Ground Referencing Night Time Light Imagery—How Critical Is It to Conduct the Measurements at the Same Time the Image Is Acquired?

Highlights

What are the main findings?

• Field measurements of night lights do not always have to be conducted alongside the satellite.
• Dimly lit areas were dark in the SDGSAT-1 image and noisy in the ISS and Haishao-1 images.

What are the implications of the main findings?

• Overpass time is expected to be more critical than the exact date when planning a field measurement campaign.
• Multi-directional measurements are critical to better predict spaceborne measurements of night lights.

Abstract

With the increasing availability of high-resolution (<50 m) spaceborne night time light imagery, it is now becoming more feasible to examine the correspondence between spaceborne and ground-based measurements of night lights. However, so far there have been very few studies that have conducted a ground-based campaign of night time brightness measurements during the overpass of a night light-sensitive satellite. Here we tested whether the correspondence between measurements is higher when ground-based measurements are conducted at the same time as the satellite overpass. We conducted measurements using a LANcube photometer along the same route on two consecutive nights (27–28 August 2025) in Brisbane, Australia, and compared them with an SDGSAT-1 (10–40 m) and Haishao-1 (10 m) images acquired concurrently in the evening and with an early morning ISS photo (8 m) acquired three months earlier. We found the correlation between ground-based and spaceborne measurements was not higher for simultaneous measurements, and the explanatory power of our model predicting night time brightness as measured from space increased when including horizontal and upwards ground-based brightness measurements alongside variables of canopy height, land use and road hierarchy. We confirmed the importance of multidirectional ground measurements and urban structure for understanding night time brightness levels measured from space.

1. Introduction

Remote sensing of night lights is one of the key approaches for monitoring human activity from space and for estimating light pollution [1]. However, the variability in night time lights is not only a function of human activities but also depends on emission directionality [2], viewing geometry [3], time of night [4], season [5], and policies for reducing light pollution [6]. Ground referencing (validating) night time lights as measured from space is a challenge due to mismatches between the spatial resolution of the spaceborne sensor and the field of view ground measurements, differences in their viewing geometry (downwards vs downwards and horizontal and all sky; [2]), differences in the units used by the sensors and in their spectral sensitivities [7], atmospheric scattering [8], limitations in the spatial coverage of the ground measurements (with networks for sky observing photometers being limited spatially; [9,10]), the sensitivity of the sensor to low light levels, and temporal mismatches in the time of measurements (satellite images acquired over a few seconds vs the collection of ground measurements whose acquisition might take minutes to hours if conducted along a pre-determined route).

Studies of remote sensing of night lights that attempted to examine the correspondence between spaceborne measurements and ground-based measurements often did not have the two sets of measurements conducted simultaneously (e.g., [2,11]; Table 1). In most cases where simultaneous measurements were conducted on the ground, they were limited to fixed point measurements using a single photometer or a network of photometers such as Sky Quality Meter (SQM) or Telescope Encoder and Sky Sensor-WiFi (TESS-W) (e.g., [10,12]; Table 1). The only examples we know of where simultaneous measurements of night lights were conducted alongside ground-based sensors over pre-determined routes took place using UAVs [13] or from a high-altitude balloon [14], but very few from spaceborne sensors (Table 1).

In daytime remote sensing, it has been shown that ground measurements of surface reflectance for the purpose of atmospheric correction do not have to be acquired simultaneously at the overpass time of the satellite [15]. In this paper, we aimed to examine whether the correspondence between ground measurements acquired using a mobile LANcube photometer and spaceborne measurements of night lights is higher when they are conducted at the same time or not. A secondary objective of ours here was to examine and compare the quality of three night time spaceborne sensors: SDGSAT-1, HaiShao-1 and astronaut photos acquired during the ISS073 mission, and to better understand the variables explaining night time brightness as measured from space.

Table 1. Selected previous studies that used ground-based sensors to calibrate and validate spaceborne night light imagery.

