5.1. Performance of Built-Up Environment Properties
Several limitations should be identified before discussing the performance of built-up environment properties. The first limitation is the time differences between GlobeLand30 land-cover data, GLAS waveform data, and DMSP/OLS radiance-calibrated NTL data. GLAS waveform data and NTL data were all acquired in 2006, while GlobeLand30 land-cover data were obtained in 2010. Due to urbanization, there are some changes in land cover between 2006 and 2010. However, the building areas in 2006 mostly remained the same as the building areas in 2010, so the accuracy of built-up environment properties would not be obviously affected. Another limitation is the small error caused by the trees around buildings. The trees within the GLAS footprint would affect LiDAR echo waveform, therefore they would affect the accuracy of building height slightly.
Building height showed a moderate and positive correlation to NTL data (R2
= 0.48); i.e., as building height increased, NTL value increased (Figure 4
a). Higher building height values correspond to more floors; therefore, they would contain more lights, which would lead to higher NTL value. These results were consistent with Kocifaj et al. [38
], and they found total lumen output-normalized radiant intensity is depicted as a function of emission zenith angle for a set of building height values. For tall buildings, the light emissions directed upwards exceed the emissions to low elevation angles. Therefore, increasing building height results in an increase of NTL. However, building height also shapes light output pattern, meaning that the relative contribution of emissions to different angles changes as building height increases. This information is difficult to obtain from DMSP/OLS NTL data, so we cannot analyze the phenomenon clearly. Figure 4
a showed that several points are below the regression line. This can be explained as although building height is high, there is only a small region within the NTL data pixel that is covered by buildings, so the number of lights is small, which leads to low NTL value. There are also several points obviously higher than the regression line in Figure 4
a. This may be caused by large building coverage in these NTL data pixels.
Building coverage did not exhibit strong correlations with NTL data. Figure 4
b showed a weak positive correlation between building coverage and NTL (R2
= 0.16). Building coverage is the ratio of building area and total area within the NTL pixel, which demonstrates the coverage of buildings. Higher building coverage values mean that NTL pixel was more covered by buildings. According to Kocifaj et al. [38
], the low correlation may be due to missing information on the light emissions to different zenith angle. Several points were obviously above the regression line, which might be caused by very high building heights. Figure 4
b showed that although some points had high building coverage values, the NTL values were still very small, which might be caused by low building heights.
Vegetation fraction had a weak negative correlation with NTL, with R2
of 0.10 (Figure 4
c). Vegetation fraction is the ratio of vegetation area and total area within the NTL pixel, which demonstrates the coverage of vegetation. Higher vegetation fraction values mean that NTL pixel was more covered by vegetation, which would affect the intensity of nighttime light. These points, which obviously deviate from the regression line, may also be interpreted as very tall buildings. According to Kocifaj et al. [39
], vegetation can reflect light from built-up elements, which implies flux emitted upward. Due to the chlorophyll concentration and vegetation nutrient change with season, the light-reflecting ability of vegetation also changes with season, which would cause seasonal NTL. Therefore, the correlation between NTL and vegetation fraction may show seasonal behavior. However, due to limited data sources, we cannot analyze the seasonal variation of the correlation between NTL and vegetation fraction in this study, which will need to be studied in the future.
Surface-area index correlated with NTL data strongly and positively, with the R2
value of 0.64; i.e., as surface-area index increased, NTL value increased (Figure 4
d). Surface-area index is a metric that corresponds to building surface area. Higher surface-area index values represent larger building surface area, so the light exposed outside would be more, which would result in higher NTL value. The correlation between NTL and surface-area index is stronger than the relationships between NTL and building height and between NTL and building coverage. This can be explained as the surface-area index containing both building height properties and the number of building pixels, which is related to building coverage properties, so it can describe human settlements more accurately, which would obviously affect the nighttime light intensity. Therefore, surface-area index has a better explanatory power for NTL value than other built-up environment properties.
5.2. Effects of Building Heights
Due to the differences in architectural styles and economic conditions, the building heights across China are diverse. According to the code for design of civil buildings in China, buildings below four floors are low-rise buildings, buildings between four floors to six floors are middle-level buildings, and buildings above six floors are high-rise buildings. The residential building module coordination standard of China indicated that single-floor height is about 2.8 m. Therefore, in this study, the maximum heights of the buildings within GLAS footprint smaller than 8.4 m are classified as low-rise buildings, the maximum height of the buildings between 8.4 m and 16.8 m are classified as middle-level buildings, and the maximum height of the buildings larger than 16.8 m are classified as high-rise buildings.
To explore whether building height would affect the relationship between radiance-calibrated NTL data and surface-area index, we conducted simple linear regressions between radiance-calibrated NTL data and surface-area index at different building heights, and the results are shown in Figure 5
. Figure 5
a shows that radiance-calibrated NTL data of low-rise buildings has a moderate positive correlation with surface-area index, with the R2
value of 0.36. Figure 5
b shows that the correlation between radiance-calibrated NTL data and surface-area index of middle-level buildings was positive and higher, with the R2
value of 0.56. Radiance-calibrated NTL data and surface-area index of high-rise buildings is also positive, and the R2
value was 0.66. Therefore, the correlation between radiance-calibrated NTL data and surface-area index of high-rise buildings is the highest, followed by middle-level buildings, and the correlation of low-rise buildings is the lowest. It can be concluded that building height affects NTL data, and the higher the building height, the stronger the correlation between NTL data and surface-area index. This can be explained by the fact that in some building areas with lower building heights, the trees around buildings may be higher than the buildings, which would lead to higher surface-area index estimation values than true values. In addition, due to the limitation of the statistical method used in Section 3.4
, there are still some lights from non-building areas that cannot be removed when the values of non-building lights are very high.
5.3. Effects of Regional Economic Development Level
To investigate whether regional economic development level would affect the relationship between radiance-calibrated NTL data and surface-area index, we conducted linear regressions between NTL data and surface-area index of the three economic zones in China, and results are shown in Figure 6
. Figure 6
a showed that radiance-calibrated NTL data of China’s eastern region had a positive correlation with surface-area index, with the R2
value of 0.53. Figure 6
b showed that the correlation between radiance-calibrated NTL data and surface-area index of China’s central region was higher (R2
= 0.71). Radiance-calibrated NTL data and surface-area index of China’s western region was also positive, and the R2
value was 0.70. Therefore, the correlation between radiance-calibrated NTL data and surface-area index of the eastern region is the lowest, and the correlation of the central and western regions are all high. It can be concluded that regional economic development level affects nighttime lights, and when regional economic development level is high, the relationship between NTL data and surface-area index is relatively weak. This may be explained by the fact that China’s eastern region has a relatively developed economy, and the public lighting infrastructure is better. Therefore, there are more lights from non-building areas (such as streets and parks) in the eastern region of China, which cannot be completely removed by the statistical method used in Section 3.4
. However, the central and western regions of China are economically underdeveloped areas. where the influences of non-building areas on NTL data are not obvious, and any such influences could have been removed in Section 3.4