Open Access
Add to BibSonomy
Abstract
Visible infrared imaging radiometer suite (VIIRS) day/night band (DNB) data has been used to detect lit boats during night as it is very sensitive to low radiances. The existing methods for boat detection from VIIRS DNB data are mainly based on thresholds that are estimated by the statistical characteristics of pixels or artificial experience. This may generate detection errors and poor adaptability due to the lack of characterization of boat lights. In this paper, a two-step threshold detection algorithm based on the point spread and the radiative characteristics of nightlight point sources is proposed, so that the interference from adjacent pixels could be reduced as much as possible and a reasonable threshold could be determined. Meanwhile, this algorithm is applied to three study areas, namely the sea area around Tianjin Port in Bohai Sea, the sea area around Shanghai Port in East China Sea, and the sea area around Port Sulphur in Gulf of Mexico. It is demonstrated that the detection precision of the proposed algorithm reaches up to 90% and the recall rate reaches up to 85% in three areas when validated by visual interpretation, and the precision is 85.71% when validated by automatic identification system (AIS) data in the study area of the sea area around Port Sulphur in Gulf of Mexico, which approximately increases by 5% compared with the previous algorithm.
© 2022 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
With the development of low light remote sensing technology, not only the data source is constantly enriched, but also the data quality is continuously improved. The U. S. Air Force Defense Meteorological Satellite Program (DMSP) Operational Linescan System (OLS) launched in September 1976 is the first to offer measurements of nighttime lights since its high sensitivity to low levels of visible light [1]. The National Oceanic and Atmospheric Administration (NOAA) National Geophysical Data Center (NGDC) has been producing DMSP nighttime lights products since 1994 [2] and these products have been used to monitor fishing fleets [3], visualize their spatial distribution and movements over time [4], and estimate the number of vessels [5,6] for many years. There are some shortcomings of DMSP OLS including coarse spatial resolution, lack of on-board calibration, six-bit quantization, saturation on bright lights, lack of a thermal band suitable for fire detection and so on [2]. In October 2011 the National Aeronautics and Space Administration (NASA) and the National Oceanic and Atmospheric Administration (NOAA) launched the Suomi National Polar-orbiting Partnership (SNPP) satellite carrying the first Visible Infrared Imaging Radiometer Suite (VIIRS) instrument [7]. The VIIRS instrument includes a Day/Night Band (DNB) which draws heritage from the low light nighttime visible sensing capability upon the DMSP OLS instrument [8]. The DNB has a similar bandpass to OLS for the low light imaging band (0.5 to 0.9 um) [7], while it has significant improvements over the OLS including full calibration, large dynamic range, increased spatial (0.74 km vs. ∼3 km) and radiometric (14-bit vs. 6-bit) resolutions [9]. In terms of spatial resolution, unlike OLS whose detector ground projection (defining OLS spatial resolution) varies from 2.2 km at nadir to 5.4 km at edge-of-scan [10], the DNB pixel size is maintained at nearly a constant value of 742 m sampling over the entire swath via a unique scan-angle-dependent sub-detector aggregation strategy [11]. Another enhancement is the DNB provides lower detection limits than the OLS (∼2E-11 Watts/cm2/sr vs. ∼5E-10 Watts/cm2/sr), which enables the detection of dimmer lighting [7]. There have been two subsequent builds of the VIIRS sensor for the Joint Polar Satellite System (JPSS) program with JPSS-1 satellite (renamed NOAA-20) and JPSS-2 satellite having launch dates of November 2017 and March 2022 respectively [12]. Performance assessments show that these VIIRS instruments are performing in general as well or better than previous VIIRS instrument operating on the SNPP satellite [13,14]. All these improvements encourage the use of low light imaging data for various remote applications, where maritime boat detection is a typical one.
