Fig. 4: Scale dependent effects of area protection on cloud cover.

Location of World Database of Protected Areas (WDPA) (opaque) and reference areas (semi-transparent) in the study region (transparent) (a). Difference in mean annual Normalized Difference Vegetation Index (NDVI) (b) and elevation (c) between the WDPA and reference areas. The difference in average April–September 07:00–15:45 cloud cover frequency (ΔCCF) between the WDPA and the corresponding reference area, per size of the WPDA area (d, e). The reference area consists of a 10 km buffer around the WDPA area, where overlapping WDPA areas are not considered. Points in (d) show ΔCCF for the individual areas. The line shows a linear regression between ΔCCF and log(WDPA size). The green line in (d) shows a linear least-squares regression, including the Pearson’s correlation coefficient (r²) and statistical significance (p) calculated with the Wald test. Boxes in (e) show the median (line/point), the interquartile range (box) and 1.5 times the interquartile range (whiskers) of the data grouped per 10-percentile of WDPA size. Each box contains 32 WDPA areas. The width of the boxes represents the range of WDPA sizes within the 10-percentile. Note that CCF is first averaged over time (based on ~131,760 individual images) and within the WDPA area and the reference areas, after which the ΔCCF is calculated. The boxes represent the variation in ΔCCF across WDPA areas only. Green boxes indicate that the median is significantly different to zero (p < 0.05, using the Wilcoxon signed-rank test). Results for a 5 km and 15 km buffer are respectively shown in Supplementary Figs. 10 and 11, illustrating a larger variation in ΔCCF with a larger buffer.