Spaceborne or Aerial Sensor	Ground-Based Data	Concurrent Acquisition	Spatial Matching of Ground Measurements	Aims of Comparison	Ref.
SDGSAT-1	TESS-4C	Yes	Fixed point measurements	To examine the impact of the transition from HPS to LED	[16]
SDGSAT-1	LANcube	No	Along a short road section		[16]
SDGSAT-1	Five-point light sources	Yes	Fixed point measurements	Vicarious radiometric calibration	[17]
High Altitude Balloon Flights	LANcube	Yes	Along roads in a town in Canada	To better calibrate spaceborne night light measurements	[14]
SDGSAT-1, VIIRS/DNB	DSLR camera with fisheye lens	No	Measurements along the coast	To estimate coastal light pollution	[18]
VIIRS/DNB	SQM, DSLR camera with fisheye lens	No			[19]
SDGSAT-1	LANcube	No	Along roads in Israel and Brisbane	To examine the effect of the directionality of ground measurements on the correspondence with satellite data	[11]
EROS-B	SQM	No	Along trails in parks		[2]
ISS	SQM	Yes	Fixed point measurements	To better understand temporal changes in lighting during the night	[20]
VIIRS/DNB	Modeling of light emission	Yes	Fixed point measurements	To better understand the calibration stability and absolute accuracy of VIIRS data and the impact of the atmosphere	[21]
VIIRS/DNB	SQM	Yes	Mostly fixed-point measurements	To calibrate and generate the atlas of artificial night sky brightness	[9]
VIIRS/DNB	TESS-W, SG-WAS	Yes	Fixed point measurements	Better calibrate and integrate ground and spaceborne measurements	[10]
UAV	SQM	Yes	Fixed point measurements as well as along a route in a playground	To understand the diurnal dynamics of city lights during the night	[13]

Table 1. Selected previous studies that used ground-based sensors to calibrate and validate spaceborne night light imagery.

Spaceborne or Aerial Sensor	Ground-Based Data	Concurrent Acquisition	Spatial Matching of Ground Measurements	Aims of Comparison	Ref.
SDGSAT-1	TESS-4C	Yes	Fixed point measurements	To examine the impact of the transition from HPS to LED	[16]
SDGSAT-1	LANcube	No	Along a short road section		[16]
SDGSAT-1	Five-point light sources	Yes	Fixed point measurements	Vicarious radiometric calibration	[17]
High Altitude Balloon Flights	LANcube	Yes	Along roads in a town in Canada	To better calibrate spaceborne night light measurements	[14]
SDGSAT-1, VIIRS/DNB	DSLR camera with fisheye lens	No	Measurements along the coast	To estimate coastal light pollution	[18]
VIIRS/DNB	SQM, DSLR camera with fisheye lens	No			[19]
SDGSAT-1	LANcube	No	Along roads in Israel and Brisbane	To examine the effect of the directionality of ground measurements on the correspondence with satellite data	[11]
EROS-B	SQM	No	Along trails in parks		[2]
ISS	SQM	Yes	Fixed point measurements	To better understand temporal changes in lighting during the night	[20]
VIIRS/DNB	Modeling of light emission	Yes	Fixed point measurements	To better understand the calibration stability and absolute accuracy of VIIRS data and the impact of the atmosphere	[21]
VIIRS/DNB	SQM	Yes	Mostly fixed-point measurements	To calibrate and generate the atlas of artificial night sky brightness	[9]
VIIRS/DNB	TESS-W, SG-WAS	Yes	Fixed point measurements	Better calibrate and integrate ground and spaceborne measurements	[10]
UAV	SQM	Yes	Fixed point measurements as well as along a route in a playground	To understand the diurnal dynamics of city lights during the night	[13]

Our hypotheses were the following:

• That the correlations between ground-based and spaceborne measurements will not be much higher for the same night measurements than for different night measurements, due to the high variability in space and time of light emissions and the variability in the viewing angles of the spaceborne measurements.
• That the correlations between ground-based and spaceborne measurements will be higher at a finer spatial resolution (≤10 m) than at 40 m resolution.
• That the correlations between ground-based and spaceborne measurements will be higher for the green band than for the other spectral bands because of atmospheric scattering in the blue band and because of the inclusion of the near infra-red in the “red” band of the SDGSAT-1 [16].
• That the correlations between ground-based and spaceborne measurements will be higher when using the sum of lights, including the horizontal measurements of the LANcube, than when only including the upwards measurements of the LANcube, given that the horizontal measurements will include additional light sources that may be partly sensed from space [2].
• That dimly lit areas will be below the threshold of spaceborne sensors [18], and that in densely built areas with high-rise buildings, the ground-based measurements will under-represent the night time brightness levels measured from space, given that street-level measurements will be less representative of light sources originating from high-rise buildings.
• That areas along major roads and in non-residential areas will be associated with higher night time brightness levels [11].

