1.Introduction
As a definition of Maritime Domain Awareness (MDA) (IMO, 2010), all associated activities have become a major topic in monitoring the security, safety, economy, or environment of the oceans (Kanjir et al., 2018). Above all, the awareness of boats is the main issue in areas of fisheries management, illegal fishing, and maritime traffic, etc (Dekker et al., 2013; Kanjir et al., 2018). As the awareness system, Vessel Monitoring System (VMS), such as AIS and V-PASS, has been widely used to collect the presence and activity of boats (IMO, 2000; Cho and Choi, 2018). Under the mandatory standards and regulations, the VMS instruments must be installed on the most of boats (FAO, 1998; IMO, 2000; IMO, 2006). However, there are some disadvantages to collect the accurate information about boats (Kanjir et al., 2018). First, the mandatory standards and regulations prescribe different requirements for installation depending on the size, type, and so on (FAO, 1998; IMO, 2000; IMO, 2006). Second, it is difficult to figure out the boats far away from the coasts since the receiving distance is determined by the communication frequency. Above all, if illegal, unreported, and unregulated (IUU) fishing boats turn off the VMS instruments, it will not be able to identify their position and behavior (FAO, 2001; Park et al., 2020). International Plan of Action (IPOA) addressed these issues as key restrictions on the prevention and eliminating of IUU fishing boats (FAO, 2001).
Recently, satellite-based data has been increasingly used for detection and monitoring the boats as a cooperative surveillance system (Ryu et al., 2018; Ryu et al., 2020). In contrast of VMS, satellite-based data can be used to detect and monitor the boats without being bound by the above constraints (Kanjir et al., 2018; Park et al., 2020). As satellite-based data, optical image or Synthetic Aperture Radar (SAR) data has been used in many publications (Kanjir et al., 2018). However, there are other constraints to simply use these data in night-time. Unlike the day-time, the influence of external energy source is relatively small in the night- time optical image. Above all, the night-time optical image is sensitively affected by the unexpected factors (e.g., Lunar irradiation effect) (Cao et al., 2013; Miller et al., 2013; Lee et al., 2019). Even though the SAR data can be used for detection in the night-time, there is a limitation to observe a wide distribution of boats due to the narrow spatial range. Therefore, the detection of night-time boats requires to understand and approach the night-time optical image with a different concept.
In 2015, U.S. NOAA developed an automatic boat identification system using VIIRS DNB data (Elvidge et al., 2015; Elvidge et al., 2018). They developed the VIIRS Boat Detection (VBD) algorithm to recognize the location of night-time fishing boat with low temporal latency. The detection results have been reported at the homepage (http://payneinstitute.mines. edu/eog/viirs-boat-detection-vbd/) in the vector format of daily/weekly/monthly. These results are actively utilized as the location of night-time fishing boats for behavior and spatiotemporal pattern analysis in various studies (Elvidge et al., 2018; Geronimo et al., 2018; Luo et al., 2018; Hsu et al., 2019; Oh et al., 2019; Park et al., 2020). Although the detection results are highly available and used in many publications, it is difficult to immediately grasp the lights that can be estimated as the night-time fishing boats. The latest results are not available to the general public because they are accessible after 45 days later. The time is the restriction to figure out the night-time fishing boats using the NOAA results. To overcome this restriction, the VBD algorithm can be directly executed and applied to VIIRS DNB data. However, it is difficult to figure out the result by applying it directly because NOAA only provides the result vector, not the algorithm.
Moreover, it is difficult to estimate the threshold applied to the VBD algorithm even if the algorithm is reproduced by referring to the publication of Elvidge et al. (2015). This is because moonlight-free night-time data (new moon) are used to apply the algorithm in that publication. The aspect of threshold is more prominent when comparing with moon phase (Fig. 1). VIIRS DNB data is consisted of the sum of artificial radiance (e.g., shipborne lights) and Earth’s surface reflectance caused by moonlight (Cao et al., 2013). Therefore, these lunar irradiation effect must be corrected appropriately to produce the detection results by applying a specific threshold (Miller et al., 2013).
Fig. 1. Thresholds for VBD algorithm. The data was accessed from U.S. NOAA. This study assumed that the minimum value of Spike Median Index (SMI) is the threshold of the algorithm. Under this assumption, this study compared the thresholds with the moon phases. QF-1, QF-2, and QF-3 indicate the detection intensity: strong, weak, and light under detection, respectively.
