search
Contact Us
HOME

  • Crossref logo
Journal Search Engine

to

Search Advanced Search Adode Reader(link)
Download PDF Export Citaion korean bibliography PMC previewer
ISSN : 1229-3431(Print)
ISSN : 2287-3341(Online)
Journal of the Korean Society of Marine Environment and Safety Vol.30 No.6 pp.527-540
DOI : https://doi.org/10.7837/kosomes.2024.30.6.527

Estimation of Overtopping Discharge Using Real-time Monitoring, Numerical, Empirical, and Neural Network Methods

Jong-Yoon Mun*, Wan-Hee Cho**, Khawar Rehman***†
*Executive Director, Marine Information Technology Inc., Seoul, Korea
**Laboratory Chief, Marine Information Technology Inc., Seoul, Korea
***Lead Researcher, Marine Information Technology Inc., Seoul, Korea

* First Author : perz@mitkorea.com, 02-2029-7871


† Corresponding Author : khawar_rehman@mitkorea.com, 02-2029-7871
September 23, 2024 October 21, 2024 October 28, 2024

Abstract


Wave overtopping due to storm surges and high waves, such as swells, is a major coastal flooding hazard, which necessitates accurate predictions for the safety of coastal facilities and people. This study investigates the wave overtopping characteristics, including wave heights and discharge, along the South Korean coastline using numerical, empirical, neural network, gradient boosting, and computer-vision-based models. Wave heights were calculated using coupled ADvanced CIRCulation (ADCIRC) and Simulating Waves Nearshore (SWAN) models using meteorological data obtained from the Korea Meteorological Administration (KMA), Japan Meteorological Agency (JMA), National Centers for Environmental Prediction (NCEP), and European Centre for Medium-Range Weather Forecasts (ECMWF) for the entire east coast of South Korea. The Samcheok Port, located on the east coast of South Korea, was selected as the pilot area for installing a closed-circuit television (CCTV) system to detect wave overtopping. The stills captured from the CCTV videos were analyzed using You Only Look Once (YOLO) to detect and quantify run-up events in the pictures. The performance of the numerical model was qualitatively and quantitatively assessed by comparing the predicted wave characteristics with observed values. The performance of the numerical model was deemed excellent in terms of wave height prediction, with minimum root mean square errors (RMSEs) of 0.60 m and 0.44 m for typhoon and wind conditions, respectively. In the prediction of wave period, the RMSE values for typhoon and strong wind conditions were 1.68 m and 1.84 m, respectively. The study findings confirm that the real-time video monitoring of waves can facilitate reliable prediction of wave overtopping characteristics, which can enable real-time and rapid hazard assessments and provide warnings for the protection of coastal communities.


Wave overtopping , Real-time monitoring , YOLO , Coastal hazards , Coastal flooding

실시간 비디오 모니터링, 수치적, 경험적 및 신경망 방법을 통한 월파량 추정

문종윤*, 조완희**, 레만카와르***†
*㈜해양정보기술 전무이사
**㈜해양정보기술 연구소장
***㈜해양정보기술 수석 연구원

초록


폭풍해일 및 너울과 같은 고파랑으로 인해 발생되는 월파는 심각한 연안 침수 위험을 초래하며, 연안 시설과 주민의 안전을 위해 정확한 예측이 필요하다. 본 연구는 수치적, 경험적, 신경망, 그레디언트 부스팅(gradient boosting) 및 컴퓨터 비전 기반 모델들을 사용하여 해안선 인근의 파고와 월파량을 포함한 월파 특성을 조사하였다. 동해안을 대상으로 한국 기상청(KMA), 일본 기상청(JMA), 미국 국립환경예측센터(NCEP), 유럽 중기기상예보센터(ECMWF)의 기상데이터를 사용하여 ADCIRC 모델과 SWAN 모델을 결합하여 파고를 계산하였다. 월파 감지용 CCTV가 설치된 동해안의 삼척항을 대상지역으로 선정하였다. CCTV에서 촬영된 영상들을 YOLO를 사용하여 분석하였으며, 화면 내의 처오름 현상을 감지하였다. 수치모형의 성능은 예측된 파도 특성과 관측값을 비교하여 정성적, 정량적 측정을 통해 평가하였다. 수치모형의 성능은 파고 예측에서 우수한 것으로 분석되었으며, 태풍과 비태풍 조건에서 파고는 각각 0.60m와 0.44m의 최소 RMSE이고 주기는 각각 1.68m와 1.84m의 RMSE로 분석되었다. 본 연구결과에 의하면 실시간 모니터링은 월파 특성에 대한 신뢰할 수 있는 예측 가능성을 가진다. 실시간 모니터링은 해안지역 보호를 위한 신속한 위험 평가 및 실시간 경보 제공에 활용될 수 있다.


월파 , 실시간 모니터링 , YOLO , 해안 위험 , 해안 침수

    1. Introduction

    According to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC), the coastal flooding is projected to intensify due to climate change induced sea-level rise, increased frequency of typhoons, and storm surges (Oppenheimer et al., 2019). The IPCC AR6 synthesis report predicts a global mean sea-level rise of approximately 50 cm to 100 cm by the end of this century (Lee et al., 2023). The rising sea levels, reduced return periods and increased severity of extreme coastal phenomena is challenging the effectiveness of existing coastal defenses in preventing coastal flooding around the world.