2. Materials and Methods

2.1. Study Area

Our study area was placed in the inner western suburbs of Brisbane, Queensland, Australia, to the west of the Brisbane River, between the university suburb of St Lucia in the south, Kelvin Grove in the north, and Toowong and Indooroopilly in the west, covering both residential streets and major roads.

2.2. Spaceborne Images

We used three sources of night time light images, all acquired on weekdays, as detailed in

Table 2. Satellite images used in this study.

Satellite	Date	Time of Acquisition	Spatial Resolution	Spectral Bands	Comments
Astronaut photo	Wednesday 21 May 2025	02:56 am	8 m	RGB	ISS073-E-120440 NIKKON Z 400 mm f/2.8
SDGSAT-1	Wednesday 27 August 2025	21:12 pm	10 m	panchromatic
SDGSAT-1			40 m	RGB
Haishao-1	Thursday 28 August 2025	20:14 pm	10 m	panchromatic

. Astronaut photos represent the first source of multispectral night time light images from space [1] and have been used for various studies of urban areas [22]. However, they are provided as digital images without georeferencing or radiometric calibration, and their quality and spatial resolution can vary a lot between different ISS missions, depending in part on the camera used [23,24]. The astronaut photo we used here was acquired after midnight on 21 May 2025, three months before our field campaign. Given that Brisbane has a sub-tropical climate, most trees in Brisbane have their leaves on throughout the year, and thus seasonal changes related to leaf-fall and its possible impact on night time brightness as measured from space [25] are negligible or non-existent within this study area. In addition, based on our acquaintance with the measured routes, there were no major changes in land use or in lighting types during those three months.

The SDGSAT-1 satellite was launched in November 2021 and is the first multispectral night time sensor providing free color night time images at a moderate spatial resolution of 40 m and panchromatic images at a spatial resolution of 10 m for the scientific community as well as for humanitarian purposes [26,27]. The SDGSAT-1 satellite acquires its images around 21:30 pm (local time) when people are more active [6], in comparison with the VIIRS/DNB sensor that acquires its images around 01:30 am and opens new research avenues for ecological and urban applications [28,29].

More recently, in December 2024, the Haishao-1 satellite was launched, offering a panchromatic night time lights band centered around 0.7 μm at a spatial resolution of 10 m [30].

We georeferenced all images to the road network of Brisbane using OpenStreetMap (OSM) data with more than 15 ground control points each and used a spline transformation, validating that the images were well aligned with the road network along the roads that were measured on the ground. The SDGSAT-1 bands were calibrated from DN values to at-sensor radiance values using the gain and bias coefficients provided in the metadata of the SDGSAT-1 imagery, following the approach described in [27]. The ISS and Haishao-1 images were analyzed using their DN values with no radiometric calibration. Radiometric calibration of spaceborne sensors is critical for examining changes in radiation or in night time brightness; however, given that calibration functions from DN to at-sensor-radiance are linear [17,27], the conversion from DN to radiance will not affect correlations or the explanatory power of regression models between ground-based and spaceborne measurements.

2.3. Ground Measurements of Night Lights

We coordinated the timing of the image acquisition by the SDGSAT-1 and Haishao-1 for two consecutive nights, 27 and 28 of August 2025, in the early evening hours, when the moon was less than 25% full, near the time of the moonset, and in cloud-free conditions. We used a LANcubeV2 photometer [11,31] to acquire ground measurements of night lights along the same route on both evenings, around the time of the image acquisition by the two satellites (on 27 August 2025 between 20:14 and 21:50 pm, and on 28 August 2025 between 19:39 and 20:57 pm). The LANcube acquired data every 0.6 s with sensors in five directions (S1: upwards, S2: backwards, S3: right, S4: forwards, S5: left) in three bands (R, G, B) and calculated the lux values in each direction for each measurement. Overall, more than 7000 GPS locations were measured.