For the purpose to figure out the daily detection immediately, it is important to set an appropriate threshold as well as reproduce the VBD algorithm. Therefore, this study aims to identify the frequency of lunar irradiation effect and then to relatively correct the VIIRS DNB data considering the frequency. In this study, the moon phase is regarded as the main frequency of lunar irradiation effect. Considering the moon phase, it is determined which time series appeared in the representative values of VIIRS DNB data. Through the time series, the relational expressions are derived and then used as the offsets for relative correction. After the offset correction, a specific threshold is determined for the application of reproduced VBD algorithm.
This study purposes on the rapid detection of night-time fishing boat using a single specific threshold. For this purpose, this study is conducted as following. First, this study uses VIIRS DNB data received directly from the Korea Ocean Satellite Center (KOSC) of the Korea Institute of Ocean Science & Technology (KIOST) to draw the detection results more quickly. Sequentially, the relative correction is performed to offset the lunar irradiation effect in the consideration of moon phase. Third, the VBD algorithm is reproduced by referring to the publication of Elvidge et al. (2015), and then a specific threshold for applying the algorithm is determined. Finally, the detection results are evaluated by cross-matching with other data (AIS and V-PASS) regarded as ground truth.
2. Materials and Methods
1) S-NPP VIIRS DNB data
This study uses VIIRS DNB data as the night-time optical image. This data has been evaluated to detect the shipborne lights even in dark condition ofthe ocean (Waluda et al., 2004;Brown et al., 2005; Elvidge et al., 2015; Straka et al., 2015; Kanjir et al., 2018).This data was recorded by the sensor mounted on Suomi National Polar-orbiting Partnership (S-NPP) satellite. It was developed by U.S. NOAA and U.S. National Aeronautics and Space Administration (NASA), which launched in October 2011 (Seaman, 2013). In a sun synchronous orbit of 834 km, it is currently operating for global observation. Every day, it observes the area around Korean peninsula between UTC 15:30 and UTC 18:30, taking about 101 minutes to pass through a single orbit. At this time, the observed raw data is received directly to the KOSC of KIOST. After reception, it is preprocessed at the ground system of KOSC, and then released for public (http://kosc.kiost. ac.kr/). On the ground system, the first reception data is processed byTeraSCAN software which convertsthe digital number (DN) into the radiance. The processed data can be accessed in about one hour after it is received. Thus, in terms of access time, it can be used more efficiently for night-time fishing boat detection rather than receiving the data from NOAA Comprehensive LargeArray-data Stewardship System (CLASS) (Lee et al., 2019).
In this study, the received DNB data was processed using an open-source-based Geospatial Data Abstraction Library (GDAL) process (version 3.0.4, released 2020. 01.28) and MATLAB programming. The DNB data was used after resizing around the Korean peninsula (North latitude 51°; South latitude 21°; West longitude 110°; East longitude 150°). For the targeting of nighttime fishing boats, the land surface should be eliminated due to the inference of land lights. So, this study used a full resolution coastline provided by GNOMEOnlineOceanographicDataServer(GOODS) and defined the surround coastline as the buffer zone (2 pixels) to sufficiently eliminate the effect of land lights.
2) Relative correction considering the moon
The lunar irradiation effect is a major factor affecting the radiance of night-time DNB data (Cao et al., 2013). Thus, this study performed several preprocessing steps to search the lunar irradiation effect. The first step is to remove the cloud affected pixels to completely understand this effect in the ocean (Fig. 2). In this step, cloud masking was performed using the product of VIIRS Cloud Mask (VCM) algorithm (Godin, 2014). After the masking, the extracted ocean pixels may contain the artificial lights. These pixels should also be removed because they can stem from the shipborne lights and unnecessary to understand the lunar irradiation effect. The DNB data has sufficient pixels to completely sample the ocean pixels. So, for the second step, the outliers were drastically eliminated to sample the complete and certain ocean pixels. These outliers were defined as the median absolute deviation (MAD) in the range of 1, as shown in following equation. Because the unit of radiance is too small to understand the lunar irradiation effect, the unit was converted from ‘Watts/ cm2/sr’ to ‘nanoWatts/cm2/sr’. The unit was then adjusted to the common logarithms scale (log10), which means the stretched DNB (sDNB) data.
\(\begin{array}{c} s D N B=\min \left[100 \times \text { real }\left(\log _{10} D N B\right), 255\right] \\ M A D=\operatorname{median}(|s D N B-\operatorname{median}(s D N B)|) \\ \quad \text { Outlier } \geq s D N B+M A D \\ \quad \text { Outlier } \leq s D N B-M A D \end{array}\)
Fig. 2. Cloud masking to eliminate the cloud affected pixels. (a) VIIRS DNB data masking lands. (b) Product of VCM algorithm. (c) Extracted ocean pixels masking the lands and clouds.
