addition, we calculated the percentage change and the compound annual growth rate of confirmed uncomplicated malaria incidents. For each LGA, we extracted the yearly count of confirmed uncomplicated malaria cases for the years 2017–2024 from the DHIS2 routine surveillance data. We treated each LGA year as one observation and computed growth rates per LGA over the full interval from 2017–2024. Compound annual growth requires the number of years between the start and end points of the observations. For the 2017 (baseline) to 2024 (endpoint) window, the growth interval is n = 2024 − 2017 = 7. For each LGA, we therefore calculated the compound annual growth rate (CAGR) of confirmed uncomplicated malaria as follows:

CAGR_i ​= (V_i,2024)​^1/7 (1)

(V_i,2017)​

​where V_i,2017​ and V_i,2024​ are the baseline and endpoint annual case counts for LGA, respectively. CAGR provides the constant annualized rate that links the beginning and ending values under geometric compounding. This is the same “geometric endpoint” growth measure widely used in official statistics and international monitoring (24). Because it is geometric, CAGR smooths year-to-year volatility and summarizes long-term changes with a single annualized figure.

We analyzed routine LGA-level counts of confirmed uncomplicated malaria cases from 2017–2024, classified each LGA as upland or riverine based on proximity to the coastline, and used one-way ANOVA in SPSS to compare group means. The dataset comprises 8 LGA-year observations (4 upland; 4 riverine). We report group descriptives with 95% confidence intervals (CIs), the ANOVA F test, and effect sizes (etasquared, epsilonsquared, and omegasquared) with CIs. Interpretation follows standard guidance for ANOVA and effect sizes (25).

Annual global Moran’s I was computed via GeoDa software to assess the spatial autocorrelation of confirmed uncomplicated malaria across Bayelsa’s eight LGAs. Moran’s I measure the similarity of values among neighboring areal units relative to the overall mean; its slope equals the regression slope in the Moran scatterplot (26–28). GeoDa software typically evaluates significance via permutation testing (e.g., 999 randomizations) to produce a pseudo p value and z score (29, 30). With N = 8 LGAs, the expected value of Moran’s I under the null hypothesis of spatial randomness is E[I] = − 1/(N − 1) = − 1/7 ≈ − 0.1429 (26, 27). The values near this benchmark are consistent with spatial randomness.

The number of confirmed uncomplicated malaria incidents in each LGA was the dependent variable, while the various environmentally related variables constituted the explanatory variables. Pearson correlation analysis was used to investigate the associations between the number of confirmed uncomplicated malaria incidents and potential environmental predictors. The average malaria incidence by LGA was correlated with environmental, infrastructural, and access covariates: VIIRS nightlight radiance, soil moisture, top soil pH (0–20 cm), SAVI, NDWI, NDVI, NDBI, EVI, land surface temperature (LST), elevation, average daily precipitation, total road length, river length, number of healthcare facilities, area of built-up land, and area of cropland. Pearson r values and 1-tailed p-values were calculated (n = 8).

Furthermore, to identify the predictors of confirmed uncomplicated malaria, exploratory and ordinary least squares (OLS) regressions were conducted. The ArcGIS Pro’s exploratory regression tool was used to screen combinations of candidate predictors for LGA--level malaria incidence in Bayelsa State and retained models that (a) explained a large share of variance (adjusted R²), (b) were parsimonious (lowest Akaike information criterion (AICc), and (c) met OLS diagnostics—residual normality (Jarque–Bera, JB), homoscedasticity/stationarity (Koenker–Breusch–Pagan, K(BP)), acceptable multicollinearity (maximum VIF < 7.5), and no residual spatial autocorrelation (Global Moran’s I; SA p > 0.05). These diagnostics follow established guidance for spatial regression and model selection (31, 32). In addition, we used the ordinary least squares (OLS) model to fit ArcGIS Pro with malaria counts as the dependent variable and four covariates identified through exploratory regression: VIIRS night-time lights (VIIRS_NIGH), mean NDVI (Mean_NDVI), mean elevation (Elevation), and total river length (Rivers). ArcGIS Pro reports both classical and heteroskedasticity-robust (White) standard errors; when the Koenker–Breusch–Pagan test is significant, interpretation should rely on robust p values (33). Multicollinearity was assessed via variance inverse factors (VIFs). The exploratory and OLS regressions were conducted via ArcGIS Pro 3.4 (34).