2.4. Spatial and Statistical Analysis

We aggregated the LANcube point measurements within grid cells of 40 m that were defined so as to fit the 40 m pixels of the SDGSAT-1 RGB bands. We calculated within each grid cell the average lux values in each direction (thus accounting for variable car speed and number of LANcube measurements within each grid cell), as well as the average values for the different night time light sensors (Table 2). Altogether this resulted in a sample size of 1057 grid cells of 40 m. We examined the correlations between the ground-based measurements and spaceborne measurements using the Spearman rank correlation coefficient (r_s) to examine the hypotheses presented above. To examine the overall explanatory power of spaceborne measurements as a function of ground-based measurements, we also included the following datasets: ETH (Eidgenössische Technische Hochschule) Global Canopy Height (2020) product at 10 m resolution, which utilizes deep-learning fusion of Sentinel-2 and GEDI LIDAR data [32], residential and non-residential building height classes extracted from the Global Human Settlement Layer (GHSL) GHS-BUILT-C (2018 epoch) product [33,34], and OpenStreetMap road classes [35]. We merged the road classes of Motorway, Trunk and Primary into a Highway class and the road classes of Secondary, Tertiary and Unclassified into a Major road class. Four stepwise regression models were run each time for one of the spaceborne datasets of night time brightness after a logarithmic transformation of the spaceborne dataset at the grid cell level. The explanatory variables were the sum of lux values as measured by the LANcube in all directions, canopy height, percent area covered by residential buildings, percent area covered by non-residential buildings, percent area covered by buildings higher than 6 m, percent area covered by buildings higher than 15 m, length of highways, length of major roads, and length of residential roads. We assessed the spatial autocorrelation for the residuals of the stepwise models, using Global Moran’s I based on a 4-nearest-neighbor inverse-distance spatial weights matrix (distance decay exponent = 2), using GeoDa 1.22.0.21 spatial data software [36]. As a further robustness test, we reran the stepwise models, including only every 4th grid cell, thus running the models also for a reduced sample size of 265 grid cells.

An additional examination of the quality of the images was performed within four representative dark areas. Three of these four areas were completely unlit: a section of the Brisbane River (58 ha), a section of the Indooroopilly Golf Course (67 ha) and a section of Mount Coot-tha Forest (223 ha). An additional area we examined included dimly lit residential streets in a low-density residential area in the university neighborhood of St Lucia, between Carmody Road in the north and Hawken Drive in the south (20 ha). The most common street light type in that specific area was compact fluorescent (43 out of 62 street lights), followed by LED (14 street lights), with an average lux value (in the upwards direction) of 2.45 lux (based on >1500 LANcube measurements conducted in those streets on 11 September 2025). To quantify the variability of night time light levels as measured from space in those areas, we created a point layer with a uniform distribution every 40 m within those four areas and calculated the average and standard deviation of the DN values of each of the bands of the ISS, Haishao-1 and SDGSAT-1 images. Given that in the SDGSAT-1 image, a DN value of one is the lowest value, whereas in the ISS and Haishao-1 images, zero is the lowest DN value, we subtracted a value of one from the SDGSAT-1 bands for this analysis. To assess differences in night light brightness levels as a function of street light types, we used a GIS layer from Energy Queensland that we intersected with the 40 m grid cells. Each grid cell was associated with the street light type that was most common within it. Grid cells with an equal number of different street lights were excluded from our analysis (6% of all grid cells). We also excluded the grid cells within the university campus (8% of all grid cells) from our analysis, because they were not covered by the street light dataset of Energy Queensland.

3. Results

The three night time satellite images of Brisbane had similar spatial resolution, and were highly correlated, especially between the higher spatial resolution night time images, in the following descending order: SDGSAT-1 at 10 m and the sum of the three ISS bands (r_s = 0.795), SDGSAT-1 at 10 m and Haishao-1 (r_s = 0.729), and Haishao-1 and the sum of the ISS bands (r_s = 0.667). The correlation in overall brightness (radiance) between the multispectral bands of SDGSAT-1 (sum of the three bands after their conversion into radiance at-sensor) that have a spatial resolution of 40 m, was lower, and amounted to about 0.610 with the Haishao-1 and to 0.599 with the sum of spectral bands of the ISS. Some notable differences between the three images for example were in the lighting of sports ovals and stadiums, as can be seen in the St Lucia campus of the University of Queensland (bottom part of all sub-figures of Figure 1).

From the point of view of image quality, the ISS image was the most ‘noisy’ (showing bright pixels in unlit areas, such as the Brisbane River) and the SDGSAT-1 image was the least ‘noisy’ (Figure 2). In addition, the ISS image had a significant NW-SE smear of the pixels, whereas the other images were ‘sharper’. In all three spaceborne platforms, many dimly lit residential streets with street lights appeared to be either unlit or ‘noisy’, and thus the night time brightness in them appeared to be below the detection threshold of these sensors (Figure 2). Amongst the four dark areas we compared, Mt Coot-tha was the darkest, followed by the Indooroopilly Golf Course, the dimly lit residential streets in St Lucia and the Brisbane River having similar values (

Table 3. Average DN values of the different bands of each of the spaceborne platforms, in four dark or dimly lit areas.

