4. Results
4.1. Visual assessment
We applied the VZA-COLD algorithm to all the selected tiles (Fig. 1) to examine its change detection performances for various kinds of human-related NTL changes over different regions. The annual and day-of-year NTL change maps from 2013 to 2021 were created for every pixel. To demonstrate the algorithm’s abilities in monitoring NTL changes, we investigated a range of urban and peri-urban regions with human activity changes corresponding to different land use, demographic, and socioeconomic typologies (Stokes and Seto, 2019). The results, shown in Figs. 9-16, illustrated how the algorithm can accurately capture NTL changes caused by the major types of both short-term and long-term transitions. These factors include, but are not limited to, urbanization processes in sub-urban areas, non-residential constructions of public facilities, land cultivation of a new modern agriculture field, redevelopment of a pre-existing urban area triggered by the economic growth, de-electrification derived by the renovation of lighting technologies and environmental policies, armed conflicts, and power grid loss caused by natural hazards. Meanwhile, the identified changed areas covered a wide range of human footprints with different land cover and land use types, including the highly populated urban areas, urban green space, suburban and rural areas, agricultural fields, roads, and barren land regions with human activities.
Urbanization with constructions of residential and non-residential developments is one of the most prevalent human-driven land cover and land use changes. Fig. 9 showed the urban expansion process of a new residential community in the suburban area of Melbourne, Australia that converted the agriculture fields to impervious surfaces. A gradual NTL change was captured by the algorithm at this site, which is consistent with the built-up period of this new settlement from 2017 to 2021 (Fig. 9c). Figs. 10-11 showed the construction actions of a newly built international airport in the suburban areas of Beijing, China (Fig. 10) and the Olympic Parks in Rio de Janeiro, Brazil (Fig. 11). Multiple NTL changes were identified for the new international airport, which aligned well with the different construction stages (Fig. 10c). The timing of these three identified changes agreed with the start of land clearance in February 2016, the time when the major construction of the airport was finished in 2018, and its opening date in September 2019. At the Olympic Parks in Rio, an increase in artificial light emissions was caused by the new facilities and the large gathering event of the Olympic Games in summer 2016, and this abrupt NTL change was also successfully captured (Fig. 11). In addition to urban developments, the land cultivation engendered by the food consumption in agricultural land can also be captured by the DNB time series. Fig. 12 showed a new modern organic vegetable greenhouse built with LED plant light at night in the low light area of Canada. According to the time series result of the selected pixel (Fig. 12c), a dramatic increase of the NTL was captured in 2017 after the greenhouse was put into use. Redevelopment driven by the population and economic growth, and de-electrification caused by new technologies and environmental policies can shift intensities of the artificial NTL over pre-existing developments without land conversions. Fig. 13 showed the redevelopment of urban areas in Abidjan, Ivory Coast. Foreign investments promoted both GDP and population density of the urban environment of Abidjan (Ramiaramanana et al., 2021), which was detected in 2014. The large-scale renovation of LED streetlights in the suburban areas of Milan that were planned by environmental policies encouraged by the International Registered Exhibition in 2015 (World EXPO) can also be detected from the annual NTL change map (Fig. 14). A significant drop in the NTL radiances was captured in 2014 after the new energy-saving LED streetlights with less upward emission were installed.
Changes in human behavior at night, armed conflict, and power grid loss can lead to short-term NTL shifts. VZA-COLD successfully detected these short-term changes in a timely manner. In urban areas with high dynamics of NTL caused by armed conflicts such as the Syrian Civil War in Raqqa, the algorithm identified multiple NTL changes between the stable periods with relatively short durations (Fig. 15). In September 2017, Puerto Rico was hit by two powerful hurricanes, and the abrupt power outage and gradual restoration were successfully identified by the VZA-COLD algorithm (Fig. 16).