Additionally, the local scale outliers were removed using a sliding window. The sliding window was set to 3 on the window scale, which can detect the outliers most drastically. For the time series of sDNB radiance, the moon phase was used in this study. The moon phase was calculated using the lunar calendar date available at the Korea Astronomy and Space Science Institute (KASI). In this study, the following equation was used to calculate the moon phase numerically and simply.
\(\begin{array}{c} \text { Caculated moon phase }=15-\mid \text { Moon phase }-15 \mid \\ \text { Full moon } \approx 15 \\ \text { New moon } \approx 0 \end{array}\)
As a subordinary frequency of lunar irradiation effect, seasonal differences were further analyzed in this study. The distribution of sDNB radiance was separated according to the season in the criteria of solar term (Spring: February 4th to May 4th, Summer: May 5th to August 6th, Autumn: August 7th to November 6th, Winter: November 7th to February 3rd). The seasonal distribution was used to derive the curve-fitting expressions in the consideration of moon phase. The curve-fitting expressions were derived using a non- linear model library, which can regress the optimized relation between the moon phase and sDNB data. Ultimately, the derived seasonal expressions were used as the offsets to relatively correct the lunar irradiation effect in the DNB data.
3) VIIRS Boat Detection (VBD) algorithm
U.S. NOAA developed the VBD algorithm focused on the spike since the night-time fishing boat can be estimated as the shipborne light. In the publication of Elvidge et al. (2015), the spikes could be detected using Spike Median Index (SMI). The SMI was applied using a median filter [3, 3], a nonlinear operation to reduce ‘salt and pepper’ noise (Lim, 1990). The median filter can be more effectively used to detect the spikes because an unrepresentative pixel in a neighborhood has a minimal effect on the median. In this study, the median filtering (Fig. 3(b)) was applied to the sDNB data which is relatively corrected the lunar irradiation effect (Fig. 3(a)). It calculates the median value taking into account each pixel of sDNB data. Then, it returns the pixel value sorting into the numerical order to generate the SMI data. The SMI data is generated by subtracting the median filtering from the sDNB data (Fig. 3(c)). In the SMI data, the spikes are detected by setting a threshold. NOAA detected spikes by setting the different thresholds each day. However, this study aims to detect the spikes setting a constant specific threshold as the sDNB data was relatively corrected considering the moon phase. The detected spikes calculated the detection strength using Spike Height Index (SHI) and S3 algorithm. The SHI has two objects within the algorithm. It removes the false detection and divides the spikes into the two categories: strong detection (QF-1) and weak detection (QF-2) (Fig. 3(d)). The SHI is calculated as the ratio between the spike’s radiance and the average radiance of adjacent pixels along the vertical and horizontal directions. The threshold of SHI is set on the smaller of two directions.
\(\begin{array}{c} \text { Vertical SHI = } \left(D N B(j, k)-\left(\frac{D N B(j-1, k)+D N B(j+1, k)}{2}\right) \div D N B(j, k)\right. \end{array}\)
\(\begin{aligned} &\text { Horizontal SHI = } \left(D N B(j, k)-\left(\frac{D N B(j, k-1)+D N B(j, k+1)}{2}\right) \div D N B(j, k)\right. \end{aligned}\)
Fig. 3. Spike detection process in a part of VIIRS DNB data (Jeju Island). (a) sDNB data relatively corrected the lunar irradiation effect considering the moon phase. (b) Median filtering[3, 3] applied to sDNB data. (c) Detected spikes using SMI. (d) SHI applied to the DNB data.
In another index for calculating the detection strength, the S3 algorithm was used to rate the sharpness in each sDNB data. In some cases, the detected spike may be affected by clouds. This spike shows a relatively low detection strength. Therefore, the S3 algorithm was used to measure the sharpness to identify this spike (Fig. 4). The S3 algorithm is calculated by converging two sharpness measure (based on spectral slope and spatial variance) (Vu et al., 2011). If the measurement is less than 0.4, the spike is divided into the light under detection (QF-3). For certain parameters of the S3 algorithm, in this study, block sizes of spectral and spatial sharpness are sets to 8 and 2, respectively. The S3 measurement is calculated by multiplying each square root of the two sharpness measures.
Fig. 4. (a) Detected spikes using the SMI. (b) Sharpness measure based on spectral slope. (c) Sharpness measure based on spatial variance. (d) S3 algorithm converging the two sharpness measures. If the S3 measurement is less than 0.4, it is divided into the light under detection (QF-3).