Satellite	Resolution (m)	Band	Brisbane River	Golf course	Mt Coot-tha	St Lucia dimly lit streets
ISS photo	8	Red	3.210	2.409	2.114	3.234
		Green	3.204	2.421	2.117	3.234
		Blue	3.102	2.414	2.115	3.081
Haishao-1	10	Pan	0.769	0.702	0.474	0.879
SDGSAT-1	10	Pan	0.043	0.005	0.007	0.129
	40	Red	11.390	0.428	0.024	14.758
		Green	22.263	2.560	0.251	15.935
		Blue	8.632	1.371	0.328	1.742

). The panchromatic band of the ISS photo presented the darkest values for all four areas (in DN values), followed by the Haishao-1 (Table 3). Based on the LANcube measurements in the upwards direction, streets led by LED (n = 229 grid cells) were the most brightly lit, followed by high-pressure sodium (HPS, n = 323 grid cells) and compact fluorescent (CFL, n = 110 grid cells) (Figure 3a). Unlit grid cells (i.e., grid cells with no street lights within them; n = 236) were not dark, and their brightness levels were mostly affected by adjacent light sources. However, in all three spaceborne images, the difference in night time brightness between LED and HPS was smaller than in the ground-based measurements (Figure 3b–d).

As for multispectral information, in the ISS photo analyzed in this study, the correlations between the three color bands were very high (r_s > 0.985 for all band combinations), indicating little multispectral information, whereas the correlations between the three color bands of the SDGSAT-1 were lower (r_s ranging between 0.91 and 0.97 for the different band combinations) indicating better color information.

The correlations between the two consecutive ground measurements of the LANcube were much higher than between the two consecutive measurements of the spaceborne sensors. The correlation between the S1 (upwards) lux measurements of the two nights was 0.927, and was even higher when summing up the lux values in three directions (S1 + S3 + S5, upwards, right and left), being 0.950 between the two nights. When examined on a map, the consistency of the ground-based measurements in these directions can be well appreciated (Figure 4). When summing the lux values in three directions, the measurements were less ‘noisy’ than when examining the values in the upwards direction, with a lower coefficient of variation (standard deviation divided by the average) in both nights (being 1.16 when summing the three measurements directions, vs 1.27 in the upwards direction alone).

The correlations between the ground-based lux measurements and the spaceborne measurements were highest with the ISS photo (which was acquired three months before the ground-based LANcube measurements), second highest for the SDGSAT-1 at 10 m, followed by the Haishao-1 and last by the SDGSAT-1 multispectral bands at 40 m (Table 3). Across all comparisons, the correlations were higher between the spaceborne measurements and the ground-based measurements when we summed the lux measurements of the LANcube in all five directions, than in the three directions (upwards, right and left), and than when examining the correlations only with the upwards (S1) measurements (

Table 4. Spearman’s rank correlation coefficient between the spaceborne images and the ground-based LANcube lux measurements. The five sensors of the LANcube analyzed here are the S1 (upwards), S2 (backwards), S3 (right), S4 (forward) and S5 (left). All correlations were statistically significant (p < 0.001).

LANcube		S1	S1 + S3 + S5	S1 + S2 + S3 + S4 + S5	S1	S1 + S3 + S5	S1 + S2 + S3 + S4 + S5
Imagery date		27 August 2025			28 August 2025
ISS photo 8 m	21 May 2025	0.654	0.742	0.773	0.640	0.726	0.768
Haishao-1 10 m	28 August 2025	0.496	0.592	0.619	0.499	0.616	0.638
SDGSAT-1 10 m	27 August 2025	0.584	0.689	0.722	0.574	0.673	0.713
SDGSAT-1 40 m		0.451	0.540	0.563	0.452	0.541	0.575

).

Summing the brightness measurements in all directions thus incorporates the different light sources, with the upwards direction representing street lights, the left and right directions representing artificial light from residential and commercial sources, and the back and front directions representing artificial light from passing cars. In both dates (27–28 August 2025), the average contribution of upward measurements amounted to 24% of the sum lux values in all directions, whereas when adding the left and right directions, the average contribution amounted to 46% of the sum lux values in all directions (Figure 5). Thus, almost half of the sum lux values was contributed from measurements in the forward and backward direction, which may be partly attributed to light from passing cars (traffic). In more brightly lit roads this contribution (of light coming from the front or back of our photometer) decreased.