The potential use of VIIRS DNB images to detect lit boats during night has been demonstrated by quite a few studies [15,16]. Boats operating lights at night appear as “bright spots” in DNB data, which are essentially peak points with higher radiances than surrounding backgrounds. The detection of lit boats is essentially to separate these peak points from backgrounds, and the threshold method is exactly appropriate to address this issue. Elvidge et al. [17] initiate the development of an automatic system for reporting the detection of lit boats based on the threshold method from VIIRS DNB data. In their study, spike median index (SMI) and spike height index (SHI) are defined, and scattergrams of SMI vs. DNB radiances and SHI vs. DNB radiances are analyzed respectively. Thresholds are selected by an analyst after numbers of adjustments in consideration of these scattergrams to generate a list of candidate boat detections, discard ionospheric energetic particle detections and rate boat detections as either strong or weak. Cozzolino et al. [18] propose an algorithm to detect the high gradients between the boat lights and the dark ocean and to obtain a threshold value. A horizontal gradient calculation is carried out and an image of gradients is obtained as the difference of radiance between the pixel (i) and pixel (i+1). For each gradient value that exceeds the mean value, the radiance associated to this gradient is taken as a possible threshold value. The final threshold is then calculated by averaging the minimum and maximum radiance values from the vector of possible threshold values. This final threshold will allow the identification of the pixels illuminated by a boat. Lebona et al. [19] apply an adaptive threshold method commonly used in Synthetic Aperture Radar imagery called constant false alarm rate (CFAR) to DNB data. The aim of this method is to ensure the probability of a false alarm detection is constant. The threshold is set depends on the statistics of surrounding area to keep the percentage of the background pixel values higher than this threshold constant. Pixels above this threshold are declared unusually bright and therefore likely to be samples from the target. Guo et al. [20] design a set of improved algorithms based on the method proposed by Elvidge et al. in [17]. First, the maximum entropy method (MaxEnt) is used as the threshold selection strategy. Specifically speaking, the threshold is calculated to maximize the entropy sum of the target and the background according to probabilities of pixel radiances indicated by a histogram. Then a local spike detection (LSD) algorithm is proposed to remove the false identification pixels near the target boat pixels and illuminated by the boat lights. However, in fact this LSD algorithm simply compares pixel values within 8-neighborhood and takes the local maximum as a local final detection.
To sum up, methods mentioned above estimate thresholds based on statistics and address DNB boat detection issue from the perspective of digital image processing, lacking the consideration of boat lights diffusion and radiation characterization, which may lead to detection errors and poor adaptability.
- (1) As a result of atmospheric scattering and sensor diffraction and defocusing, boat light sources may diffuse into more than one pixel. Radiances of background pixels may exceed the threshold because of adjacency effect caused by this diffusion, leading to false alarms; While radiances of target pixels may fall below the threshold because of energy dissipation, leading to missing alarms.
- (2) Detections based on statistics-dependent thresholds lacking analysis of boat lights radiation properties are vulnerable to noises, since the calculation results of thresholds may not be effective when there are a lot noises in statistical samples. To remedy this defect, artificial experience after review of a large number of datasets is conducted in some threshold selection algorithms, which is labor intensive.
In this paper, an improved threshold method is proposed with the spread and the radiative characteristics of point light sources considered. First, the point spread functions of the atmosphere and VIIRS DNB sensor are calculated; Second, a fundamental radiative transfer process is analyzed to estimate the radiance range of boat lights in DNB images; Third, a two-step threshold detection algorithm based on spatial diffusion and radiance characteristics of boat lights is designed.
The rest of this paper is organized as follows. Section 2 describes the method in detail. Section 3 introduces the results of applying this method to different study areas. Section 4 draws the conclusions.
2. Data and method
2.1 Data description
VIIRS data are divided in three levels: Raw Data Records (RDRs or level 0), Sensor Data Records (SDRs or level 1), and Environmental Data Records (EDRs or level 2) [18]. Records processed by this method are within the SDRs level and can be downloaded from the Comprehensive Large Array-data Stewardship System (CLASS) of NOAA. In this paper, three scenes of moonless DNB images in the SDRs level over three study areas are selected for validating the proposed algorithm, namely, the sea area around Tianjin Port in Bohai Sea (Area 1: 38.28°N to 38.66°N, 117.99°E to 118.56°E; April 9, 2016), the sea area around Shanghai Port in East China Sea (Area 2: 30.87°N to 31.39°N, 122.47°E to 123.12°E; October 24, 2017), and the sea area around Port Sulphur in Gulf of Mexico (Area 3: 28.43°N to 28.97°N, 88.91°W to 89.56°W; January 3, 2019).
