Abstract
Rapid expansion of remote sensing applications in urban landscapes has increased the opportunities for cost-effective inference of vegetation cover proxies using spectral indices of greenness. However, as these studies and data synthesis efforts advance, an important question arises about the statistical comparability and robustness of greenness indicators across different remote sensing sources. This study tested whether and to what extent the statistical distribution of a popular spectral greenness indicator (Normalized Difference Vegetation Index, or NDVI) changes among popular image products when the latter are acquired over the same study area very closely in time. Using the city of Sacramento, California, USA as a test case study, NDVI metrics were extracted from four remote sensing products at 1m, 10m, 30m, and 500m spatial resolution at nearly the same observation time in June 2022. Comparison of statistical distributions of NDVI indicated that central values tend to progressively increase with coarser pixel sizes, making the city appear on average “greener” at moderate to coarser resolutions than at higher resolutions. In contrast, metrics of dispersion such as standard deviation, and the occurrence of extreme high and low index values decreased with coarser spatial resolution. Aggregation of high-resolution products to coarser pixel scales of other products did not produce equivalent statistical distributions. Together, results indicate that the magnitude and variation of remote sensing-derived greenness indicators in cities must be interpreted relative to both the source sensor product and its spatial resolution. Future work should investigate this scale sensitivity more in-depth expanding to a larger sample of cities in different bioclimatic regions and develop generalizable guidance on navigating across different scales of widely used remote sensing datasets.