In both the Haishao-1 and the SDGSAT-1 images, there were grid cells in which both sensors recorded no radiance, mostly up to sum lux values of 10 lux (Figure 6). These locations were mostly associated with roads lit by CFL (Figure 7). CFL lights have very specific emission peaks, some of them in the short blue, below the spectral sensitivity of both SDGSAT-1 and Haishao-1 (Figure 8). As for the same-day vs. one-day-apart comparisons, the correlations for both the Haishao-1 and the SDGSAT-1 were only slightly higher when compared with the ground measurements taken on the same day than with the ground measurements taken on another day; these differences were quite small (Table 4). Band-wise, the highest correlations were found for the green band across the three multispectral sensors, and the lowest correlations were found for the red band (

Table 5. Spearman rank correlation coefficients between the three multispectral sensors, the LANcube, the ISS photo and the SDGSAT-1 image, within each of the three spectral bands. All correlations were statistically significant (p < 0.001). Note that the ‘red’ band of SDGSAT-1 includes the NIR as well.

	Red	Green	Blue
ISS and SDGSAT-1	0.581	0.612	0.586
ISS and LANcube	0.610	0.670	0.665
SDGSAT and LANcube	0.418	0.478	0.472

).

The multivariate stepwise regression model with the highest explanatory power was for the ISS night time brightness values (Adj R² of 0.643), followed by that of the SDGSAT-1 image at 10 m (Adj R² of 0.525) (

Table 6. Multivariate linear stepwise regression models, explaining the spaceborne borne night time brightness (after a logarithmic transformation), as a function of the sum LANcube lux values (sum of all directions), and additional variables representing canopy height, GHSL morphological settlement zones (residential and non-residential classes, and building height classes), and OpenStreetMap road classes. The table shows the standardized coefficients of each of the variables that were found to be statistically significant. The Global Moran’s I was calculated on the residuals of the regression model, and its pseudo p-value was calculated using 999 permutations. In addition to the full model (all grid cells, n = 1057), we show the model results for a robustness test, including only every 4th grid cell (n = 265).

Response Variable	ISS		Haishao-1		SDGSAT-1 10 m		SDGSAT-1 40 m
Model for n = 1057
Adjusted R2	0.643		0.370		0.525		0.295
Global Moran’s I	0.317		0.394		0.487		0.501
Pseudo p-value	0.001		0.001		0.001		0.001
Model for n = 265
Adjusted R2	0.605		0.324		0.477		0.273
Explanatory variable for n = 1057	t	Pr > |t|	t	Pr > |t|	t	Pr > |t|	t	Pr > |t|
LANcube Sum lux	17.109	<0.0001	11.636	<0.0001	13.446	<0.0001	7.642	<0.0001
Canopy height	−9.183	<0.0001	−2.354	0.019	−6.673	<0.0001	−2.535	0.011
GHSL residential	−2.147	0.032
GHSL non-residential	2.927	0.003	3.215	0.001	4.387	<0.0001	5.974	<0.0001
GHSL > 6 m
GHSL > 15 m
OSM Highways	14.492	<0.0001	6.353	< 0.0001	11.253	<0.0001	7.081	<0.0001
OSM Major roads	10.334	<0.0001	5.576	<0.0001	6.082	<0.0001	2.652	0.008
OSM Residential	−4.540	<0.0001	−5.308	<0.0001	−5.487	<0.0001	−3.187	0.001

). The residuals of the models presented moderate to substantial positive spatial autocorrelation among neighboring grid cells, with Global Moran’s I values ranging between 0.317 (for the ISS) and 0.501 (for the SDGSAT-1 multispectral brightness). Hence the statistical significance of our models may be somewhat lower than the reported p-values. Rerunning the stepwise models including only every 4th grid cell (n = 265), all models were still highly statistically significant, and the adjusted R² values only decreased slightly (Table 6).

Six of the explanatory variables were found to be statistically significant in all four stepwise models (including all grid cells, n = 1057). The variable of the sum of lux values as measured by the LANcube had the highest standardized coefficient in all models, followed by the length of the highways (canopy height was moderately correlated with the length of highways, r_s = −0.345). Two variables had negative standardized coefficients in all models: canopy height and the length of residential roads (these two variables were not correlated with each other, r_s = 0.097). In the model of the ISS night time brightness, an additional explanatory variable was included in the model, also with a negative standardized coefficient: the percent area of residential buildings within a grid cell, whereas the percent area of non-residential buildings always had a positive coefficient in all four models. The height of buildings was not found as a significant variable in any of our models.