In addition, the aerosol optical depth (AOD) over study areas is also used to analyze the diffusion and radiation characterization of boat lights. The AOD at 550 nm with a spatial resolution of 0.75° over study areas is extracted from the European Centre for Medium-Range Weather Forecasts (ECMWF) dataset, which is generated from assimilated model using external data of AOD retrieved from Moderate-resolution Imaging Spectroradiometer (MODIS) (Level 2, collection 5) instruments on board of Terra and Aqua satellites [21]. MODIS data are chosen for their reliability and availability in near-real time. General description of MODIS AOD over land is given by Levy et al. [22,23], and over ocean by Remer et al. [24,25]. The AOD at 550nm extracted from ECMWF dataset for the three studied DNB images is 0.6928, 0.1518, 0.1184 over Area 1, Area 2, Area 3, respectively.
2.2 Method
2.2.1 Analysis of the diffusion characteristics of boat lights
The point spread function (PSF) describes the response of an imaging system to a point source or point object. The PSF in many contexts can be thought of as the extended blob in an image that represents an unresolved object. Thus, the PSF of the atmosphere and the sensor is vital for analyzing the diffusion characteristics of boat lights. The total PSF $PS{F_{total}}$ can be calculated as following:
Where, $PS{F_{atm}}$ is the PSF of the atmosphere, $PS{F_{sensor}}$ is the PSF of DNB sensor. Among them, $PS{F_{atm}}$ could be calculated with an empirical function as follows [26]: Where, $\delta$ is the AOD@550nm, s is the distance from a certain pixel to the center pixel (km), andThese formulas are general and suitable for all the spectral bands, and the effectiveness is illustrated by numerical experiments in [26].
Furthermore, the knife-edge method [27] is used to calculate the PSF of DNB sensor $PS{F_{sensor}}$ in this paper. The first step in the $PS{F_{sensor}}$ construction is to find targets with sharp contrasts. For nighttime data of VIIRS DNB, targets commonly used in the daytime such as dams are not clear anymore, while edges formed by Antarctica ice shelves provide the necessary high contrast and sharp transition [28]. Figure 1 shows one such observation of the Ross Ice Shelf on the night of 25 February 2013. Figure 2 shows the $PS{F_{sensor}}$ construction process. Edge extraction results from the DNB radiance image are shown in Fig. 2(a)(b). Then edge locations are used to construct the edge spread function (ESF) (shown in Fig. 2(c)(d)) by curve-fitting. After that the differentiation of the ESF is taken to get the line spread function (LSF) (shown in Fig. 2(e)(f)). Finally, the LSF along track and cross track are combined and interpolated to get the PSF (shown in Fig. 2(g)).
2.2.2 Fundamental radiometry equations for DNB nightlight
The radiance received by DNB sensor ${L_{DNB}}$ is mainly constituted by three parts: ① ${L_{tar - atm}}$, the radiance emitted by boat lights and pass through the atmosphere; ② ${L_{moon}}$, the moon radiance reflected by surface and pass through the atmosphere; ③ ${L_{path}}$, the atmospheric path radiance. Among them, ${L_{moon}}$ and ${L_{path}}$ are calculated using the MODerate resolution atmospheric TRANsmission (MODTRAN).
For a point source such as the boat light with the electrical power ${P_e}$ and the electrical to radiant power conversion efficacy $\varepsilon$, the light intensity ${I_{target}}$ can be estimated by Eq. (5). The lamps used are assumed to be High Pressure Sodium (HPS) or Light Emitting Diode (LED) lamps, whose electrical to radiant power conversion efficacy $\varepsilon$ are 30% [29,30].
An extended radiance-based source can be considered as a series of point sources with radiant intensity ${I_{target}}$ per unit area. If one defines the unit area as a pixel, the equivalent extended source radiance ${L_{target}}$ can be estimated by the following expression [31]:
Where, ${A_{pixel}}$ is the area of the pixel which is $742 \times 742$ m2.To include atmospheric effects, ${L_{tar - atm}}$ can be described as [31]:
Where, ${M_s}$ is the multiple scattering factor, and $\tau$ is the total transmittance from the surface to the satellite, which is calculated using the MODTRAN radiative transport code.For an active Lambertian illumination source, one can state [31]:
Where s is the spherical albedo, and $\overline \rho $ is the average surface reflectance. Since the spherical albedo across the 500-700 nm spectral range (HPS and LED peak emission spectra within the bandpass of the DNB) is less than 0.4 [31,32] and seawaters have a low reflectance ($< 0.03$) over three study areas [33–35], we expect ${M_s} \approx 1$, and ${L_{tar - atm}}$ can be expressed as:In this study, the power range of boat lights is estimated by investigating the types, power and quantity of commonly used lamps on boats. According to related regulations and rules such as Convention on the International Regulations for Preventing Collisions at Sea, there are two types of boat lights outside the cabin: navigation lights (∼60/65 Watts per lamp) for indicating the boat’s position, status, size and so on, floodlights (∼300/500/2000/3000 Watts per lamp) for illuminating or searching [36,37]. There are also detailed regulations about the quantity of each type of lights in these documents. The total power of lights on one boat approximately ranges from 650 W to 300000 W for calculating ${L_{tar - atm}}$. The calculated ${L_{DNB}}$ ranges from 1.22 to 562.88 $nW/(c{m^2}sr)$ over Area 1, ranges from 2.21 to 1021.68 $nW/(c{m^2}sr)$ over Area 2, and ranges from 2.32 to 1071.51 $nW/(c{m^2}sr)$ over Area 3.