3. Results
1) Relative correction of VIIRS DNB data
This study purposes to determine a single specific threshold to be applied to the VBD algorithm by correcting the lunar irradiation effect in the VIIRS DNB data. For this purpose, this study retrieved the lunar irradiation effect which affects the DNB data. As shown in Fig. 5, the distribution of sDNB radiance is primary different according to the moon phase. It could be found in the 3-years data (2017-2019) as the same periodical cycle. In this cycle, the sDNB radiance increased as the moon phase was closed to 15. It means that the lunar irradiation effect increased as it closer to the full moon. Through this cycle, this study could consider the moon phase as the frequency of this effect. In addition, the distribution was different according to the season. Fig. 6 shows that the deviation increased as the moon phase closed to 15, but there were differences among the seasons in 3-years data (2017-2019). In spring, the radiance showed a common feature with dense distribution. In contrast, other seasons showed the sparse distributions, which density is much more sparse as closer to winter. In the consideration of these features, it was found that the radiance varies according to the moon phase seasonally.
Fig. 5. Time series of sDNB radiance distribution found in 3-years data. The black and gray lines indicate the sDNB radiance and the calculated moon phase, respectively. The sDNB radiance eliminated the outlier off the range of 1MAD and the outlier of local scale using the sliding window. (a) 2017. (b) 2018. (c) 2019.
Fig. 6. Seasonal distribution of sDNB radiance according to the moon phase. It shows the seasonal differences in 3- years data (2017-2019). (a) Spring. (b) Summer. (c) Autumn. (d) Winter.
Through the seasonal distribution, the curve-fitting expressions could be derived considering the moon phase (Fig. 7). As a result of curve-fitting, a two-term power series model could be used to derive the optimized expressions in the data of spring, autumn, and winter. In summer data, a two-term exponential model could be used to derive the optimized expression. The derived expressions showed the seasonal differences similar with the distribution of sDNB radiance. Between the expressions and distributions, a high correlation was found in spring (R2= 0.99, root-mean-square deviation (RMSE) = 0.55) and summer (R2= 0.99, RMSE = 0.66), whereas a relative low correlation was contained in autumn (R2= 0.55, RMSE = 5.07) and winter (R2= 0.73, RMSE = 5.42).
Fig. 7. Derived seasonal expressions using the 3-years data (2017-2019). The expressions were derived from the curve-fitting. The expressions were used as the offsets to relatively correct the lunar irradiation effect. (a) Spring. (b) Summer. (c) Autumn. (d) Winter. Black dots indicate the distribution of sDNB radiance. Orange lines indicate the derived expressions.
The derived seasonal expressions were used as the offsets to relatively correct the sDNB radiance (Fig. 8 and Table 1). Before the correction, the mean value of sDNB radiance ranged from 38.0978 to 41.7581, indicating a variation of approximately 3.66. The standard deviation (STD) ranged from 6.1018 to 11.8283, with a variation of approximately 5.73. After the correction, the mean value was in the range of -1.3533 to 1.1726, which variation was decreased to 2.53. The STD ranged from 2.7248 to 6.3586, which variation was also decreased to 3.63. As in the before correction case, there were differences in the distribution of corrected radiance over the seasons. However, the radiance was corrected consistently over all seasons (Table 1). There was a slight deviation and a constant distribution.
Fig. 8. Results of relative correction in 3-years data (2017-2019). The derived seasonal expressions were used for offset correction. (a) Spring. (b) Summer. (c) Autumn. (d) Winter. The black and red dots indicate the before and after correction, respectively.
Table 1. Statistics of 3-years data (2017-2019) before and after the relative correction in each season
2) Threshold determination of VBD algorithm
In this study, the lunar irradiation effect was relatively corrected only considering the moon phase seasonally. Through the correction, this study could confirm the relatively constant radiance distribution (Fig. 8). Using the corrected data, this study aimed to determine a single specific threshold. The corrected data was compared with the NOAA detection results at the same pixel location on the same date to determine the threshold. As a comparison method, the SMI data was used to match the corrected data and NOAA detection result. The comparison showed that the thresholds of NOAA detection have a mean of 0.1239, a STD of 0.1141, and a minimum value of 0.0350. On the other hand, the corrected DNB data showed a mean of 0.2957, a STD of 0.0051, and a minimum value of 0.2834. The STD was significantly decreased compared to the thresholds of NOAA detection (Fig. 9). Through the comparison, it could be found that there is not necessary to set a different threshold for each day. In this study, the minimum value of 0.2834 was determined as the threshold to be applied to the reproduced VBD algorithm.
