Abstract
Spatial and chromatic visual processing is tuned to the statistical structure of the visual environment, and these statistics have been described for a wide variety of natural and unnatural scenes. However, the extent to which these statistics vary across natural environments or over time (e.g., because of changes in the time of day, the seasons, or the weather) remains uncertain. To characterize these variations, we analyzed images from the US National Parks Air Quality Webcameras that were acquired over the course of one year (2015). In theory, a panoramic image was taken every 15 minutes across 20 (static) locations at different parks for a total of 693,500 images. In practice, we were able to access N = 539,390 images (78%) which were cropped to 1024 x 1024 pixels, and thresholded to reject images taken at night. Color statistics were based on the (uncalibrated) RGB values and included variations in the mean and gamut of the scene, while spatial analyses were based on greyscaled images and included contrast, fractal dimension, and amplitude spectrum slope. For each we compared the range of variation within location (over time) vs. across locations (over space). The results suggest that some statistical properties of scenes (e.g., amplitude slope and fractal dimension) tend to be more stable across temporal vs. spatial contexts while others (e.g., mean chromaticity) tend to show greater variation within the same scenes over time compared to location. In further analyses we compare these variations over short (e.g., within-day) and long (e.g., seasonal) timescales. Our results point to the degree to which simple image statistics can vary across a range of natural environments, and thus to the range and timescales of adaptations that would be required to adjust to a specific environment.