ScienceWatch@KIOST
Not a member yet
29648 research outputs found
Sort by
Genetic diversity and structure for conservation genetics of goldeye rockfish Sebastes thompsoni (Jordan and Hubbs, 1925) in South Korea
No full textSebastes thompsoni is a cold-water rockfish of commercial and ecological value off the coast of Korea, requiring conservation management. We analyzed seven microsatellite loci to assess genetic diversity, population structure, and historical effective population size (Ne) of five populations obtained from the South and East Seas of Korea in 2018. The observed heterozygosity (HO = 0.759–0.816) was higher than previously reported, and none of the STRUCTURE, DAPC, or AMOVA analyses detected geographic differentiation among samples from the South and East coasts of Korea, indicating a single population within these coasts. There was genetic flow between the five groups, with migration rates ranging from 4.1 to 19.11. However, the current Ne of all populations is estimated to be <1000, and VarEff-based reconstructions indicate a recent, severe bottleneck following an expansion approximately 600–1200 years ago (100–200 generations ago). This suggests that genetic diversity loss may persist in the future due to long-term habitat loss, fishing pressure, and ocean current fluctuations. Therefore, S. thompsoni should be established as a single management unit covering the Korean Peninsula coast, and habitat protection, overfishing control, genetic management type resource release using various mother and broodstock, and periodic genetic monitoring should be promoted. This study provides evidence to guide efforts to secure long-term genetic resilience and sustainable management of S. thompsoni in Korean coastal waters.11Nsciescopu
해양 기원 블랙 카본 오염: 부산 항구와 황해의 사례 연구
No full textBlack carbon (BC) from marine shipping is a growing global pollutant of concern, with impacts on both coastal and marine environmental quality. Busan Port, ranked among the world’s top ten most polluted ports, and the Yellow Sea, a semi-enclosed basin influenced by significant anthropogenic and natural inputs, illustrate the scale and complexity of this issue. Seasonal monitoring from fall 2020 to summer 2021 revealed non-refractory submicron aerosol mass concentrations of 3.6–11.4 µg·m⁻³ at Busan Port, with persistently high BC levels (1.9–2.2 µg·m⁻³) accounting for 26.5% of total aerosol mass—indicative of dominant shipping-related sources. Elevated elemental carbon was observed in port areas (1.2–1.5 µg·m⁻³) relative to surrounding cities. In the Yellow Sea, episodes with air mass from China showed nitrate dominance (38.1 ± 0.37%) alongside secondary aerosol formation and intense BC contributions (0.6–4.1 µg·m⁻³). Together, these results reveal a persistent and interconnected marine-derived BC contribution, underscoring its growing role in degrading marine air quality.2
Guided Sampling for Diffusion Models: A Single-Image Cloud Removal Method for Dynamic Tidal Environments
No full textRecent advancements in Earth observation technology have led to the growing importance of optical satellite imagery in various applications, including crop growth monitoring, disaster assessment, and environmental change detection. However, persistent cloud cover, estimated to obscure around 66% of the Earth's surface, remains a significant challenge. Most optical bands cannot penetrate clouds, particularly in dynamic environments like tidal flats, where the land–water boundary shifts drastically with each tidal cycle. Conventional cloud removal methods frequently depend on multi-temporal data or external sensors. However, identifying suitable reference scenes for tidal flats is complicated by rapid tidal fluctuations, making automated processes prone to error.
We present a denoising diffusion probabilistic model (DDPM) for cloud removal using only a single optical satellite image. This approach overcomes the limitations of multi-temporal or data-fusion methods by eliminating the need for a separate reference scene. This is particularly advantageous in tidal flats where accurately matched acquisitions are scarce. This approach's core is a gradient-based guided sampling technique, drawing inspiration from classifier guidance in the diffusion process. Instead of generating missing pixels randomly, the model incorporates gradient information extracted from cloud-free portions of the same scene, allowing it to reconstruct obscured areas that align with real-world tidal behaviors.
