How Environment and Urbanization Shape Bird Diversity in Sri Lanka

Abstract: This study presents a comprehensive analysis of bird diversity across Sri Lanka by integrating spatial, temporal, and environmental data. Bird observation records were combined with environmental variables, including weather conditions, air pollution, the Normalized Difference Vegetation Index (NDVI), land cover, elevation, and Artificial Light At Night (ALAN), and rigorously preprocessed to ensure data quality. Spatial analyses were conducted on multiple grid scales (2 km, 5 km, 10 km) to evaluate patterns in species richness while minimizing sampling bias through spatial thinning. Temporal trends were assessed using effort-corrected metrics including rarefied richness and occupancy rates to account for variations in observation effort over time. Environmental drivers of bird diversity were examined using multivariate statistical models, including Poisson Generalized Linear Models (GLMs) and correlation analyses, to identify key associations between ecological factors and species richness. Additionally, community structure, dominance patterns, and beta diversity were analyzed to understand variations in species composition across regions and time. The study found that land-cover type is a stronger predictor of bird diversity than individual continuous variables such as NDVI or temperature alone. Urbanization, measured by ALAN, exhibits nuanced scale-dependent effects, supporting high abundances of a few generalist species while reducing overall richness. The findings provide actionable insights into the patterns and drivers of avian diversity in Sri Lanka, offering a scalable and reproducible framework for biodiversity research and conservation planning.

• The paper confirms that categorical habitat types are better predictors of bird diversity in Sri Lanka than continuous environmental variables, highlighting the importance of land-cover type over measures like NDVI
• The presence of artificial light at night, which exemplifies urban areas, was revealed to exhibit a limited nuanced relationship with avian diversity; species abundance declined significantly where there is a surplus of light, providing evidence of biotic homogenization where only a few species dominate.
• The study found notable increases in species turnover across different regions and months implying geographically diverse conservation strategies.

Overview and Motivation

This paper presents a national-scale assessment of avian diversity in Sri Lanka, integrating citizen-science occurrence records with satellite-derived environmental indicators to identify the dominant drivers of species richness, community structure, and temporal trends. Sri Lanka is an appropriate study system because it combines high ecological heterogeneity—tropical lowland rainforest, montane cloud forest, dry-zone scrubland, and coastal wetlands—within a small geographic extent, while facing intensifying urbanization and land-use change. The authors address three questions: whether categorical habitat variables explain richness better than continuous variables such as NDVI or climate; whether artificial light at night (ALAN), as an urbanization proxy, is associated with reduced diversity and biotic homogenization; and whether observed patterns are stable across spatial resolutions of 2 km, 5 km, and 10 km (2607.00582).

Data Integration and Bias Correction

The analytical dataset combines approximately 1.55 million bird records from GBIF/eBird (2014–2024) with MODIS NDVI and IGBP land-cover classification (17 classes), VIIRS nighttime radiance, MERRA-2 reanalysis variables (temperature, rainfall, wind speed, humidity, aerosols), and SRTM elevation, matched by coordinates and observation month. Continuous covariates were averaged within grid cells and categorical variables assigned their modal class, ensuring point occurrences were matched to the broader environmental baseline.

Because eBird observations are strongly clustered—55.99% in western/southern coastal districts versus only 1.8% in the east—the authors applied grid-based spatial thinning retaining exactly one record per species per grid cell per district. A key methodological result is the stability of island-wide species count at 429 across all tested resolutions, supporting the robustness of district-level summaries to thinning scale. District-level rarefaction (300 subsamples) equalized sampling effort for fair comparisons, and temporal analyses relied exclusively on effort-corrected metrics: rarefied annual richness, richness normalized per 100 sampled cells, and occupancy computed on a "stable panel" of repeatedly surveyed cells.

Habitat Structure Dominates Over Continuous Variables

The central finding is that categorical land-cover type is a stronger predictor of bird diversity than individual continuous environmental variables. A Kruskal–Wallis test on records across IGBP classes yielded $H > 1000$ ($p < 0.001$), with evergreen forests and woody savannas supporting substantially higher and more stable richness than human-altered landscapes. In contrast, correlations between NDVI and raw counts were negligible (Pearson $r = -0.022$; Spearman $\rho = -0.094$), and richness showed only weak positive association ($\rho = 0.033$), with statistically significant $p$-values attributable to the very large sample size rather than ecological effect size. The implication is that conservation value cannot be inferred from greenness alone; habitat structure, composition, and disturbance regime matter more than productivity measured as NDVI.