2.2.3 Detection algorithm
A two-step improved threshold detection algorithm is proposed in this study based on radiation and diffusion characterization of boat lights at night. In the first step of the algorithm, the pixels of active light sources are detected using a self-developed spatial filter mask according to $PS{F_{total}}$, so that the contribution of other adjacent pixels could be removed. If the radiance of the target pixel is greater than the maximum value of noises (extracted from DNB data collected on moonless nights over deep ocean) after removing the contribution of other adjacent pixels, the target pixel is considered as an active point light source. The corresponding flow chart for the first step of the proposed algorithm is shown in Fig. 3. In this study, the spatial filter mask D of $5 \times 5$ pixels is defined as Table 1 based on $PS{F_{total}}$, and the derived masks for three study areas are shown in Table 2, Table 3 and Table 4 respectively. The expansion formula to calculate the output response of a DNB image processed by the filter mask D is:
Where $L(x,y)$ is the original radiance of a pixel at coordinates $(x,y)$ in an $M \times N$ DNB image, and ${L_{active}}({x,y} )$ is the corresponding processed radiance after the first step, $i,j ={-} 2, - 1,0,1,2$, $x = 1,2,\ldots ,M$, $y = 1,2,\ldots ,N$.
Table 1. Filter Mask D for Performing Spatial Filtering Operations on the Image
Table 2. Spatial Filter Mask for the Sea Area around Tianjin Port in Bohai Sea
Table 3. Spatial Filter Mask for the Sea Area around Shanghai Port in East China Sea
Table 4. Spatial Filter Mask for the Sea Area around Port Sulphur in Gulf of Mexico
In the second step of the algorithm, these marked active light sources are further filtered according to the radiative thresholds of boat lights so that false alarm sources could be eliminated. Since the radiance of a point light source may spread around within a spread range of $5 \times 5$ pixels area, the radiance of the entire spread range should be taken into consideration when the total radiance of a marked active light source is calculated. In addition, if there are other marked active light sources in the spread range of the one in processing, the radiance increment caused by these sources should be removed first. These operations can be described by Eq. (11) to Eq. (13). First, if there are other active light sources with the radiance of $L({{x_{sur}},{y_{sur}}} )$ at coordinates $({{x_{sur}},{y_{sur}}} )$ in addition to the central active light source at coordinates $({{x_{center}},{y_{center}}} )$ in a certain $5 \times 5$ pixels area, the radiance increment ${L_{sub - sur}}({{x_{subset}},{y_{subset}}} )$ of a pixel at coordinates $({{x_{subset}},{y_{subset}}} )$ caused by these sources is shown as Eq. (11):
Next, ${L_{sub - sur}}({{x_{subset}},{y_{subset}}} )$ is subtracted from the original radiance $L({{x_{subset}},{y_{subset}}} )$ to make the central active light source the only derivation of radiances in this area:
Then the total radiance of the central active light source ${L_{total}}({{x_{center}},{y_{center}}} )$ is calculated:
In the end, the total radiance of the central active light source is compared with the threshold derived from the simulated radiance at the top of atmosphere ${L_{DNB}}$. The corresponding flow chart for the second step of the proposed algorithm is shown in Fig. 4.
2.3 Performance assessment
The performance of the new method is also measured in relation to its ability to estimate the actual position of each identified boats. In this paper, artificial visual interpretation and Automatic Identification System (AIS) data are used to evaluate the detection effect of this algorithm. The visual inspection is to preliminarily verify whether the automatic detection of “bright spots” in DNB images is right. Due to the limitation of the resolution of DNB images, visual interpretation couldn’t determine whether the “bright spots” in images are boats or not. Therefore, the AIS data of sea area in Gulf of Mexico downloaded from NOAA Office for Coastal Management are used to further verify whether the detected “bright spots” are boats. On basis of these two validation approaches, the precision P and the recall rate R (see Eq. (14) and Eq. (15)) are calculated for detection performance assessment of the proposed method in the paper.