We evaluated our approach using synthetic datasets with varying levels of cloud coverage, ranging from 10% to 50%, and compared it against conventional inpainting and repainting strategies. Key performance metrics included the Structural Similarity Index (SSIM), Peak Signal-to-Noise Ratio (PSNR), correlation coefficient (CC), and Root Mean Squared Error (RMSE). The proposed guided diffusion method consistently outperformed conventional sampling methods in terms of both numerical accuracy and visual fidelity across all tests. Significant tidal structures, such as intricate land–water transitions, were preserved even under extensive cloud coverage. These findings emphasize the method's effectiveness in reconstructing fine-scale details critical for accurate tidal flat analysis.
To further validate the spatial coherence of our reconstructions, we integrated UAV-derived high-resolution Digital Elevation Models (DEM) and tidal data to delineate land–water boundaries. We then compared these boundaries with those extracted from our restored satellite images using the Normalized Difference Water Index (NDWI). The strong agreement between the two indicates that our method captures the dynamic tidal features with high fidelity, confirming its potential as a viable alternative to multi-temporal or multi-sensor cloud removal approaches in rapidly evolving coastal environments.
In summary, using only a single optical image, the proposed diffusion model-based cloud removal technique is a powerful solution for thick and moderate cloud coverage in tidal flats. By avoiding complicated scene selection or additional sensor data, this gradient-based guided sampling approach is well-suited for real-time environmental monitoring, large-scale topographic mapping, and the generation of reliable training datasets for machine learning. These high-quality reconstructions can also accelerate the creation of accurate, temporally consistent data for predictive analyses in coastal systems, especially during rapid tidal shifts. In conclusion, our method enables more precise and efficient coastal management and supports research into sediment transport, habitat evolution, improved data assimilation, and broader environmental dynamics in intertidal zones.1
Particle flux in the central Yellow Sea in summer
No full textSinking particles were collected using an array of surface-drifting sediment traps at a station in the central Yellow Sea from 31 July to 4 August 2024 to understand export and resuspended particle fluxes during the summer season. During a short-term deployment for 115 hours, the stratified condition of the water column was maintained with a clear thermocline depth at 40 m. The average sinking particle fluxes at 30 m, 40m, and 50m depths were 273 ± 114, 1,458 ± 486, and 3,205 ± 224 mg m-2d-1, respectively. Fluxes measured in the lower layers of the thermocline are probably overestimation of the net sinking particle flux because of the flux from resuspended sediment particles. The increase in total particle flux with depth corresponded well with the vertical profiles of suspended particulate material (SPM) concentration. SPM concentration increased dramatically from 35 m to 70 m water depth, corresponding to the thermocline depth. In addition, the vertical profile of aluminum (Al) concentration in SPM increased monotonically from 35 m to the seafloor. These findings clearly demonstrated that the active resuspension of bottom sediment in the benthic layer significantly contributes to the carbon flux in the central Yellow Sea. Thus, we concluded that export particle flux should be measured below the surface mixed layer or at the thermocline layer.1