4. Discussion

This study was the first to conduct multidirectional ground measurements along multiple road sections concurrently with the overpass of two high-resolution spaceborne night light sensors with an early evening overpass—SDGSAT-1 and the recently launched Haishao-1. We found that conducting same date measurements of night lights at the ground level and from above were not critical for comparing the two perspectives. This may at first seem a surprising result, given that night lights are highly volatile and temporally dynamic [39]. While the turning on and off of lights within residential areas is an individual-level decision, during the early evening hours (before 10 pm) most people are still awake and have some indoor lights that can spill outside via windows, and the turning off of lights becomes more noticeable after midnight [4,40]. In a study examining the dynamics of the urban nightscape, it was actually found that the turning-off time of lights is quite consistent between nights (at least for residential sources) [4]. Moreover, in residential areas (in contrast with commercial or industrial areas), street lights can be a major source of artificial lights [41], having very consistent lighting dynamics, and in most countries and cities are not dimmed or turned off during the night (as is also the case in Brisbane). Therefore, it is not surprising that ground measurements of night lights can be well correlated with spaceborne measurements conducted on other nights. Our finding that the correlation between ground-based measurements of night lights over two nights was much higher than the correlation between spaceborne measurements of night lights over the same two nights also reinforces our conclusion and the relative stability of artificial light sources during the early evening, compared with the variability in spaceborne measurements of night lights, which are also affected by variable atmospheric conditions [21], viewing geometries [3], and georeferencing errors [42]. Within our case study of Brisbane, the highest correlation between ground-based and spaceborne imagery was actually found for the ISS photo, acquired after midnight, three months before the field campaign that was conducted in the early evening hours. This result however, may also be affected by other factors, such as road orientation and the different viewing geometries of each of the three spaceborne sensors and the slightly higher spatial resolution of the ISS image that we used. However, to account for anisotropic effects as observed by night time light sensors at moderate spatial resolutions (between 5 and 40 m), new approaches should be developed, following previous studies that examined this for the VIIRS as well as for drone imagery [3,43,44]. An additional factor that could partly explain the higher correlation that was obtained by the ISS image for the correlation between the ground-based and spaceborne measurements is related to the spectral sensitivity of the different sensors. Both the SDGSAT-1 and Haishao-1 cover a wide spectral range that includes the near infra-red (NIR) and do not cover the short blue, whereas the LANcube and digital cameras used on the ISS have only red, green and blue bands that do not include the NIR (Figure 8).

While some of the light sources were obviously turned off after midnight (e.g., in some sports facilities within the University of Queensland St Lucia campus), most of our study area was composed of residential buildings: 63% of all grid cells had more than 25% of residential areas, whereas only 4% of the grid cells had more than 25% of non-residential areas. We therefore expect that when conducting a similar study in a major commercial urban area, or in an area where lights are dimmed or turned off after midnight (as in rural France; [6]), coordinating the time at night of the ground-based and spaceborne measurements will be more critical than in our case study of Brisbane. Central commercial areas and activity hubs in Los Angeles were found to exhibit high fluctuations in night time levels between different nights compared with residential areas [45], hence when comparing ground-based and spaceborne measurements acquired on different nights, we recommend preferring land uses with less fluctuations. Unless street lights are intentionally dimmed or turned off, they tend to be temporally stable during the night in comparison with lights from windows [46], hence roads where street lights represent a major source of artificial light may be suggested as potential sites for calibrating between ground-based and spaceborne measurements of artificial lights. The time can be coordinated within general time frames of early evening, late evening, after midnight, and before dawn, depending on the local dynamics of night lights turning on and off. We expect that in brightly lit urban and suburban areas, the moonlight will have a negligible effect on the correspondence between ground-based and spaceborne measurements, given that the full moon has a brightness of less than 0.3 lux [47].