The spatial distribution and concentration of confirmed uncomplicated malaria show that, cumulatively, from 2017–2024, Yenagoa LGA (Bayelsa’s State capital) contributed 78,808 of 201,994 cases (39.02%), followed by Sagbama (24,914; 12.16%), southern Ijaw (22,215; 11.00%), Nembe (18,305; 9.06%), Kolokuma/Opokuma (17,920; 8.87%), Ekeremor (14,887; 7.37%), Ogbia (12,779; 6.33%), and Brass (12,526; 6.20%). Annually, Yenagoa’s share ranged from 28.25% (2018) to 49.2% (2019) and stood at 39.5% in 2024, demonstrating a persistent concentration of reported burden in that LGA throughout the period (Table 1). Yenagoa LGA had the highest confirmed uncomplicated malaria incidence in 2024, with 25,735 cases (39.5% of the state total), followed by Sagbama, with 8,327 cases (12.8%), and southern Ijaw, with 7,379 cases (11.3%). Mid-range counts were observed in Nembe (5,905; 9.1%) and Ogbia (5,844; 9.0%), while Kolokuma/Opokuma (4,582; 7.0%) and Ekeremor (4,426; 6.8%) counts were lower, and Brass (2,951; 4.5%) reported the lowest counts. All the LGAs reached their highest values in 2024. Relative to 2017, the rank positions in 2024 were stable at the top (Yenagoa first, Sagbama second, Southern Ijaw third), with upwards movement for Nembe (eighth to fourth) and Ogbia (seventh to fifth), and a decline for Brass (fourth to eighth) and Ekeremor (fifth to seventh).

The temporal trend in confirmed uncomplicated malaria incidence at the state level revealed that the number of cases rose from 10,745 in 2017 to 65,149 in 2024. This represents a 6.06-fold increase, corresponding to a compound annual growth rate (CAGR) of 29.4% (Table 2). The annual totals and year-to-year (YoY) changes were as follows: 2017, 10,745; 2018, 15,698 (+ 46.1%); 2019, 20,867 (+ 32.9%); 2020, 16,174 (− 22.5%); 2021, 16,279 (+ 0.6%); 2022, 21,800 (+ 33.9%); 2023, 35,642 (+ 63.5%); and 2024, 65,149 (+ 82.8%) (Table 2). The period was characterised by an increase in the incidence of confirmed uncomplicated malaria from 2018–2019, a clear reduction in 2020, a minimal net change in 2021, recovery in 2022, and a step change beginning in 2023 that accelerated in 2024. The total of 2024 was 3.85 times greater than the average annual total from 2017–2022 (16,927).

At the LGA level, the data revealed that confirmed uncomplicated malaria increased by 2206.64% in Nembe, 1057.23% in Ogbia, 805.53% in Kolokuma/Opokuma, 555.33% in Yenagoa, 378.84% in southern Ijaw, 355.35% in Ekeremor, 343.16% in Sagbama, and 154.62% in Brass LGAs from 2017–2024. The corresponding approximate CAGRs were as follows: Nembe, 56.5%; Ogbia, 41.9%; Kolokuma/Opokuma, 37.0%; Yenagoa, 30.8%; southern Ijaw, 25.1%; Ekeremor, 24.2%; Sagbama, 23.7%; and Brass, 14.3% (Table 3). The 2024 surge relative to each LGA’s 2017–2022 median was highest in Ogbia (7.67 times), followed by a cluster of approximately fourfold in Sagbama (4.20 times), southern Ijaw (4.19 times), Yenagoa (4.18 times), and Nembe (4.17 times). Ekeremor (3.14 times) and Kolokuma/Opokuma (2.67 times) were moderate, and Brass (2.37 times) presented the smallest relative increase.

The year-specific patterns revealed that the 2019 state-level rise was strongly influenced by Yenagoa, which accounted for 49.2% of the cases that year (10,272 of 20,867). The 2020 decline was broad-based (Table 4), with the steepest LGA-level declines from 2019–2020 in Ogbia (− 44.7%; 671–371), Kolokuma/Opokuma (− 37.9%; 2,188–1,357), Ekeremor (− 37.7%; 1,550–965), Yenagoa (− 28.5%; 10,272–7,341), and southern Ijaw (− 27.0%; 1,637–1,195). The subsequent increase was pronounced in 2022–2023 in Ogbia (+ 157%; 1,030 to 2,647), Sagbama (+ 112%; 2,014 to 4,273), Yenagoa (+ 86%; 7,729 to 14,391), Kolokuma/Opokuma (+ 58%; 2,269 to 3,591), Ekeremor (+ 32%; 1,782 to 2,351), Nembe (+ 35%; 2,734 to 3,689), southern Ijaw (+ 13%; 3,021 to 3,416), and Brass (+ 5%; 1,221 to 1,284). It intensified again from 2023–2024: Brass (+ 130%; 1,284–2,951), Ogbia (+ 121%; 2,647–5,844), southern Ijaw (+ 116%; 3,416–7,379), Sagbama (+ 95%; 4,273–8,327), Ekeremor (+ 88%; 2,351–4,426), Yenagoa (+ 79%; 14,391–25,735), Nembe (+ 60%; 3,689–5,905), and Kolokuma/Opokuma (+ 28%; 3,591–4,582).