Fig. 9. Comparison of thresholds between this study and the detection results of U.S. NOAA. This study derived the thresholds by matching with the detection results of NOAA at the same pixel location on the same date. Black and red dots indicate the thresholds of NOAA and this study, respectively.
3) Cross-matching with other data
This study reproduced the detection algorithm by referring to the publication of Elvidge et al. (2015). U.S. NOAA has detected the boats by raising the threshold when the moon phase approached to the full moon. However, this study was able to detect the boats with a single specific threshold (0.2834) regardless of the moon phase. For the numerical verification of detection result, cross-matching was performed with other data. As the ground truth data, AIS and V-PASS, provided by the Korea Coast Guard (KCG), were used to evaluate the detection result. For a precise comparison, this study limited the matching area around the Jeju Island. AIS and V-PASS record the boat’s trajectories as the location of vector format. Thus, these data were rasterized to fit the pixel grid of DNB data based on the location. Through the rasterization, it was able to compare with the detection results.
Fig. 10(a) is the DNB data observed on October 17, 2018 which time is 17 h 01 m PM (UTC). In this day, the calculated moon phase was 9.3, which was close to a waxing gibbous. At the same time, AIS and V-PASS data were rasterized to 123 pixels and 407 pixels, respectively, with a total of 462 pixels (68 pixels matched each other) (Fig. 10(b)). As a result of cross- matching with NOAA detection, it showed 39.53% of precision, 29.65% of recall, and 97.80% of accuracy (F1-score = 0.34, False Positive Rate (FPR) = 0.88%) (Fig. 10(d)). On the other hand, this study showed 30.57% of precision, 38.27% of recall, and 97.17% of accuracy (F1-score = 0.34, FPR = 1.69%) (Fig. 10(c)). In both results, the overall values of performance indexes were not significantly different. For the AIS and V-PASS data, some position errors may occur because they recorded based on the receiving time at ground station. So, this study performed an additional cross-matching by creating a buffer zone of 1 pixel size (Fig. 10(e) and (f)). As a result, the accuracy was slightly decreased due to the increasement of FPR resulted by buffer zone. However, the precision and recall were increased and the overall values were not significantly different in both results (Table 2).
Fig. 10. Result of cross-matching on October 17, 2018. (a) VIIRS DNB data accessed from KOSC of KIOST. (b) AIS and V-PASS regarded as the ground truth. (c) Comparison between the detection of this study and the ground truth. (d) Comparison between the detection of NOAA and the ground truth. (e) Comparison between the detection of this study and the ground truth after the application of 1-pixel buffer zone. (f) Comparison between the detection of NOAA and the ground truth after the application of 1-pixel buffer zone.
Table 2. Result of performance test before and after the application of 1 pixel buffer zone (October 17, 2018)
4. Discussion and Conclusion
Up to date, U.S. NOAA leads to study the detection of night-time fishing boat using the satellite data. They have used the S-NPP VIIRS DNB data to detect the shipborne light. The detection results have been used for the location of night-time fishing boats in various studies (Elvidge et al., 2018; Geronimo et al., 2018; Hsu et al., 2019; Park et al., 2020). Nevertheless, it is difficult to figure out and utilize the daily detection results rapidly. This is because the NOAA releases the daily detection results after 45 days later for public. To overcome this difficulty, there is a way to reproduce the detection algorithm directly. But, due to the lunar irradiation effect, it is also difficult to determine the threshold. As shown in Fig. 1, the NOAA has determined the different thresholds for each day considering the lunar irradiation effect. It is time- consuming to estimate the threshold quantitatively every day. If a single specific threshold is determined, the time can be reduced to figure out the detection. Therefore, this study aims to apply a single specific threshold to rapidly detect the night-time fishing boats in each data while reproducing the algorithm.
In order to determine the threshold, it needs to correct the lunar irradiation effect in the VIIRS DNB data. For the correction of lunar irradiation effect, this study identified the frequency of this effect first. In this study, the distribution of sDNB radiance was primary different according to the moon phase. In addition, the distribution was different according to the season. Considering these features, it could be found that the radiance varies according to the moon phase seasonally. Thus, this study could regard the moon phase as the seasonal frequency of lunar irradiation effect. Through the seasonal distribution, the curve-fitting expressions could be Moon Phase based Threshold Determination for VIIRS Boat Detection derived considering the moon phase. The derived expressions were used as the offset to relatively correct the DNB data. After the correction, the mean STD was strikingly decreased from 8.35 to 4.40. Although there were some differences in the distribution, the radiance was corrected consistently in all seasons. To correct the lunar irradiation effect more precisely, a number of factors should be considered in combination such as the distance of the moon or the angle of the sensor (Elvidge et al., 1999; Wang et al., 2017). Nevertheless, this study relatively corrected this effect only considering the moon phase seasonally. As a result, the STD of the thresholds was significantly decreased compared to the thresholds of NOAA. Also, it is possible to set a constant threshold without determination of the different thresholds each day. Consequently, unlike the detection of NOAA, this study can achieve the constant results applying the single specific threshold to the detection algorithm by the relative correction.
