2.2. Study design
We selected 189 points using a random sampling method (minimum distance between points was 473 m; Fig. 1), stratified according to the main three land-uses: agricultural, n = 86; forest, n = 50; and urban = 53. To characterise the landscape around each point, we considered composition and configuration variables within a buffer of 500 m radius. Composition was expressed as the percentage cover of seven categories of land-use (Carta de Ocupação do Solo maps available for Portugal; IGP, 2020) that were a priori considered as potentially relevant to bird distributions within different (decreasing) levels of human disturbance: i) urban and industrial, ii) intensive agriculture, iii) rice-fields, iv) extensive agriculture, v) shrubland, vi) plantation forest, and vii) native forest (e.g. montado). Landscape configuration was calculated using the Shannon-Weiner diversity index of the seven land use categories (SHDI). Landscape composition and configuration variables were calculated using QGIS 3.26.3-Buenos Aires andlandscapemetrics R package (Hesselbarth et al., 2019). We scaled all landscape variables to better evaluate collinearity (Cade, 2015), testing for multicollinearity through the Spearman correlation coefficient. We retained all variables since Spearman’s Rho was lower than 0.63 in all cases (Dormann et al., 2013). For a more detailed description, see Supporting Information 1.