Bayelsa State lies within the Niger Delta, where malaria transmission is perennial but ecologically heterogeneous. Inland/upland LGAs are dominated by freshwater systems, while riverine/mangrove zones experience tidal influence and brackish conditions. Because dominant vectors such as Anopheles gambiae sensu stricto and Anopheles. coluzzii prefer sunlit, shallow freshwater, whereas brackish habitats are more suitable for more localized Anopheles. melas, differences in landscape settings may translate into measurable differences in malaria burden (2, 16). We assessed whether confirmed uncomplicated malaria incidence rates differ between upland and riverine LGAs from 2017–2024.

Across the eight LGAs, from 2017–2024, malaria counts were higher in upland LGAs than in riverine LGAs and showed markedly greater dispersion in upland settings. Upland LGAs (n = 4) reported a mean of 4,189 cases per LGA per year (SD = 5,064; SE = 895), with a 95% confidence interval (CI) for the mean of 2,363-6,015 and a range of 371 − 25,735. Riverine LGAs (n = 4) averaged 2,123 cases (SD = 1,490; SE = 263), 95% CI 1,586–2,660, ranging from 256–7,379 (Table 5). The mean difference was 2,067 cases, and the ratio of group means was approximately 1.97 (upland ≈ twice riverine). Relative variability was substantially greater in upland LGAs (coefficient of variation ≈ 121%) than in riverine LGAs (≈ 70%), reflecting heterogeneous burdens and occasional high-count years in upland areas. Overall, across all observations, the mean was 3,156 cases (SD = 3,847; 95% CI 2,195–4,117; range 256–25,735). Upland LGAs report higher rates of confirmed uncomplicated malaria than Riverine LGAs do in Bayelsa State (2017–2024).

A one-way ANOVA was used to test whether the mean number of confirmed uncomplicated malaria cases differed between upland and riverine LGAs (2017–2024). The analysis revealed a statistically significant group effect (F_{(1, 62)} = 4.904, p = 0.030) (Table 6). Thus, the upland and riverine LGAs do not have the same average malaria counts. Given the descriptive results reported earlier (upland mean ≈ 4,189 vs riverine mean ≈ 2,123) (Table 5), the direction of the effect indicates higher counts in upland LGAs.

According to the ANOVA results (Table 6), the between-group mean square (68,326,756) is approximately 4.9 times greater than the within-group mean square (13,934,067.56), producing the observed F ratio (4.904). In terms of practical importance, effect size estimates computed for this comparison indicate that the group factor accounts for approximately 6–7% of the total variance (η² ≈ 0.073; ω² ≈ 0.057), representing a small to moderate effect by conventional benchmarks.

All Moran’s I coefficients are negative, indicating that the global pattern of confirmed uncomplicated malaria is one of spatial dispersion (i.e., a tendency for dissimilar values to be neighbours, a “checkerboard” pattern) rather than clustering (26, 27). The coefficients are small in absolute value (|I| < 0.20), and crucially, they hover around the null expectation for N = 8 ( ≈ − 0.1429). The Moran’s I was very close to the null value in 2017 (− 0.141), 2020 (− 0.149), and 2021 (− 0.131), while the Moran’s I was more negative than the null value in 2022. The most dispersed year was 2022 (− 0.168), while the year 2023 (− 0.069) indicates the weakest negative autocorrelation. There is mild interannual fluctuation with a trough (mostly negative) in 2022, followed by a rebound toward zero in 2023 and a slight decrease in 2024 (Table 7). No monotonic trend is evident. Given the small number of LGAs (N = 8) and the proximity of most estimates to the null expectation ( ≈ − 0.143), the global pattern is most consistent with spatial randomness in many years (30)

Nigeria remains a high burden setting where malaria risk is heterogeneous over short distances due to environmental, infrastructural, and intervention coverage gradients (33). Such heterogeneity can generate local clusters even when the statewide (global) pattern appears randomly. Across 2017–2024, confirmed uncomplicated malaria in Bayelsa State exhibited weak negative global spatial autocorrelation, with coefficients clustered around the null expectation for the eight LGAs. The statewide pattern is largely consistent with spatial randomness, punctuated by year-to-year fluctuations.