Where, $tp$ is the number of spots existed both in detection results by the proposed method and validation dataset (selections from visual interpretation or AIS data), $fp$ is the number of spots only existed in detection results, and $fn$ is the number of spots only existed in validation dataset.3. Results
The method proposed in this study is applied to three moonless images of VIIRS DNB over the sea area around Tianjin Port in Bohai Sea (lunar phase angle: −147.10°), the sea area around Shanghai Port in East China Sea (lunar phase angle: −124.52°), and the sea area around Port Sulphur in Gulf of Mexico (lunar phase angle: 148.85°), and selections of visual interpretation and AIS data are used to validate detection results.
3.1 Boat lights detection results
The detection method described above is applied to three studied images, and the corresponding detection results are shown in Fig. 5, where red crosses represent detected boat lights. From Fig. 5, It can be seen that the level of background noises in three study areas differs greatly. In fact, the mean value of noises is 0.025 $nW/(c{m^2}sr)$, 0.344 $nW/(c{m^2}sr)$, 0.069 $nW/(c{m^2}sr)$ for Area 1, Area 2, Area 3, respectively. Although the AOD information mentioned in Section 2.1 shows the AOD for Area 1 is highest, the level of noises for Area 2 is much higher than the other two areas. The significant difference in noises between Area 2 and the others may make noises rather than the AOD dominate the detection performance.
3.2 Validation of detection results with visual interpretation and AIS data
As described in section 2.3, detection results are firstly compared with the visual interpretation results. The conduct of the visual interpretation is as follows: For dispersed “bright spots”, visual interpretation is conducted by observing the original DNB images directly and marking the “bright spots” on the images; For gathering “bright spots”, it is hard to mark each “bright spot” in a mass of them by just observing the original image, so visual interpretation is conducted with the help of the three-dimensional image, and if the radiance of one pixel is higher than most of its neighboring pixels in the three-dimensional image, it is marked as a target.
The comparison results of three study areas are shown in Fig. 6, and the precision P and the recall rate R are shown in Table 5. It can be seen that the P and R are 93.75% and 91.84% for Area 1; 91.46% and 87.21% for Area 2; 97.96% and 87.27% for Area 3. It is noted that the P reaches up to 90% and the R reaches up to 85% in three areas, indicating less false alarms than missing alarms. For one thing, this may result from the overestimated lit boats using visual interpretation since it is somewhat subjective. And for another, quite a few missing detected boat lights are due to their weak brightness, reflecting the limitation of the proposed algorithm on low-radiance target detection which is vulnerable to noises, and this may be caused by the coarse estimation of noises and non-adaptive threshold. Thus building a fine noise model and developing the adaptive threshold method are taken into consideration in future study for reducing missing alarms further.
Table 5. Calculation Results of Performance Assessment Index
Furthermore, detection results of the sea area around Port Sulphur in Gulf of Mexico are also validated by AIS data, and the results are shown in Fig. 7. Due to the fact that there are much more boats collected by AIS dataset than those imaged by VIIRS DNB, the missing alarm rate is much higher than that in practice, leading to an impractical recall rate R. Therefore, only the precision P is concerned when detection results are validated with AIS data. The calculated P of this study area is 85.71%. The false alarms may be caused by management omissions, namely some lit boats detected by the proposed method have not installed or turned on the AIS transmitter. It may also be due to the false detection of lights from other sources instead of boats appearing as “bright spots” in DNB images, which needs to be further eliminated to improve the precision of the algorithm in the future.
3.3 Comparisons with previous study
In order to further evaluate the proposed threshold method in this study, the detection performance over Area 3 is compared with that of the algorithm proposed by Elvidge et al. in [17]. The comparison of validation results with visual interpretation between the method proposed in this study and that in [17] over Area 3 is shown in Fig. 8. The calculated P of the algorithm proposed by Elvidge et al. in [17] is 84.62% and R is 80.00%, and both of them are lower than those of the method proposed in this study (P is 97.96% and R is 87.27%).
Fig. 8. The comparison of validation results with visual interpretation between two algorithms over the sea area around Port Sulphur in Gulf of Mexico.
