1. Introduction
One recent trend in optical remote sensing is to increase observation frequencies to meet the urgent need for effectively monitoring of ephemeral events or phenomena on Earth from space. For example, Sentinel 2 can revisit the Earth every 10 days at the equator with one satellite, but 5 days with 2 satellites under cloud-free conditions, which results in 2–3 days revisiting time at mid-latitudes [
1]. MODIS (Moderate Resolution Imaging Spectroradiometer) sensors can visit the entire Earth surface twice a day through the constellation of Terra and Aqua, and commercial small or nano-satellite constellations, such as Jilin-1 [
2] and Planet Labs’ Dove [
3], enable even higher observation frequencies of up to hours. Increasing the number of satellites can surely help to increase the temporal resolution of remote sensing observations, but there are still challenges at the night side, when there is no sunlight available to illuminate the Earth surface. Such a situation can be even worse in the polar regions, where sunlight is not available for almost half the year [
4].
Diurnality is a common ephemeral phenomenon, frequently observed in many animals, plants, as well as some natural processes. For example, many animals’ daily activities strictly depend on sunlight, and they are active during the day and sleeping at night or vice versa. Natural plants use photosynthesis to convert light energy into chemical energy during the day and uptake and transport of water through the process of transpiration at night. The ocean temperature often shows differences during the day and night cycles, mainly caused by the influences of the sun [
5,
6,
7,
8]. Diurnality (circadian rhythms for organisms) are mainly caused by day and night cycles, driven by Earth revolving both around the sun and on its own axis. Lunar rhythms are also embedded in the life cycles of many organisms. Fluctuating light levels reflected by the Moon also have a startling impact on life on Earth. For example, some animals prefer to live by the light of the Moon. For many animals, particularly birds, the Moon is essential to migration and navigation [
9,
10]. In addition, although coral reproduction is affected by weather, water temperature and other factors, it is also found that most corals choose to spawn during or near a full moon [
11].
Human beings used to be a member of the diurnal club. During our early history, human activities at night were also greatly confined until they learned how to use fire to light up their living spaces. Artificial illumination thus has been allowing human beings to break through its natural diurnality so their social, cultural, and economic activities can extend into night, creating cities and a night-time economy [
12]. Nowadays, lighting is essential for human beings, first for convenience and for safety. Artificial nightlight is thus an important and reliable indicator of human activities, directly at night and indirectly during daytime. The artificial illumination of buildings, transportation corridors, parking lots, and other elements of the built environment have become a hallmark of many contemporary urban settlements and urban activities [
13].
With the advent of low-light detecting technologies, nightlight remote sensing makes it possible to detect artificial lights from space, forming a convenient and powerful tool to characterize and understand human being’s altered diurnality. Since the first night-time light scene was captured using the operational linescan system (OLS) aboard Defense Meteorological Satellite Program (DMSP) satellites, detecting artificial light has become the main staple of nightlight remote sensing. Since then, whether it is mono-spectral (Visible Infrared Imaging Radiometer Suite/Day-Night-Band (VIIRS/DNB), Scientific Application Satellite-C High Sensitivity Technological Camera (SAC-C HSTC), Scientific Application Satellite-D High Sensitivity Camera (SAC-D HSC), CubeSat Multispectral Observing System (CUMULOS), Luojia1-01 (LJ 1-01), Earth Remote Observation System-B (EROS-B) or multi-spectral (Aerocube 4, International Space Station (ISS), Aerocube 5, Landsat-8, Jilin-1), new sensors have been developed with the main focus of detecting and identifying self-luminous objects at night [
14].
Applications using these night-time remote sensing data include mapping urban areas [
15,
16,
17,
18,
19], estimating population, GDP, and poverty [
20,
21,
22,
23], monitoring disasters and conflicts [
24,
25,
26], as well as understanding the influence of light pollution [
10,
27,
28,
29]. These studies often focus on urban lights or self-luminous bodies, such as fisheries, oil, gas extraction, etc., with very few exploring the potential of nightlight remote sensing data to study natural processes.
Ironically, it seems to have long been ignored that the original purpose of DMSP/OLS was to detect clouds under moonlight illumination [
30]. Although moonlight has been an important factor that affect many nocturnal animals and plants, mainstream nightlight remote sensing image processes either try their best to totally avoid moonlight, or to remove the moonlight component from observations with tuned algorithms. For example, to generate annual DMSP/OLS composites, only sunlit and moonlight-free observations have been used, and moonlit observations are simply discarded [
31]. To produce the VIIRS/DNB daily black marble product, an algorithm was developed to remove moonlight components from daily nightlight observations [
32].
