Spatiotemporal Characterization of VIIRS Night Light

TL;DR

This study analyzes nearly a decade of VIIRS night light data to distinguish true changes in light sources from imaging artifacts, revealing pervasive heteroskedasticity and stability of anthropogenic lights on subannual scales.

Contribution

It introduces a comprehensive spatiotemporal analysis of VIIRS night light variability, highlighting the impact of imaging process variability on observed changes.

Findings

Heteroskedasticity of night light variability across all regions.

Anthropogenic night light remains remarkably stable on subannual time scales.

Most observed variability results from imaging process effects rather than actual changes.

Abstract

The VIIRS Day Night Band sensor on the Suomi NPP satellite provides almost a decade of observations of night light. The daily frequency of sampling, without the temporal averaging of annual composites, requires the distinction between apparent changes of imaged night light related to the imaging process and actual changes in the underlying sources of the light being imaged. This study characterizes night light variability over a range of spatial and temporal scales to provide a context for interpretation of changes on both subannual and interannual time scales. This analysis uses a combination of temporal moments, spatial correlation and Empirical Orthogonal Function (EOF) analysis. A key result is the pervasive heteroskedasticity of VIIRS monthly mean night light. Specifically, the monotonic decrease of temporal variability with increasing mean brightness. Anthropogenic night light is remarkably stable on subannual time scales. Overall variance partition derived from the eigenvalues of the spatiotemporal covariance matrix are 88%, 2% and 2% for spatial, seasonal and interannual variance in the most diverse geographic region on Earth (Eurasia). Heteroskedasticity is present in all areas for all months, suggesting that much, if not most, of observed month-to-month variability may result from luminance of otherwise stable sources subjected to multiple aspects of the imaging process varying in time. Given the skewed distribution of all night light arising from radial peripheral dimming of bright sources, even aggregate metrics using thresholds must be interpreted in light of the fact that much larger numbers of more variable low luminance pixels may statistically overwhelm smaller numbers of stable higher luminance pixels and cause apparent changes related to the imaging process to be interpreted as actual changes in the light sources.