The comparison of validation results with AIS data between two algorithms over Area 3 is shown in Fig. 9. The calculated P of the algorithm proposed by Elvidge et al. in [17] is 80.77%. It is noted that the precision of the proposed method in this study increases by about 5%. It can be illustrated by both Fig. 9 and the calculated P that the false alarm of the algorithm proposed in this study is lower than that of the algorithm proposed by Elvidge et al. in [17]. This is attributed to two reasons: For the area with dense boat lights, the interactions among adjacent lights are removed by our algorithm based on the point spread function, so the false alarm could be reduced; For the area with sparse boat lights, the threshold based on radiative characteristics of boat lights is more reasonable and stricter than the threshold based on artificial experience.
Fig. 9. The comparison of validation results with AIS data between two algorithms over the sea area around Port Sulphur in Gulf of Mexico.
4. Conclusion
In this paper, a two-step threshold method has been developed for detecting lit boats at night from VIIRS DNB data. This detection method is fully automated and the point spread and the radiative characteristics of nightlight sources are taken into consideration so that the interference from adjacent pixels could be removed and a reasonable threshold could be calculated. In the first step, a template for illustrating the spatial diffusion of light source is proposed according to the PSF of atmosphere and DNB sensor, and active light sources could be detected using this template. On basis of this, in the second step, the variation of theoretical radiance at the top of atmosphere is simulated with radiative transfer equation as a reasonable threshold, and the observed radiance of active light sources are compared with the threshold so as to detect lit boats.
This algorithm is applied to three study areas under moonless conditions, namely the sea area around Tianjin Port in Bohai Sea, the sea area around Shanghai Port in East China Sea, and the sea area around Port Sulphur in Gulf of Mexico. It is demonstrated that the detection precision and recall rate of the proposed detection algorithm are 93.75% and 91.84%, 91.46% and 87.21%, 97.96% and 87.27% for three study areas respectively when validated by visual interpretation. Since the spatial footprint of DNB is too large to recognize whether detected “bright spots” are boats or not by visual interpretation, AIS data downloaded from NOAA Office for Coastal Management are used to validate the detection results of the sea area around Port Sulphur in Gulf of Mexico, and it is demonstrated that the detection precision is 85.71%. This would illustrate that the proposed algorithm is reliable to a certain extent.
In future work, spectral characterization of boat lights would be taken into consideration to better distinguish lit boats from noises and other light sources and improve the performance of the detection algorithm; Much more validation effort would also be conducted to analyze the types of detected boats.
Funding
National Key Research and Development Program of China (2018YFB0504600).
Acknowledgments
We thank the NOAA’s Comprehensive Large Array-Data Stewardship System and European Centre for Medium-Range Weather Forecasts for providing the VIIRS DNB data and the aerosol optical depth data that are available online. The comments and recommendations by the anonymous reviewers are also greatly appreciated.
Disclosures
The authors declare no conflicts of interest.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
References
1. S. D. Miller and R. E. Turner, “A lunar spectral flux database for quantitative viirs day/night band environmental applications,” in Proceedings of the 4th Symposium on Future National Operational Environmental Satellites (2008), pp. 20–24.
2. C. D. Elvidge, P. Cinzano, D. R. Pettit, J. Arvesen, P. Sutton, C. Small, R. Nemani, T. Longcore, C. Rich, J. Safran, J. Weeks, and S. Ebener, “The Nightsat mission concept,” Int. J. Remote Sens. 28(12), 2645–2670 (2007). [CrossRef]
3. K. Cho, R. Ito, H. Shimoda, and T. Sakata, “Fishing fleet lights and sea surface temperature distribution observed by DMSP/OLS sensor,” Int. J. Remote Sens. 20(1), 3–9 (1999). [CrossRef]
4. P. G. Rodhouse, C. D. Elvidge, and P. N. Trathan, “Remote sensing of the global light-fishing fleet: an analysis of interactions with oceanography, other fisheries and predators,” Adv. Mar. Biol. 39(01), 261–303 (2001). [CrossRef]
5. C. M. Waluda, C. Yamashiro, C. D. Elvidge, V. R. Hobson, and P. G. Rodhouse, “Quantifying light-fishing for Dosidicus gigas in the eastern Pacific using satellite remote sensing,” Remote Sens. Environ. 91(2), 129–133 (2004). [CrossRef]
6. S. I. Saitoh, A. Fukaya, K. Saitoh, B. Semedi, R. Mugo, S. Matsumura, and F. Takahashi, “Estimation of number of pacific saury fishing vessels using night-time visible images,” Int. Arc. Photogramm Remote Sens. Spat. Inf. Sci. 38(8), 1013–1016 (2010).