With the recent development of low-light detection technology, the nightlight remote sensing community started to realize that moonlight can be a very useful illumination source for detecting weather and climate parameters at night, instead of being treated as a noise source for city light detection [
33,
34]. Miller et al. [
4] made a detailed insight of many potential applications for nocturnal low-light visible satellite observations and presented a long list of key variables that could be obtained under moonlight using VIIRS/DNB from space. They found that sometimes night-time moonlight remote sensing even showed advantages over the daytime sunlight remote sensing. These include the detection of snow cover, rainfall distributions across arid/semi-arid surfaces, the ability to peer through optically thin clouds to reveal sea ice, and the detection of oceanic currents, etc. Although these studies demonstrated a comprehensive potential for night-time low-light measurements, quantitative assessment is still needed.
How to quantitatively assess the potential of moonlight remote sensing needs a thorough investigation and more questions must be addressed. Currently, there exist many different satellite sensors, as mentioned above, with different characteristics in terms of spectra and spatial resolutions. Furthermore, drones, as a new near-ground remote-sensing platform, also have great potential to study changes in lighting at night [
14]. These sensors have not been discussed for Earth observation under moonlight, thus, it is of great significance to analyze and compare them in the context of radiometric correction of nightlight remote sensing data and for the design of next generation night-time sensors. Another important question is the quantitative characterization of non-self-lighting objects under night-time low-light environments, considering that there is no rigorous quantitative analysis to date, such as land cover classification under moonlight lighting conditions.
We first compare the differences in night-time observations under moonlight using different sensors, the mono-spectral VIIRS/DNB night-time image, the multi-spectral night-time photos taken by astronauts from the International Space Station, and UAVs. We then explore the potential of nightlight remote sensing through land cover classification under night-time low-light conditions, with a specific focus on detecting non-self-lighting features at night. Finally, we propose a new concept of nightlight remote sensing—moonlight remote sensing, which uses moonlight as a stable lighting source to observe the Earth’s surface, and which focuses on night-time remote sensing mechanisms and applications under lunar illumination. With these distinct characteristics, moonlight remote sensing is different than traditional nightlight remote sensing, as well as from daytime optical remote sensing.
2. Study Area and Data
The potentiality of moonlight remote sensing in this study was evaluated using two ISS multi-spectral moonlight images, acquired on 24 December 2010 and 28 November 2015; UAV moonlight imagery acquired on 20 June 2021; and VIIRS/DNB imagery acquired on 1 November 2015 (with a full moon). These images with different spatial and spectral resolutions covered three regions, Calgary in Canada, Komsomolsk-on-Amur in Russia, and a small part of the Guangming District, in Shenzhen, China. The Calgary images cover a land area of about 825.56 km
2, located in the south of Alberta, Canada. Calgary is the fourth largest city in Canada and is one of the most livable cities in North America in both 2018 and 2019 and has high living standards. This region has a temperate continental climate, warm in summer, cold and dry in winter, and with four distinctive seasons [
35,
36]. Komsomolsk-on-Amur is a city in Khabarovsk Krai, Russia, located on the west bank of the Amur River in the Russian Far East, characterized by a humid continental climate. There is a long period of snow and ice coverage because of the high latitude [
37]. The Guangming District, a recently developed area in Shenzhen, has been planned as the Shenzhen Science City in recent years and is one of the core areas for the construction of a comprehensive national science center in the Guangdong–Hong Kong–Macao Bay Area. After more than 40 years of reform and opening, the urbanization process of Shenzhen has reached a high-level with rapid economic and social development [
38].
Figure 1 shows these three regions and the corresponding nightlight imagery, including information of sensor parameters. These parameters include spacecraft nadir points and altitudes, UAV height, moon altitudes, azimuths, and illumination, as well as cloud cover percentages. In addition, local times when the images were captured were also recorded. The ISS imagery was taken with digital single-lens reflex (DSLR) cameras since 2001, and was the first dataset providing colorful space-borne nightlight images with moderate spatial resolutions (often between 5 m and 200 m). The true-color images in this study were taken with Nikon D3S and D4 DSLR cameras. Both the Nikon D3S and Nikon D4 DSLR cameras were equipped with a Bayer filter in front of the sensor, and were comprised of red (R), green (G), and blue (B) microfilters. The focal lengths of the Nikon DSLR cameras in this study were 180 mm and 400 mm, respectively. These sensors, mounted on a specially designed device to compensate for the movement of the ISS, can take multi-spectral images in visible wavelengths, making it capable of detecting ground features under faint illumination from space [
39,
40,
41,
42]. In addition, a Hasselblad L1D-20c camera was mounted on a DJI MAVIC2 Pro UAV, with a focal length of 35 mm. Both Nikon and Hasselblad cameras are equipped with complementary metal oxide semiconductor (CMOS) sensors. VIIRS is a temperature-controlled charge coupled device (CCD) sensor, and is one of the key instruments aboard the Suomi National Polar-orbiting Partnership (S-NPP) satellite. It is a passive whiskbroom scanning imaging spectroradiometer, taking measurements from 0.4 to 12.2 μm in 15 reflective solar bands (RSB), including a panchromatic DNB and seven thermal emissive bands [
43]. VIIRS has gathered high-quality nightlight images since 28 October 2011, at a spatial resolution of 750 m and in a broad band, covering the 500 to 900 nm spectral region for the DNB [
44,
45] (
Table 1).