Previously, the detection and monitoring of boats have been mainly studied using the VMS or satellite-based data. However, the VMS and satellite-based data have their own constraints for detection and monitoring of night-time fishing boats in spite of useful advantages. On the contrary, the VIIRS DNB data has the advantage to detect the night-time fishing boats. Nevertheless, it is hard to detect the boats more precisely than the VMS or other satellite-based data in terms of spatial or time resolution. Therefore, it would be efficient to use the detection result and other data together as a cooperative system.
In around the Korean peninsula, the catches of offshore fisheries are decreasing every year (Kim et al., 2014; Lee et al., 2017). Particularly, the catches of night-time fisheries are getting more worse due to the IUU fisheries (Lee et al., 2017; Park et al., 2020). This phenomenon causes the social economic problem in the domestic of South Korea and international disputes in the Northeast Asian Sea (Lee and Jung, 2014). Recently, the South Korea makes an effort to surveil the IUU fisheries using the satellite-based data (Ryu et al., 2018; Ryu et al., 2020). Most of all, there is increasingly attention to detect the night-time fishing boats. Considering these points, this study suggests the necessary data to surveil the IUU fisheries, which can be used for prevention and damage reduction. It is expected that provides the important data to prepare the countermeasures and responses for damage reduction. Particularly, if the IUU fishing boats invade the exclusive economic zone (EEZ) or contiguous zone, it can be used as a more efficient surveillance system for detection, monitoring, and arrest.
Acknowledgements
This research is a part of the 2021 projects entitled “Development of satellite based system on monitoring and predicting ship distribution in the contiguous zone” funded by the Korea Coast Guard (20200495).
References
- Baugh, K., F.C. Hsu, C.D. Elvidge, and M. Zhizhin, 2013. Nighttime lights compositing using the VIIRS day-night band: Preliminary results, Proceedings of the Asia-Pacific Advanced Network, 35: 70-86. https://doi.org/10.7125/APAN.35.8
- Brown, C.E., M.F. Fingas, J.P. Monchalin, C. Neron, and C. Padioleau, 2005. Airborne oil slick thickness measurements: realization of a dream, Proc. of 2005 the Eighth International Conference on Remote Sensing for Marine and Coastal Environments, Halifax, NS, CAN, May. 17-19, p. 8.
- Cao, C., X. Xiong, R. Wolfe, F. DeLuccia, Q. Liu, S. Blonski, G. Lin, M. Nishihama, D. Pogorzala, H. Oudrari, and D. Hillger, 2013. NOAA Technical Report NESDIS 142A, NOAA, Washington. DC, USA.
- Cho, K., R. Ito, H. Shimoda, and T. Sakata, 1999. Technical note and cover fishing fleet lights and sea surface temperature distribution observed by DMSP/OLS sensor, International Journal of Remote Sensing, 20: 3-9. https://doi.org/10.1080/014311699213569
- Cho, S.J. and H.J. Choi, 2018. Recent Trends and Their Implications of Marine Activities Mapping for Marine Spatial Planning, Journal of the Korean Society for Marine Environment & Energy, 21(4): 270-280 (in Korean with English abstract). https://doi.org/10.7846/JKOSMEE.2018.21.4.270
- Croft, T.A., 1978. Nighttime images of the earth from space, Scientific American, 239(1): 86-101. https://doi.org/10.1038/scientificamerican0778-86
- Dekker, R., H. Bouma, E. den Breejen, B. van den Broek, P. Hanckmann, M. Hogervorst, A. Mohamoud, R. Schoemaker, J. Sijs, R. Tan, and A. Toet, 2013. Maritime situation awareness capabilities from satellite and terrestrial sensor systems, Proc. of 2013 Maritime Systems and Technologies (MAST Europe), Gdansk, POL, Jun. 4-6, pp. 1-6 .