Malaria transmission in southern Nigeria is perennial and intense, driven by efficient Anopheles vectors and a warm, humid deltaic environment (33). While climate sets broad suitability, within-state heterogeneity often reflects how people live, move, and access care (5, 10). We examined LGA-level correlations between average malaria incidence and environmental/infrastructural covariates in Bayelsa to understand which landscape features align with the observed patterns.

Bayelsa’s Niger Delta ecology ensures perennial malaria receptivity, with Anopheles gambiae s.l. and Anopheles funestus sustained by warm temperatures and high humidity (2, 34). Within-state heterogeneity is therefore expected to reflect human environmental modification, population aggregation, mobility, and differential access to testing and care (5, 10). The built environment indicators, such as the area of built-up land (r = 0.954, p < 0.001), exhibited the strongest positive correlations with confirmed uncomplicated malaria. Nightlights (r = 0.706, p = 0.025) supported this pattern, which is consistent with these variables acting as complementary proxies for settlement density and urban/peri-urban expansion (5, 35). There was also a strong positive association between the incidence of confirmed uncomplicated malaria and the length of roads in each LGA (r = 0.912, p = 0.001, r²≈0.83). Roads both concentrate people and create larval habitats through borrow pits, drainage failures, and puddling along verges; they also facilitate parasite flow via human movement (9). The number of confirmed uncomplicated malaria incidents was also positively correlated with the number of healthcare facilities (r = 0.808, p = 0.008). This likely reflects detection/reporting intensity, such that LGAs with more facilities test and report more, inflating the observed incidence relative to areas with poorer access (33). This may also reflect the rational allocation of facilities to higher burden LGAs. The number of confirmed uncomplicated malaria incidents was also positively correlated with the area of cropland (r = 0.711, p = 0.024). Agricultural landscapes, especially low-lying and poorly drained fields, irrigation ditches, ponds, and wheel ruts, frequently generate sunlit, shallow water bodies favored by major vectors (13, 14). The vegetation indices (SAVI, r = 0.108; NDVI, r = 0.162; EVI, r = 0.142) and NDWI (r = 0.129) showed little explanatory power. LST (r = 0.108) and precipitation (r = − 0.250) were also weak. Elevation (r = 0.399) and soil pH (r = 0.586, p = 0.063) tended to increase but were not significant. River length was weakly negative (r = − 0.284). In Bayelsa’s consistently warm, rainy environment, climatic gradients between LGAs are modest and remain near the thermal optimum for Plasmodium falciparum transmission, limiting their explanatory value at this spatial scale (3, 36). Large rivers and tidal channels are typically poor larval habitats compared with small, sunlit, stagnant pools (2).

The exploratory regression analysis yielded three potential models for explaining variations in confirmed uncomplicated malaria among LGAs in Bayelsa State. These candidate models had very high fits (AdjR² ≥ 0.97) and passed key diagnostics (Table 6). AICc differences and diagnostics, however, identify Model 1 as the strongest, with Model 2 being a plausible alternative and Model 3 receiving comparatively less support. Model 1 (best-supported) identified increased VIIRS Nighttime Light (+), reduced NDVI (-), increased Elevation (+) and increased length of streams/rivers (+) as the most likely predictors of confirmed uncomplicated malaria among LGAs in Bayelsa State (Table 8).

The VIIRS nighttime lights (positive) proxy settlement density, electrification, and economic activity. Brighter, denser LGAs typically exhibit more peri-urban habitats (blocked drains, borrow pits, water containers) that support Anopheles gambiae s.l., sustaining focal transmission (5, 6, 37, 38). The negative NDVI suggests that greener and more vegetated LGAs are less prone to malaria at this scale, which is consistent with the preference of dominant West African vectors for small, sunlit, shallow freshwater pools rather than shaded habitats. Increased greenness often corresponds to canopy covers or wetlands that are less productive for these vectors (2, 14). The positive elevation aligns with Bayelsa’s ecology. The slightly higher inland/upland LGAs provide fresher water bodies, whereas low-lying, saline/brackish tidal zones are less suitable for Anopheles. gambiae s.s./Anopheles. coluzzii; coastal Anopheles. melas is more saline-tolerant but have a narrower distribution (2, 16). Additionally, the positive relationship between the length of rivers/streams likely captures floodplain edges, borrow pits, and human settlements along waterways, where overbank flooding and poor drainage create discrete larval habitats and increase exposure (14). This model combines the lowest AICc with a very strong fit and acceptable diagnostics, indicating the best trade-off between explanatory power and parsimony (Table 8). Its predictors also match established transmission mechanisms: anthropogenic settlement intensity coupled with freshwater availability.

