Uncovering fishing area patterns using convolutional autoencoder and Gaussian mixture model on VIIRS nighttime imagery
Seong, J.C.; Jang, J.; Yang, J.; Choi, S.H.; Hwang, C. (2026). Uncovering fishing area patterns using convolutional autoencoder and Gaussian mixture model on VIIRS nighttime imagery. ISPRS International Journal of Geo-Information 15(1): 25. https://dx.doi.org/10.3390/ijgi15010025
In: ISPRS International Journal of Geo-Information. MDPI AG: Basel. e-ISSN 2220-9964, more
|   |
| Author keywords |
VIIRS DNB VNL; Korean EEZ; convolutional autoencoder; Gaussian Mixture Model; fisheries; GeoAI |
| Authors | | Top |
- Seong, J.C.
- Jang, J.
- Yang, J.
|
|
|
| Abstract |
The availability of nighttime satellite imagery provides unique opportunities for monitoring fishing activity in data-sparse ocean regions. This study leverages Visible Infrared Imaging Radiometer Suite (VIIRS) Day/Night Band monthly composite imagery to identify and classify recurring spatial patterns of fishing activity in the Korean Exclusive Economic Zone from 2014 to 2024. While prior research has primarily produced static hotspot maps, our approach advances geospatial fishing activity identification by employing machine learning techniques to group similar spatiotemporal configurations, thereby capturing recurring fishing patterns and their temporal variability. A convolutional autoencoder and a Gaussian Mixture Model (GMM) were used to cluster the VIIRS imagery. Results revealed seven major nighttime light hotspots. Results also identified four cluster patterns: Cluster 0 dominated in December, January, and February, Cluster 1 in March, April, and May, Cluster 2 in July, August, and September, and Cluster 3 in October and November. Interannual variability was also identified. In particular, Clusters 0 and 3 expanded into later months in recent years (2022–2024), whereas Cluster 1 contracted. These findings align with environmental changes in the region, including ocean temperature rise and declining primary productivity. By integrating autoencoders with probabilistic clustering, this research demonstrates a framework for uncovering recurrent fishing activity patterns and highlights the utility of satellite imagery with GeoAI in advancing marine fisheries monitoring. |
|
