Full Article Viirs Night Time Lights
A total of five global grids are produced as a VIIRS night-time lights suite. The global grids are huge, 86,400 grid cells wide and 33,600 high. This exceeds the size limitation for the standard geotiff format, so the global grids are cut into six tiles. The first grid tallies the number of usable CVGs or observations that pass through the primary filtering to remove pixels contaminated by sunlight, moonlight, stray light, lightning, and HEP (Figure 6). The histogram (Figure 7) of the CVG image shows that 0.003% of grid cells have zero CVGs and 97% of the grids cells have more than 100 CVGs in 2015. The histogram shows a series of peaks and troughs that appear to be latitude related variations in the usable CVGs.VIIRS night-time lightsAll authors
Christopher D Elvidge, Kimberly Baugh, Mikhail Zhizhin, Feng Chi Hsu & Tilottama Ghosh
Figure 6. Grid tallying the number of usable coverages (CVG) for 2015. The tally includes pixels that passed filtering steps for sunlight, moonlight, stray light, and HEP detections.
Figure 7. Histogram of coverage tallies for 2015.
The second grid, referred to as CF_CVG, tallies the number of CVG pixels that passed the cloud filter (Figure 8). The histogram of the CF-CVG grid (Figure 9) indicates that 99% of grid cells had 20 or more CF_CVGs and only 0.003% had zero CF_CVGs. This indicates that outside of a few low CVG areas, the vast majority of areas have sufficient numbers of cloud-free observations to generate a high-quality night-time lights product.VIIRS night-time lightsAll authors
Christopher D Elvidge, Kimberly Baugh, Mikhail Zhizhin, Feng Chi Hsu & Tilottama Ghosh
Figure 8. Grid tallying the number of cloud-free coverages (CF_CVG) for 2015. The tally includes the CVG pixels minus those pixels deemed to be cloudy by the VIIRS cloud mask.
Published online:
Figure 9. Histogram of the cloud-free coverages for 2015.
Visual inspection indicates that the overall background rises as latitude increases. This was also the case for the DMSP night-time lights. Figure 10 shows the average background radiance as a function of latitude. The average background varies from −0.05 to 1.075 nW, a range of approximately 1 nW. The background is lowest between 50°S and 50°N, with average radiances under 0.2 nW. The average background radiance drops slightly below zero between 10°S and 50°S latitude. This is attributed to the use of northern hemisphere dark night ocean data in establishing the dark current level for night-time DNB calibration. The background then rises in two stages from 51° to 65°S, probably due to the influence of the aurora Australis. The background rises rapidly from 50°N to 70°N. Then drops at about the same rate out to 75°N. The rise and fall of the background north of 50° is believed to be a reflection of auroral activity.VIIRS night-time lightsAll authors
Christopher D Elvidge, Kimberly Baugh, Mikhail Zhizhin, Feng Chi Hsu & Tilottama Ghosh
Figure 10. Average background radiance as a function of latitude for 2015.
Three varieties of annual VIIRS DNB cloud-free composites were produced for 2015. Examples of the three are shown in Figures 11(a-c). The first is the RCFC, which retains biomass burning, aurora, and background. The second is the ORCFC, with biomass burning and portions of the aurora filtered out. The third composite, night-time lights (VNL), is generated by removing background from the ORCFC. The data for the three are floating point values and the units are average radiances in nW cm−2 sr−1.VIIRS night-time lightsAll authors
Christopher D Elvidge, Kimberly Baugh, Mikhail Zhizhin, Feng Chi Hsu & Tilottama Ghosh
Figure 11.
Figure 12 shows a log-scaled histogram of the distribution of radiance present in the VIIRS night-time lights. The histogram peaks at 1.4 nW with more than 1.3 million grid cells per bin. The numbers fall off rapidly as radiance increases and at the extreme high end, the numbers of grid cells per bin drop to single digits. Above 13,000 nW, the number of grid cells per bin is under 10, and above 22,000 nW, there are one or two grid cells per bin. Las Vegas, Nevada peaks out at 2800 nW, leaving 1475 grid cells brighter than Las Vegas. It is believed that the majority of these are gas flares. Below 1.4 nW, the number of grid cells per bin also drops precipitously to 14,000 grid cells at 1 nW. This decline is likely an expression of the VIIRS DNB detection limits.VIIRS night-time lightsAll authors
Christopher D Elvidge, Kimberly Baugh, Mikhail Zhizhin, Feng Chi Hsu & Tilottama Ghosh
Figure 12. Histogram of the 2015 VIIRS night-time lights. There are vastly larger numbers of low radiance grid cells versus high radiance cells.
We wanted to determine whether population centres were being removed from the VNL product as a result of outlier and background removal. Our procedures intentionally seek to remove biomass burning. The glow that surrounds urban centres. To investigate the effectiveness of the outlier and background removal, we extracted Landscan 2015 (Bright, Rose, and Urban 2016) population counts for three varieties of background. Australia was selected for the analysis because it has annual biomass burning and vast areas having scant population.
The first background set consists of the grid cells where the average of high end outliers was 5 nW or greater. These are grid cells with biomass burning activity. The remaining background grid cells were divided into two classes using a radiance threshold of 0.1 nW. Those background grid cells with radiance 0.1 nW or higher are concentrated around lighting features, indicating that these are associated with atmospheric scatter, commonly referred to as glow. The third background class is the largest in terms of grid cell numbers, representing normal background largely free of biomass burning and glow.
Figure 13 shows the results from the Landscan analysis, with Landscan population count on the x axis and grid cell count on the y axis. Note that a log scale is used on the y axis to accommodate the wide range of values. The large data cloud is from the VNL product, showing population counts where lighting was detected. The curves on the left are the fit lines for the three varieties of background: outlier removed, glow, and normal background. The triangle marks the grid cell counts for normal background having population count of zero. This group represents 98.7% of the normal background. The outlier removed grid cells had the lowest population counts of the three varieties of background. The glow has only a slight increase in population count when compared to the normal background. These results indicate that the outlier removal process is not removing population centres from the night-time lights product. Similarly, the glow assigned to background falls outside of population centres and is nearly identical to the normal background lacking glow. Our conclusion is that the outlier, glow, and background removal are working as intended.VIIRS night-time lightsAll authors
Christopher D Elvidge, Kimberly Baugh, Mikhail Zhizhin, Feng Chi Hsu & Tilottama Ghosh
https://doi.org/10.1080/01431161.2017.1342050
Published online:
26 June 2017
Figure 13. Comparison of Landscan population counts from four sets of grid cells in Australia. The data points are from the VIIRS night-time lights. The three trend lines show results from three varieties of background. ORM refers to grid cells where the average of high end outliers was 5 nW or more. GLOW are background grid cells where the average radiance was 0.1 nW and higher, BKG is the remainder of the background grid cells after removal of ORM and GLOW.
Featured Products