- Elvidge, C.D., K.E. Baugh, E.A. Kihn, H.W. Kroehl, and E.R. Davis, 1997. Mapping city lights with nighttime data from the DMSP Operational Linescan System, Photogrammetric Engineering and Remote Sensing, 63(6): 727-734.
- Elvidge, C.D., K.E. Baugh, J.B. Dietz, T. Bland, P.C. Sutton, and H.W. Kroehl, 1999. Radiance calibration of DMSP-OLS low-light imaging data of human settlements, Remote Sensing of Environment, 68(1): 77-88. https://doi.org/10.1016/S0034-4257(98)00098-4
- Elvidge, C.D., K.E. Baugh, M. Zhizhin, and F.C. Hsu, 2013. Why VIIRS data are superior to DMSP for mapping nighttime lights, Proc. of 2013 the Asia-Pacific Advanced Network, Hawaii, USA, Jan. 12-18, vol. 35, pp. 62-69.
- Elvidge, C.D., M. Zhizhin, K. Baugh, and F.C. Hsu, 2015. Automatic boat identification system for VIIRS low light imaging data, Remote Sensing, 7(3): 3020-3036. https://doi.org/10.3390/rs70303020
- Elvidge, C.D., K. Baugh, M. Zhizhin, F.C. Hsu, and T. Ghosh, 2017. VIIRS night-time lights, International Journal of Remote Sensing, 38(21): 5860-5879. https://doi.org/10.1080/01431161.2017.1342050
- Elvidge, C.D., T. Ghosh, K. Baugh, M. Zhizhin, F.C. Hsu, N.S. Katada, W. Penalosa, and B.Q. Hung, 2018. Rating the effectiveness of fishery closures with visible infrared imaging radiometer suite boat detection data, Frontiers in Marine Science, 5: 132. https://doi.org/10.3389/fmars.2018.00132
- FAO (Food and Agriculture Organization of the United Nations), 1998. FAO Guidelines for Responsible Fisheries-Fishing Operations-1 Suppl. 1-1. In: Vessel Monitoring Systems, Food and Agriculture Organization of the United Nations, Rome, ITA.
- FAO (Food and Agriculture Organization of the United Nations), 2001. International Plan of Action to Prevent, Deter and Eliminate Illegal, Unreported and Unregulated Fishing, Food and Agriculture Organization of the United Nations, Rome, ITA.
- Geronimo, R.C., E.C. Franklin, R.E. Brainard, C.D. Elvidge, M.D. Santos, R. Venegas, and C. Mora, 2018. Mapping fishing activities and suitable fishing grounds using nighttime satellite images and maximum entropy modelling, Remote Sensing, 10(10): 1604. https://doi.org/10.3390/rs10101604
- Godin, R., 2014. Joint Polar Satellite System (JPSS) Ground Project, Code, 474: 474-00033, NASA Goddard Space Flight Center Greenbelt, MD.
- Hsu, F.C., C.D. Elvidge, K. Baugh, M. Zhizhin, T. Ghosh, D. Kroodsma, A. Susanto, W. Budy, M. Riyanto, R. Nurzeha, and Y. Sudarja, 2019. Cross-matching VIIRS boat detections with vessel monitoring system tracks in Indonesia, Remote Sensing, 11(9): 995. https://doi.org/10.3390/rs11090995
- IMO (International Maritime Organization), 2000. International Convention for the Safety of Life at Sea (SOLAS) 1974, REGULATION 19 - Carriage requirements for shipborne navigational systems and equipment, SOLAS Chapter V Safety of Navigation, IMO, London, GBR, pp. 13-18.
- IMO (International Maritime Organization), 2006. International Convention for the Safety of Life at Sea (SOLAS) 1974, REGULATION 19 - Carriage requirements for shipborne navigational systems and equipment: 1. Application and requirements, SOLAS Chapter V Safety of Navigation, IMO, London, GBR, p. 13.