In the Poisson GLM with HC1 robust standard errors ($n = 1{,}736$ cells), significant terms included NDVI (negative, $p \approx 0.047$), nighttime radiance (positive, $p \approx 0.002$), mean temperature (negative, $p \approx 0.002$), and selected land-cover contrasts; elevation, rainfall, and aerosol extinction were not significant. The reported Cox-Snell pseudo $p < 0.001$0 reflects in-sample fit only—no hold-out validation or overdispersion testing was performed—and should not be read as out-of-sample predictive performance, particularly given that rank agreement in the corresponding OLS model was modest (Spearman $p < 0.001$1, $p < 0.001$2) and climate-only models achieved $p < 0.001$3.

Urbanization, ALAN, and Biotic Homogenization

ALAN exhibited nuanced, scale-dependent effects. Strong positive correlations between ALAN and MERRA-2 aerosol variables confirm co-location of infrastructure, activity, and pollution, while LOWESS regression shows a steep decline of NDVI with rising ALAN, indicating progressive habitat degradation along the urban gradient. Yet bird responses are heterogeneous: highly lit areas show reduced Shannon evenness, elevated Berger-Parker dominance, and lower pairwise Jaccard dissimilarity between urbanized cells—direct quantitative evidence of biotic homogenization, whereby a few synanthropic generalists dominate while overall richness declines. Notably, the hexbin analysis of per-record counts against radiance shows that apparent large urban flocks are better explained by uneven observer effort than by genuinely larger urban aggregations, a candid qualification of the homogenization evidence.

Air pollution showed negligible global correlation with richness ($p < 0.001$4) but strong taxon-specific associations—for example, Merops persicus with $p < 0.001$5 against several pollutant metrics—implying pollution sensitivity depends on species traits such as feeding ecology rather than operating uniformly across communities.

Temporal Patterns

Monthly richness peaked in January–March and November–December, and cross-correlation analysis indicated that richness aligns best when rainfall leads by several months, consistent with delayed vegetation-mediated responses. Effort-corrected rarefied richness and district-year occupancy rates confirmed these trends are not artifacts of sampling variation. After standardizing richness per 100 sampled cells, most heavily sampled districts show a stabilizing or gradually declining trend from 2014–2024, which the authors interpret as consistent with ongoing anthropogenic pressure suppressing avian complexity. Pairwise Jaccard dissimilarity revealed substantial compositional turnover between districts, implying that conservation strategies should be region-specific rather than uniform island-wide. Species-level log-linear trends under Benjamini–Hochberg FDR correction confirmed heterogeneity of change across taxa. The authors appropriately note that these slopes represent effort-corrected apparent observation trends rather than confirmed population dynamics, given detection probability was never explicitly modeled.

Limitations and Open Questions

The paper is explicit about several constraints. Spatial autocorrelation was not modeled via spatial autoregressive weights, so all GLM outputs are associational summaries rather than causal diagnostics. Detection probability was not estimated, limiting inference on true occupancy. No cross-validation, alternative count-model comparison (e.g., negative binomial), or formal overdispersion testing was conducted despite acknowledged overdispersion risk. Only classes with sufficient observations appear in the NDVI-by-land-cover visualization, so a subset of the 17-category scheme drives much of the interpretation. Open questions include whether the ALAN-richness relationship at fine scales reflects observer effort rather than ecology, how taxon-specific pollution sensitivities map onto life-history traits, and whether the declining effort-standardized richness trend continues beyond heavily sampled districts.

Conclusion

By unifying eBird occurrence data, remote sensing, climate reanalysis, and pollution metrics within a single bias-corrected pipeline, this work demonstrates that land-cover class—not continuous productivity or climatic measures—is the primary correlate of avian richness in Sri Lanka, while urbanization promotes biotic homogenization through dominance by a few generalists. The multi-scale sensitivity analysis, spatial thinning, rarefaction, and FDR-controlled temporal regressions constitute a reproducible framework adaptable to other taxa and geographies, though its conclusions remain associational pending explicit treatment of spatial dependence and detectability.