Figure 2

Change in relative primary productivity (scaled NDVI) across the calendar year in the two study areas. We used data from Terra MODIS66 and CORINE Land Cover 201267 to extract NDVI values (2014–2018) for each 250-m2 pixel classified as suitable foraging habitat (Supplementary Table S4) within a 3-km radius of each colony68. We used GAMMs to model NDVI against date smoothed with penalised regression splines, interacting with colony area, with pixel ID as a random effect. The interactions between date and the two levels of study area were significant in the model (p < 0.0001).