7. C. D. Elvidge, K. Baugh, M. Zhizhin, and F. C. Hsu, “Why VIIRS data are superior to DMSP for mapping nighttime lights,” in Proceedings of the Asia-Pacific Advanced Network (2013), pp. 62–69.
8. T. E. Lee, S. D. Miller, F. J. Turk, C. Schueler, R. Julian, S. Deyo, P. Dills, and S. Wang, “The NPOESS VIIRS day/night visible sensor,” Bull. Am. Meteorol. Soc. 87(2), 191–200 (2006). [CrossRef]
9. S. D. Miller, S. P. Mills, C. D. Elvidge, D. T. Lindsey, T. F. Lee, and J. D. Hawkins, “Suomi satellite brings to light a unique frontier of nighttime environmental sensing capabilities,” Proc. Natl. Acad. Sci. 109(39), 15706–15711 (2012). [CrossRef]
10. C. D. Elvidge, K. E. Baugh, E. A. Kihn, H. W. Kroehl, and E. R. Davis, “Mapping city lights with nighttime data from the DMSP Operational Linescan System,” Photogramm. Eng. Remote Sens. 63(6), 727–734 (1997).
11. C. F. Schueler, T. F. Lee, and S. D. Miller, “VIIRS constant spatial-resolution advantages,” Int. J. Remote Sens. 34(16), 5761–5777 (2013). [CrossRef]
12. D. Moyer, J. McIntire, X. Xiong, and K. Thome, “Preliminary JPSS-3 VIIRS polarization sensitivity and comparison with S-NPP, JPSS-1 and -2,” in IGARSS 2020-2020 IEEE International Geoscience and Remote Sensing Symposium (2020), pp. 6246–6249.
13. X. Xiong, A. Angal, J. Butler, H. Chen, K. Chiang, N. Lei, Y. Li, and K. Twedt, “Performance assessments and comparisons of S-NPP and NOAA-20 (JPSS-1) VIIRS on-orbit calibration,” Proc. SPIE 10785, 39 (2018). [CrossRef]
14. H. Oudrari, J. McIntire, X. Xiong, J. Butler, Q. Ji, T. Schwarting, and A. Angal, “An overall assessment of JPSS-2 VIIRS radiometric performance based on pre-launch testing,” Remote Sens. 10(12), 1921 (2018). [CrossRef]
15. Y. Liu, S. I. Saitoh, T. Hirawake, H. Igarashi, and Y. Ishikawa, “Detection of squid and pacific saury fishing vessels around Japan using VIIRS Day/Night Band image,” in Proceedings of the Asia-Pacific Advanced Network (2015), pp. 28–39.
16. W. C. Straka, C. J. Seaman, K. Baugh, K. Cole, E. Stevens, and S. D. Miller, “Utilization of the suomi national polar-orbiting partnership (npp) visible infrared imaging radiometer suite (viirs) day/night band for arctic ship tracking and fisheries management,” Remote Sens. 7(1), 971–989 (2015). [CrossRef]
17. C. D. Elvidge, M. Zhizhin, K. Baugh, and F. C. Hsu, “Automatic boat identification system for VIIRS low light imaging data,” Remote Sens. 7(3), 3020–3036 (2015). [CrossRef]
18. E. Cozzolino and C. A. Lasta, “Use of VIIRS DNB satellite images to detect jigger ships involved in the Illex argentinus fishery,” Remote Sens. Appl. Soc. Environ. 4, 167–178 (2016). [CrossRef]
19. B. Lebona, W. Kleynhans, T. Celik, and L. Mdakane, “Ship detection using VIIRS sensor specific data,” in 2016 IEEE International Geoscience and Remote Sensing Symposium (IGARSS) (2016), pp. 1245–1247.