Development of accurate transmission line matrix method in acoustics and its application to underwater acoustics
No full textThe transmission line matrix (TLM) method has been used extensively in the microwave field and is an important modeling method with many advantages, but its use in acoustics has been rare. We believe that one of the main reasons for this is that existing acoustic transmission line matrix method modelling studies have not presented the conditions for accurate modelling or a way to verify its accuracy. In this study, the relationship between mesh size and modelling accuracy was investigated for transmission line matrices composed of uniformly sized meshes in order to develop a practical acoustic transmission line matrix method. For this purpose, the acoustic field simulations for a rigid enclosure were performed with the acoustic transmission line matrix method and compared with the exact results from the image method, which can calculate the acoustic field without error. Additionally, we have used this method to create underwater acoustic simulators for cases where the medium is flowing or the sound source is moving. In particular, the calculation result of the simulator for the case where the sound source is moving was compared to the result of a validation experiment in a large water tank. The frequency components of the hydrophone received signal calculated for the case of a transmitter falling in a large water tank were in good quantitative agreement with the frequency components measured in the validation experiment.1
Peculiar Morphology of Montipora millepora Reveals Interspecific Competition for Space Among Two Other Major Foundation Species in Jeju Waters, South Korea
No full textAn atypical surface shape was observed in encrusting coral colonies of Montipora millepora. Initial assumptions on their origin focused on the presence of epibiotic intermediate habitat formers, such as coral-dwelling and -boring organisms. However, further investigations revealed their origin to also be substrate shape-related, prompted by overgrowing other foundation species. The unusual bumps stemmed from encrusting over specimens of the coral Alveopora japonica, and the forked, tube-like structures over holdfasts of the brown alga Ecklonia cava. Spatial distribution patterns and interspecific competition are briefly reviewed. Potential effects of morphological changes for Montipora species identification, as well as implications of altered topography in general, are mentioned.11Ysciescopu
드론 영상을 활용한 인공지능 기반 갯벌의 퇴적상 분류
No full text갯벌 표층 퇴적물의 입도 분포는 저서생물의 서식 환경을 결정하고 퇴적 환경 변화를 이해하는 데 핵심적인 역할을 한다. 세립질 퇴적물은 저서무척추동물의 다양성과 밀접하게 연관되며, 조류 흐름 등 물리적 요인에 따라 공간적으로 뚜렷한 패턴을 보인다. 그러나 기존 갯벌 퇴적상 연구는 두 가지 한계를 보였다. 첫째, 소수의 현장 정점 자료를 내삽하여 분포도를 제작하는 방식은 영상이 지닌 공간적 연속성과 세밀한 변화를 충분히 반영하지 못하였다. 둘째, 원격탐사 영상을 범주별로 분류하는 방식은 연속적인 입도 값이 단순화되어 정량적 정보가 손실되었다.
본 연구에서는 이러한 한계를 극복하기 위해 드론 RGB 영상으로부터 표층 퇴적물의 평균 입도(ϕ, Phi Mean)를 직접 예측하는 인공지능 회귀 모델을 개발하였다. 연구 대상지는 황도 갯벌이며, 다년간 수집된 약 700개의 표층 퇴적물 시료와 50cm 공간해상도의 드론 영상을 이용하여 데이터셋을 구축하였다. 모델의 백본(backbone)으로는 이미지 표현 학습에서 성능이 검증된 Swin Transformer를 적용하였으며, 단일 출력 대신 평균 입도와 더불어 분급(Sorting), 왜도(Skewness), 첨도(Kurtosis)를 동시에 예측하는 다중작업학습(multi-task learning) 구조를 설계하였다. 이를 통해 평균 입도 예측 정확도를 향상시키는 동시에 퇴적물 분포 특성에 대한 복합적인 이해를 가능하게 하였다.
성능 평가는 평균절대오차(MAE)를 통해 수행하였으며, 모델의 강건성과 확장성을 검증하기 위해 두 가지 일반화 실험을 실시하였다. 첫째, 동일 센서로 촬영된 곰소만 드론 영상에 적용하여 공간적 일반화 성능을 평가하였다. 둘째, 동일 지역(황도)을 항공사진과 같은 이종 센서 영상에 적용하여 센서 간 일반화 성능을 확인하였다. 그 결과, 제안된 모델은 서로 다른 지역과 센서 조건에서도 안정적인 예측 성능을 보였다.
본 연구는 갯벌 퇴적 환경을 시공간적으로 정밀하게 분석할 수 있는 새로운 도구를 제시한다. 이는 갯벌 변화 모니터링 기술을 고도화하고, 연안 관리·생태계 보전·복원 사업 계획 수립 등 다양한 현장 적용에 과학적 근거를 제공할 수 있을 것으로 기대된다.2