- Kanjir, U., H. Greidanus, and K. Ostir, 2018. Vessel detection and classification from spaceborne optical images: A literature survey, Remote Sensing of Environment, 207: 1-26. https://doi.org/10.1016/j.rse.2017.12.033
- Kim, D.Y., 2014. A Study on the transition of KoreanChina Fisheries Agreement and improvement of fisheries-relation issues between two countries, The Journal of Fisheries Business Administration, 45(3): 19-37 (in Korean with English abstract). https://doi.org/10.12939/FBA.2014.45.3.019
- Lee, B., Y.K. Lee, D. Kim, and S.W. Kim, 2019. Correction of Lunar Irradiation Effect and Change Detection Using Suomi-NPP Data, Korean Journal of Remote Sensing, 35(2): 265-278 (in Korean with English abstract). https://doi.org/10.7780/KJRS.2019.35.2.6
- Lee, J.S., J.G. Ryu, and H.K. Kee, 2017.Astudy on the status of Chinese fishing in the East sea off North Korea and directions for countermeasures, The Journal of Fisheries Business Administration, 48(3): 61-74 (in Korean with English abstract). https://doi.org/10.12939/FBA.2017.48.3.061
- Lee, K.N. and J.H.Jung, 2014. Estimating the fisheries losses due to Chinese's illegal fishing in the Korean EEZ, The Journal of Fisheries Business Administration, 45(2): 73-83 (in Korean with English abstract). https://doi.org/10.12939/FBA.2014.45.2.073
- Lim, J.S., 1990. Two-dimensional signal and image processing, Prentice Hall: Englewood Cliffs, NJ, USA, p. 710.
- Luo, L., M. Kang,J. Guo, Y. Zhuang, Z. Liu, Y. Wang, and L. Zou, 2019. Spatiotemporal pattern analysis of potential light seine fishing areas in the East China sea using VIIRS day/night band imagery, International Journal of Remote Sensing, 40(4): 1460-1480. https://doi.org/10.1080/01431161.2018.1524605
- Miller, S.D., W. Straka, S.P. Mills, C.D. Elvidge, T.F. Lee, J. Solbrig, A. Walther, A.K. Heidinger, and S.C. Weiss, 2013. Illuminating the capabilities of the suomi national polar-orbiting partnership (NPP) visible infrared imaging radiometer suite (VIIRS) day/night band, Remote Sensing, 5(12): 6717-6766. https://doi.org/10.3390/rs5126717
- Oh, Y., D.W. Kim, Y.H. Jo, J.D. Hwang, and C.Y. Chung, 2020. Spatial variability of fishing grounds in response to oceanic front changes detected by multiple satellite measurements in the East (Japan) sea, International Journal of Remote Sensing, 41(15): 5884-5904. https://doi.org/10.1080/01431161.2019.1685722
- Park, J.,J. Lee, K. Seto,T. Hochberg, B.A. Wong, N.A. Miller, K. Takasaki, H. Kubota, Y. Oozeki, S. Doshi, and M. Midzik, 2020. Illuminating dark fishing fleets in North Korea, Science Advances, 6(30): eabb1197. https://doi.org/10.1126/sciadv.abb1197
- Ryu, J.H., S. Lee, D.J. Kim, and J.D. Hwang, 2018. Wide Integrated Surveillance System of Marine Territory Using Multi-Platform, Korean Journal of Remote Sensing, 34(2): 307-311 (in Korean with English abstract). https://doi.org/10.7780/kjrs.2018.34.2.2.1
- Ryu, J.H., S.H. Hong, Y.H. Jo, and D.J. Kim, 2020. Coastal Remote Sensing in Korea, Korean Journal of Remote Sensing, 36(2): 231-236 (in Korean with English abstract). https://doi.org/10.7780/kjrs.2020.36.2.2.1
- Seaman, C., 2013. Beginner's guide to VIIRS imagery data, CIRA/Colorado State University, Fort Collins, CO, USA.
- Straka, W.C., C.J. Seaman, K. Baugh, K. Cole, E. Stevens, and S.D. Miller, 2015. 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 Sensing, 7(1): 971-989. https://doi.org/10.3390/rs70100971
- Vu, C.T., T.D. Phan, and D.M. Chandler, 2011. S3: A Spectral and Spatial Measure of Local Perceived Sharpness in Nature Images, IEEE Transactions on Image Processing, 21(3): 934-945. https://doi.org/10.1109/TIP.2011.2169974
- Waluda, C.M., C. Yamashiro, C.D. Elvidge, V.R. Hobson, and P.G. Rodhouse, 2004. Quantifying light-fishing for Dosidicus gigas in the eastern Pacific using satellite remote sensing, Remote Sensing of Environment, 91(2): 129-133. https://doi.org/10.1016/j.rse.2004.02.006
- Wang, Z., X. Xiong, J. Fulbright, and N. Lei, 2017. VIIRS day/night band radiometric calibration stability monitoring using the Moon, Journal of Geophysical Research : Atmospheres, 122(11): 5616-5624. https://doi.org/10.1002/2016JD026372
Cited by
- Efficient SAR Azimuth Ambiguity Reduction in Coastal Waters Using a Simple Rotation Matrix: The Case Study of the Northern Coast of Jeju Island vol.13, pp.23, 2021, https://doi.org/10.3390/rs13234865