20. G. Guo, W. Fan, J. Xue, S. Zhang, H. Zhang, F. Tang, and T. Cheng, “Identification for operating pelagic light-fishing vessels based on NPP/VIIRS low light imaging data,” Trans. Chin. Soc. Agric. Eng. 33(10), 245–251 (2017). [CrossRef]
21. V. Cesnulyte, A. V. Lindfors, M. R. A. Pitkänen, K. E. J. Lehtinen, J. J. Morcrette, and A. Arola, “Comparing ECMWF AOD with AERONET observations at visible and UV wavelengths,” Atmos. Chem. Phys. 14(2), 593–608 (2014). [CrossRef]
22. R. C. Levy, L. A. Remer, and O. Dubovik, “Global aerosol optical properties and application to Moderate Resolution Imaging Spectroradiometer aerosol retrieval over land,” J. Geophys. Res. 112(D13), 2006JD007811 (2007). [CrossRef]
23. R. C. Levy, L. A. Remer, S. Mattoo, E. F. Vermote, and Y. J. Kaufman, “Second-generation operational algorithm: Retrieval of aerosol properties over land from inversion of Moderate Resolution Imaging Spectroradiometer spectral reflectance,” J. Geophys. Res. 112(D13), 2006JD007815 (2007). [CrossRef]
24. L. A. Remer, Y. J. Kaufman, D. Tanré, S. Matoo, D. A. Chu, J. V. Martins, R. R. Li, C. Ichoku, R. C. Levy, R. G. Kleidman, T. F. Eck, E. Vermote, and B. N. Holben, “The MODIS aerosol algorithm, products, and validation,” J. Atmos. Sci. 62(4), 947–973 (2005). [CrossRef]
25. L. A. Remer, R. G. Kleidman, R. C. Levy, Y. J. Kaufman, D. Tanré, S. Matoo, J. V. Martins, C. Ichoku, I. Koren, H. Yu, and B. N. Holben, “Global aerosol climatology from the MODIS satellite sensors,” J. Geophys. Res. 113(D14), D14S07 (2008). [CrossRef]
26. S. Liang, H. Fang, and M. Chen, “Atmospheric correction of Landsat ETM+ land surface imagery-part I: Methods,” IEEE Trans. Geosci. Remote Sens. 39(11), 2490–2498 (2001). [CrossRef]
27. X. Li, X. Jiang, C. Zhou, C. Gao, and X. Xi, “An analysis of the knife-edge method for on-orbit MTF estimation of optical sensors,” Int. J. Remote Sens. 31(17-18), 4995–5010 (2010). [CrossRef]
28. L. B. Liao, S. Weiss, S. Mills, and B. Hauss, “Suomi NPP VIIRS day-night band on-orbit performance,” J. Geophys. Res. 118(22), 12,705–12,718 (2013). [CrossRef]
29. C. D. Elvidge, D. M. Keith, B. T. Tuttle, and K. E. Baugh, “Spectral identification of lighting type and character,” Sensors 10(4), 3961–3988 (2010). [CrossRef]
30. C. Cao and Y. Bai, “Quantitative analysis of VIIRS DNB nightlight point source for light power estimation and stability monitoring,” Remote Sens. 6(12), 11915–11935 (2014). [CrossRef]
31. R. E. Ryan, M. Pagnutti, K. Burch, L. Leigh, T. Ruggles, C. Cao, D. Aaron, S. Blonski, and D. Helder, “The Terra Vega active light source: A first step in a new approach to perform nighttime absolute radiometric calibrations and early results calibrating the VIIRS DNB,” Remote Sens. 11(6), 710 (2019). [CrossRef]
32. U. Feister and R. Grewe, “Spectral albedo measurements in the UV and visible region over different types of surfaces,” Photochem. Photobiol. 62(4), 736–744 (1995). [CrossRef]
33. J. Zhu, J. Li, B. Han, A. Yang, and F. Gao, “Validation for the remote sensing reflectance of geostationary ocean color imager in Bohai sea,” Acta Optica Sin. 36(s1), 0401002 (2016). [CrossRef]
34. N. Lamquin, C. Mazeran, D. Doxaran, J. H. Ryu, and Y. J. Park, “Assessment of GOCI radiometric products using MERIS, MODIS and field measurements,” Ocean Sci. J. 47(3), 287–311 (2012). [CrossRef]
35. Z. Lee, K. L. Carder, S. K. Hawes, R. G. Steward, T. G. Peacock, and C. O. Davis, “Model for the interpretation of hyperspectral remote-sensing reflectance,” Appl. Opt. 33(24), 5721–5732 (1994). [CrossRef]
36. “Convention on the International Regulations for Preventing Collisions at Sea (COLREGs),” International Maritime Organization (1972).
37. “Rules of navigation,” Suez Canal Authority (2015).