Since the ISS and UAV moonlight remote sensing images we obtained were not geo-referenced, we geometrically corrected them using Landsat-7/8 and SuperView-1 optical remote sensing data with accurate geo-referencing information. Moreover, the daytime Landsat-7/8 and Sentinel-2 and SuperView-1 optical remote sensing data were used as auxiliary data to examine the reliability of the moonlight remote sensing imagery. Landsat-7/8 and Sentinel-2 were accessed from the Google Earth Engine (GEE) platform [
46], SuperView-1 constellation is China’s first commercial satellite constellation with high agility and multi-mode imaging capability(
http://www.spacewillinfo.com/SuperView-1English/index.html#pos02 (accessed on 20 September 2021)).
6. Conclusions
Most nightlight remote-sensing studies focus on artificial lights that are emitted at night and can be observed from space, especially those of cities. Little attention has been paid to examine the potential use of reflected moonlight. The present study systematically evaluated the potential of moonlight remote sensing.
- (1)
The reliability of the moonlight remote-sensing imagery.
Using VIIRS/DNB, ISS and UAV moonlight images, the possibilities of moonlight remote sensing were discussed. VIIRS/DNB data successfully acquired spatial distribution and spectral information of land surface, such as snowy mountains, forests, farmlands, and rivers. The ISS data successfully identified snow and forests in the wilderness. In addition, the spatial distribution and texture characteristics of the land surface could be obtained as clearly as optical data in the daytime, especially for the observation of non-self-luminous objects, such as fish ponds, bare land, farmland, and even greenhouses. Therefore, it is believable that moonlight remote sensing is feasible for obtaining non-luminous land surfaces under faint lunar illumination at night, providing a practical way to increase observation frequencies of optical remote sensing.
- (2)
Land surface classification of moonlight remote-sensing imagery.
VIIRS/DNB, ISS, UAV images were classified to explore the potential of moonlight remote sensing. The overall accuracy (OA) and kappa coefficient of the VIIRS/DNB moonlight image are 79.80% and 0.45, respectively. In the low-light suburban areas of Calgary, the overall accuracy and kappa coefficient of the classification result are 87.16% and 0.77, respectively. While the overall accuracy and kappa coefficient of Komsomolsk-na-Amure are 91.49% and 0.85, respectively.
The land surface classification of UAV moonlight images well reflected the spatial distribution characteristics of each land type. The overall accuracy and kappa coefficient are 82.33% and 0.77, respectively. The above results show that these moonlight remote sensing data can be applied well to the classification of a non-self-luminous land surface at night.
- (3)
The characteristics of current moonlight remote sensing.
Finally, the characteristics of current moonlight remote sensing were compared from three aspects of bands, spatial resolutions, and sensors. First of all, multi-spectral moonlight remote sensing is more suitable for Earth observation under complex environments at night. Then, the spatial resolution of the moonlight data directly affects the application scenario of moonlight sensors; both CCD and CMOS cameras have great potential to achieve night-time Earth observations under fine lunar illumination.
The present study has systematically proved the huge potential of moonlight remote sensing in detecting non-self-emitting objects at night, which has been overlooked in traditional applications of night-light remote sensing. Although moonlight remote sensing has great potential for Earth observations, there is still more work to be done to use moonlight as an illuminating source for nightlight remote sensing. It is more difficult for the nocturnal atmospheric radiative transfer model to establish that the moonlight irradiance is much smaller than the sunlight irradiance and atmospheric changes at night are more complicated. In addition, the irradiances of moonlight under different moon phases from a new moon to a full moon also need to be carefully measured and calculated in the future. Meanwhile, studies on the nocturnal atmospheric radiative transfer model and the influence of different moon phase irradiances on the quality of nightlight data are also the basis for promoting quantitative research of moonlight remote sensing.