Land Use, Habitat Features and Bird Diversity in Northern Italy

Bird diversity reflects how well species thrive in their environment, and variations in land use and habitat structure can shift it substantially. This study uses a dataset on birds in Northern Italy, collected by a team of ecologists: 906 observations across 38 variables, covering species richness, functional richness and functional evenness alongside environmental and habitat measures.

The key predictors are dominant land use, spatial buffer zones at 100m and 250m, and habitat features including urban areas, rivers, water bodies, forests, grasslands, orchards, riparian vegetation and roads. Edge density is included to capture habitat structure.

Species richness is a count, so Poisson regression was the natural starting point, with log(SUM_AREA) as an offset so richness is modelled per unit area rather than as an absolute count. The dispersion test returned low p-values for both scales, meaning the equidispersion assumption failed. Both models were refitted as Negative Binomial, which lets variance exceed the mean.

Sites outside the 100m buffer show a sharply lower richness rate, and forest and grassland sites are lower than agricultural ones. But the interactions run the other way: outside the buffer, grassland sites gain back more than fourfold. Protection matters most where the surrounding land use is poorest.

Functional richness is continuous and strictly positive, so a Gamma GLM with a log link was used, keeping predictions positive and treating habitat effects as multiplicative. Six models were built, from a main-effects baseline through five variants adding ecologically motivated two-way interactions such as urban with water, grassland with water, and forest with riparian vegetation.

Water, forest and grassland area each reduce functional richness on their own, with grasslands the strongest of the three. Both interactions push the other way: where grassland meets water, and where urban land meets water, the joint presence moderates the decline. Residual deviance of 570.13 against a null deviance of 634.61 supports the fit.

A Gamma GLM with a log link again, this time with an interaction between maximum edge density and dominant land use, so the edge-density relationship could be read separately for each land use type. Three sparse categories, riparian vegetation or reforestation, river, and urban, were merged into a single “Other” level: the first two held one observation each and urban only 14, too few for stable interaction estimates.

Grasslands is the only significant term: functional evenness runs about 5.7% lower there than in the reference category. Edge density itself moves evenness by roughly 0.03% per unit, not significant at p = 0.142, and not one interaction term reaches significance. The effect of edge density does not meaningfully differ across land use types.