The second model is ecologically coherent and statistically sound, but its higher AICc provides less support than Model 1 does. Using Akaike weights, Model 1 receives ~ 72% of the support, and Model 2 receives ~ 19%. Model 3 diagnostics are strong, and multicollinearity is minimal, but the higher AICc and slightly lower fit make this model less competitive (Table 8).

Nighttime lights and built-up areas consistently appear in the best models, indicating that settlement density, nighttime lights, and rapid peri-urban growth structure malaria risk, likely via the creation of artificial breeding sites and increased human‒vector contact (5, 6, 39). Hydroclimate factors, such as soil moisture, precipitation, rivers, and cropland, confirm that water availability and management affect habitat persistence, especially in floodplain/upland mosaics typical of Bayelsa (14, 36). The nonsignificant Moran’s I on residuals (SA p ≥ 0.32) indicates no remaining unmodelled spatial structure, while the Koenker test p > 0.1 suggests that relationships are spatially stable at the LGA scale and that the VIF values are within accepted limits, indicating manageable collinearity (32).

The variables identified by the exploratory regression were further analyzed OLS regression in ArcGIS Pro software to identify the contributions of the identified explanatory factors to malaria incidence in Bayelsa State. All four predictors in Model 1 are highly significant under both classical and robust inference (all p < 0.001), indicating stable associations even if the residual variance is nonconstant. The variance inverse factors (VIF) values are acceptable: 1.19 (VIIRS), 3.44 (NDVI), 6.19 (elevation), and 2.73 (river) (Table 9). Although elevation shows moderate collinearity, it remains below the common thresholds (≤ 7.5) recommended for ArcGIS OLS (32) Additionally, because covariates are at different scales, the relative importance is best gauged by t-statistics. With respect to robust t values, the variable influence ranks were as follows: VIIRS (25.86) > Elevation (18.89) > NDVI (|12.88|) > Rivers (11.23) (Table 9).

The VIIRS nighttime light data show a strong positive association with confirmed uncomplicated malaria across the LGAs in Bayelsa State from 2017–2024 (β = 10,880.47; robust p < 0.001; VIF 1.19). Nighttime light radiance is a validated proxy for settlement density, electrification, and economic activity (39). Brighter LGAs are typically more urban/peri-urban, where rapid expansion, poor drainage, construction of burrow pits, and container storage create numerous sunlit freshwater habitats near households. These settings sustain Anopheles gambiae. and maintain focal transmission despite some urbanization-related protection (2). The NDVI (β = −48,331.81; robust p < 0.001; VIF 3.44) also showed a strong negative association with confirmed uncomplicated malaria. A higher NDVI indicates denser/greener vegetation, which often corresponds to shaded, forested, or permanently inundated environments that are less favorable for the sunlit, shallow pools preferred by dominant West African vectors (2) Studies repeatedly show that small, open, human-made or modified water bodies in more sparsely vegetated peri-urban/agricultural mosaics are productive larval habitats (12, 14). Elevation was also positively associated with confirmed uncomplicated malaria (β = 817.86; robust p < 0.001; VIF 6.19). Within Bayelsa’s low relief delta, slightly greater inland/upland areas are less brackish and provide fresher breeding waters favoured by Anopheles. gambiae s.s./Anopheles. coluzzii, whereas the lowest coastal/mangrove zones are more saline and better suited to Anopheles. melas, which is geographically restricted (2, 16). This gradient is consistent with greater receptivity away from tidal marshes. The length of streams/rivers was positively associated with confirmed uncomplicated malaria (β = 3.38; robust p < 0.001; VIF = 2.73). River networks and their floodplains generate abundant fringe habitats, such as burrow pits and post flood residual pools, while also concentrating settlements and human activity along banks. These conditions increase both larval habitat availability and human–vector contact (4, 40).