An Estimate of the Pixel-Level Connection between Visible Infrared Imaging Radiometer Suite Day/Night Band (VIIRS DNB) Nighttime Lights and Land Features across China

Ting Ma

doi:10.3390/rs10050723

Abstract

Satellite-derived nighttime light images are increasingly used for various studies in relation to demographic, socioeconomic and urbanization dynamics because of the salient relationships between anthropogenic lighting signals at night and statistical variables at multiple scales. Owing to a higher spatial resolution and fewer over-glow and saturation effects, the new generation of nighttime light data derived from the Visible Infrared Imaging Radiometer Suite (VIIRS) day/night band (DNB), which is located on board the Suomi National Polar-Orbiting Partnership (Suomi-NPP) satellite, is expected to facilitate the performance of nocturnal luminosity-based investigations of human activity in a spatially explicit manner. In spite of the importance of the spatial connection between the VIIRS DNB nighttime light radiance (NTL) and the land surface type at a fine scale, the crucial role of NTL-based investigations of human settlements is not well understood. In this study, we investigated the pixel-level relationship between the VIIRS DNB-derived NTL, a Landsat-derived land-use/land-cover dataset, and the map of point of interest (POI) density over China, especially with respect to the identification of artificial surfaces in urban land. Our estimates suggest that notable differences in the NTL between urban (man-made) surfaces and other types of land surfaces likely allow us to spatially identify most of the urban pixels with relatively high radiance values in VIIRS DNB images. Our results also suggest that current nighttime light data have a limited capability for detecting rural residential areas and explaining pixel-level variations in the POI density at a large scale. Moreover, the impact of non-man-made surfaces on the partitioned results appears inevitable because of the spatial heterogeneity of human settlements and the nature of remotely sensed nighttime light data. Using receiver operating characteristic (ROC) curve-based analysis, we obtained optimal thresholds of the nighttime light radiance, by equally weighting the sensitivity and specificity of the identification results, for extracting the nationwide distribution of lighted urban man-made pixels from the 2015 annual composite of VIIRS DNB data. Our findings can provide the basic knowledge needed for the further application of current nighttime light data to investigate spatiotemporal patterns in human settlements.

Highlights

Sensed anthropogenic lighting signals at night provide us with a proxy measure of the magnitude of human activity over both time and space [1–3]
In comparison with the Defense Meteorological Satellite Program (DMSP) data, current nighttime light images derived from the Visible Infrared Imaging Radiometer Suite (VIIRS) day/night band (DNB) likely enable us to obtain a detailed look at human settlements with a wider radiometric detection range and a higher spatial resolution [18–22]
The characteristics of remotely sensed nighttime light data, including the presence of an over-glow effect and the lack of textural features, create a challenge for an investigation of human activity over space at a fine scale, we can obtain a detailed look at the anthropogenic nighttime brightness over human settlements from VIIRS DNB images

Summary

Introduction

Sensed anthropogenic lighting signals at night provide us with a proxy measure of the magnitude of human activity over both time and space [1–3]. Excluding ephemeral lights and background noises, remotely sensed nocturnal radiance signals and their spatial variations in the cloud-free composite VIIRS DNB data are mainly determined by artificial lighting sources at night and their spatial arrangements in human settlements. The selection of an optimal brightness threshold is always difficult, and such thresholds are often questioned because of the diversity of their size and form and the variability in the socioeconomic status of human settlements across different regions; (iv) because of these drawbacks, most investigations are limited to regional- or sub-regional-level surveys of human activity; (v) the impact of pixel-level uncertainties and biases cannot be eliminated at the regional level when we mainly focus on artificial surface-related investigations, because we are unable to completely (or even mostly) exclude lighted non-man-made pixels

Objectives

Methods

Findings

Conclusion