In this study we also provide a first-of-a-kind comparison of the quality of three high-res night time light sensors with a spatial resolution of at least 10 m, covering the same area, ISS photo, SDGSAT-1 and the new Haishao-1 satellite [30] that was only recently launched. We found that all three sensors were not sensitive enough in dimly lit and dark areas—with the SDGSAT-1 being almost completely dark in such areas, corresponding with previous studies [18], and the ISS and Haishao-1 images being mostly ‘noisy’ in such dimly lit or dark areas. Previous studies also confirmed that VIIRS/DNB has a better and lower light detection limit [26]. Future night time lights missions should therefore be designed with lower light detection limits and higher signal-to-noise ratios [48]. The quality of ISS photos varies widely between missions, depending also on the camera and the settings used (e.g., F/stop, ISO, lens, etc.) [24]. Recent night time ISS photos from mission 073 have high spatial resolution (<10 m) but seem to lack in their multispectral quality, with less differentiation between the different bands. While the three spaceborne sensors were not sensitive enough to measure artificial lights originating from residential streets lit by CFL, the ground-based LANcube photometer was able to fill this gap, and to provide multidirectional night time brightness measurements that enabled us to better explain the spaceborne night time brightness measurements. Although the LANcube photometer is not as sensitive as the SQM photometer for real dark sky areas [6], it is well suitable for measurements in peri-urban and urban settings where artificial lights are present. One of the limitations in comparing the spectral bands of the LANcube, ISS and SDGSAT-1, is the different widths of the spectral bands on each of these sensors, especially with the SDGSAT-1 ‘red’ band actually covering both the red and infra-red regions [16] and the Haishao-1 panchromatic band also covering the infra-red region and not just the visible range (Figure 8). It is therefore recommended for future night time light missions to have a dedicated truly red band (to better match human vision and digital cameras), and to have a separate band for measuring the near-infra-red band, which is important for detecting emissions from HPS lighting, for example. Our measurements were mostly conducted within residential areas and along some major roads, and we did not cover many non-residential areas where commercial activities are present. Commercial areas are characterized by multiple light sources (such as window facades, decoration lighting, flood lighting, billboards, etc.), different from residential areas [11]. Non-residential areas were found to increase night brightness as measured from space, whereas higher canopy cover was found to decrease it. Whereas previous studies identified that building height had a significant impact on night time brightness as observed by VIIRS/DNB from different viewing zenith angles [49], and that in areas of higher buildings there was a greater decrease in night time brightness after midnight (comparing ISS and SDGSAT-1 images of Shanghai acquired at different overpass times) [50], building height did not enter our regression models. Our driving transect did not cover the central business district of Brisbane, and most of our study area was in areas of relatively low buildings, with only 22% of our grid cells having more than 25% of their area covered by buildings higher than 15 m. When conducting field campaigns for the comparison between ground-based measurements and spaceborne measurements of night time brightness, we therefore recommend focusing on areas where light sources are less variable and are less affected by obstruction from trees or buildings [44]—such as highways and newly built low-density residential areas.

We found that explaining spaceborne measurements of night time brightness improves when summing the ground-based measurements in all directions. While the contribution of light from passing cars to the measured light on the ground may seem to be an unwanted ‘noise’, we perceive this to be part of the signal, which may also affect the spaceborne measurements. Indeed, traffic has randomness; when averaged over many samples, we expect the stochastic variability to reduce to less, affecting the correspondence between ground-based and spaceborne measurements, improving the representativeness of the ground measurements [51].

5. Conclusions

Planning and conducting field campaigns alongside the overpass of a satellite is logistically challenging, even more so at night time. In this study we demonstrated that ground-based measurements of night lights can be highly correlated with high-resolution spaceborne measurements of night time lights even if they were not conducted at the same time. This finding has important implications for future studies aiming for better calibration and validation of light pollution estimates from space. While high correspondence may be achieved between measurements conducted on separate nights, we recommend that the ground measurements be conducted at a similar phase of the night (e.g., early evening, after midnight) to the overpass of the satellite. Researchers should also verify that no transition of lighting types (e.g., between HPS and LED) took place within the study area between the two acquisition dates and that no major land use changes occurred between the two dates for the comparison to be more reliable between ground-based and spaceborne measurements. We also recommend conducting the field campaign in areas where light sources are expected to be more stable between different nights (e.g., streets and major roads in residential areas) and prefer areas where there will be fewer obstructions from high-rise buildings or high trees for the comparison between ground-based and spaceborne measurements to be more consistent and less affected by satellite viewing angle differences. We call for future studies to continue exploring these avenues and to also expand the comparison between simultaneous ground-based and spaceborne measurements of artificial lights across multiple cities, with different land use classes, urban lighting regulations, architectural styles and vegetation types so that protocols for ground-based night time measurements will be developed and agreed upon.