Land use and anisotropy of artificial light observed by night time satellite

TL;DR

This study examines how land use influences the anisotropy of upward artificial night-time light by integrating VIIRS-DNB multi-angle radiance data with the CORINE 2018 land-use map. It fits a quadratic angular radiance model $L_{\mathrm{fit}}(\theta)=a\theta^2+b\theta+L_{\mathrm{fit,nadir}}$ per grid cell and normalizes coefficients to $a_{\mathrm{rel}}$ and $b_{\mathrm{rel}}$ for cross-class comparison. The key finding is that most land-use classes exhibit zenith-dominant emission ($a_{\mathrm{rel}}>0$), with only Continuous urban fabric ($111$) and Sparsely vegetated areas ($333$) showing negative $a_{\mathrm{rel}}$, while $b_{\mathrm{rel}}$ patterns vary and tend to become more negative as radiance increases, indicating more sideways emission at higher brightness. The results, though consistent with urban morphology effects, are limited by VIIRS-DNB’s spatial resolution, motivating higher-resolution, openly licensed sensors to resolve street-level emission patterns and improve cross-domain applicability.

Abstract

Upward emission of artificial light has been investigated by researchers since the commissioning of the Visible/Infrared Imaging Radiometer Suite (VIIRS) Day/Night Band (DNB) in 2011, with applications ranging from night time light mapping to quantifying socio-economical development. The wide swath of the VIIRS-DNB sensor enables detection of artificial light at multiple angles and was utilized to study emission of artificial light from cities at different angles as well as atmospheric properties. Existing studies of the relationship between the directionality and land surface features are not available for most of the Earth's surface due to the use of space-borne LiDAR as a source of proxy. To solve this problem, we compared the land use data published under the Coordination of Information on the Environment (CORINE) against the fit parameters of radiance of upward artificial light. In general, the quadratic term of the fit, which quantifies how the brightness changes when viewing closer from the horizon at a point on the Earth, is negative when the area is "Continuous urban fabric" or "Sparsely vegetated areas", and vice versa for all other investigated land use classes. However the quadratic term shifts towards negative values for brighter areas. These results indicate that while densely built areas emit more light towards the zenith than sideways, the VIIRS-DNB is unable to distinguish small densely built areas scattered around larger unbuilt areas. Therefore, sensors with higher spatial resolution will be required to resolve the light emission patterns of areas with complicated combinations of land uses.

Figures (4)

• Figure 1: Distribution of $a_{\textrm{rel}}$ values for different land use classes. For each class, the thick error bar represents the interval between the 16th and 84th percentiles, and for the thin error bar between 5th and 95th percentiles. The vertical bar shows the median value. Only subdatasets for land use classes where there are at least 200 grid cells satisfying the selection criteria (at least 20 overflights, no moon or twilight, fitted nadir radiance at least $5 \cdot 10^{-5}$$\textrm{W}/(\textrm{sr}\cdot\textrm{m}^{2})$) are shown in this Figure.
• Figure 2: Distribution of $b_{\textrm{rel}}$ values for different land use classes. For each class, the thick error bar represents the interval between the 16th and 84th percentiles, and for the thin error bar between 5th and 95th percentiles. The vertical bar shows the median value. Only subdatasets for land use classes where there are at least 200 grid cells satisfying the selection criteria (at least 20 overflights, no moon or twilight, fitted nadir radiance at least $5 \cdot 10^{-5}$$\textrm{W}/(\textrm{sr}\cdot\textrm{m}^{2})$) are shown in this Figure.
• Figure 3: Distribution of $a_{\textrm{rel}}$ term values for different land use classes, using the same plotting scheme as \ref{['fig1']}, but only for areas with $2.5 \cdot 10^{-4}$$\textrm{W}/(\textrm{sr}\cdot\textrm{m}^{2})$$\leq L_{\textrm{fit,nadir}}\leq 1.25 \cdot 10^{-3}$$\textrm{W}/(\textrm{sr}\cdot\textrm{m}^{2})$).
• Figure 4: Change in distribution of $a_{\textrm{rel}}$ at different $L_{\textrm{fit,nadir}}$ for land use classes. The x-axis is $L_\textrm{fit,nadir}$ in logarithmic scale binned in the interval of $\sqrt{3}$-fold each, from $5 \cdot 10^{-5}$$\textrm{W}/(\textrm{sr}\cdot\textrm{m}^{2})$ to $2.34 \cdot 10^{-3}$$\textrm{W}/(\textrm{sr}\cdot\textrm{m}^{2})$. For each subplot, the gray horizontal line is the zero point for $a_{\textrm{rel}}$, the black line the $log(a_{\textrm{rel}})$ fit with slope of the fit shown in scientific notation, in the unit of $log(\textrm{sr}\cdot\textrm{m}^{2}/ \textrm{W}) \cdot ^{\circ -2}$, and the class number is shown on the top right corner. Refer to Figures \ref{['fig1']} to \ref{['fig3']} for the corresponding land uses of the classes.