    The effectiveness and resilience of coastal defenses against sea conditions is vital from disaster risk reduction and coastal management perspectives. The design of effective coastal defense system depends on accurate prediction of wave overtopping that requires an understanding of the interaction between sea-level rise, storm surges, tides, and waves. The wave overtopping estimates are based on parameters including overtopping discharge, volume, flow velocity, and thickness (Koosheh et al., 2021). The design of coastal defenses such as sea-walls and dikes can be based on estimation of overtopping rate derived from empirical models (Van der Meer et al., 2018). This requires that there is a significant match between wave and structure conditions between empirical formula assumptions design requirements. Furthermore, since the majority of reliable empirical formulations for wave overtopping estimates are derived from test cases performed in Europe and UK coastal conditions, it is difficult to adopt a suitable empirical formula for other regions around the world if there is large difference between wave and structure characteristics. Wave overtopping and surge overflow predictions are also made by using physics based numerical models (Altomare et al., 2021;Jo et al., 2024;Tuan and Oumeraci, 2010;Vanneste et al., 2014). However, in addition to being computationally intensive, often times these models lack real-time wave overtopping and overflow monitoring abilities that limit their application in disaster response system and disaster risk reduction.

    More recently integrated decision support and hazard assessment systems for coastal management are being introduced through hybrid numerical and ML (machine learning) models. The advantage of these systems is utilizing big data related to coastal systems for desired purposes. Several real-time wave overtopping risk assessment models have been deployed around the world; among them SWEEP-OWWL (Stokes et al., 2021) and HIDRALERTA system (Fortes et al., 2020) are worth mentioning due to their ground-up approach.

    Considering the nature of the threats posed by climate change to coastal regions along with the underlying uncertainties in the scale of predicted changes in sea-level rise and recurrence intervals of extreme events, it is imperative to develop and integrate real-time coastal phenomena monitoring systems. The real-time monitoring and observation systems have the potential to offer broader control over assessing and mitigating coastal hazards. This study investigates the performance of several wave overtopping prediction methods including numerical model, empirical formulae, neural network (NN), extreme gradient boosting (XGBoost), and a deep learning method based on real-time CCTV (closed-circuit television) observations of waves. Samcheok port, located at the east coast of South Korea, is selected as a study area (shown in Fig. 10). We assess the capability of real-time observation system in capturing the real-time wave overtopping with the objective of establishing a safe coastal environment system by continuous observation of overtopping waves.

    The paper is divided into six sections with Section 2 detailing the vulnerability of South Korean coastal regions to wave overtopping. Section 3 investigates the reliability of data concerning waves and meteorological parameters from several agencies. Section 4 describes the wave overtopping prediction methods employed in this study, Section 5 presents the results and discussion, and Section 6 presents the conclusions of the study.

    2. Historical Statistics of Wave Overtopping Damage

    The Korean peninsula is vulnerable to damage caused by wave overtopping particularly due to typhoons that occur mainly from August to October each year. Coastal inundation, damage to structural integrity, coastal erosion, damage to ships are main concerns due to wave overtopping during storm surges and high waves such as swells. Because of the absence of automated wave overtopping monitoring systems, the damage caused by overtopping is usually assessed by manual surveys through site visitations or gathering information from electronic and print media. A survey based on internet news and site observations for wave overtopping damage assessment observed 144 instances of damage between year 2000 and 2021 with the majority of events occurring from August till October. The distribution of overtopping events along Korean Coastal length is shown in Fig. 1. The monthly distribution of damages caused by overtopping instances and the nature of the damages is presented in Table 1 and 2, respectively. Considering the number of damages caused by overtopping events as a base, it is evident from the figure that the wave overtopping risk index is higher for Eastern coast and Jeju Island. To manifest the scale of damage, Fig. 2 shows the snapshots of damages caused by wave overtopping in Busan from typhoon Hinnamnor that made landfall in South Korea on September 6-7, 2022 resulting in 3 fatalities, 8 people missing, and 1 injury according to information from Central Disaster and Safety Countermeasures Headquarters (CDSCHQ). For South Korea, the need for real-time monitoring of coastal flooding due to wave overtopping is evident by its vulnerability and damage statistics. The rising sea-levels and increased frequency and intensity of typhoons due to climate change further necessitate the requirement of a robust system for wave overtopping prediction to establish hazard mitigation and prevention measures.

    3. Meteorological Data Reliability for Wave Height Computation

    This study first assesses the suitability meteorological data for model input from Korea Meteorological Administration (KMA), Japan Meteorological Agency (JMA), National Centers for Environmental Prediction (NCEP), US, and European Centre for Medium-Range Weather Forecasts (ECMWF). For a detailed analysis, the comparisons were performed for both strong wind and typhoon conditions at 23 observation gauges operated by KMA including 10, 7, and 6 gauges in the West, South, and East Sea, respectively. The comparisons of wind velocity between the readings observed at meteorological stations and those observed by the agencies for strong wind conditions and typhoons are shown in Fig. 3 for the east coast. For the strong wind and typhoon conditions comparison periods of 67 and 55 days were selected, respectively. For both strong wind and typhoon conditions, the RMSE for ECMWF and KMA are comparable with the latter showing slightly lower RMSE. For wind records, a correlation of 0.90 and 0.91 was observed for strong wind conditions by KMA and ECMWF, respectively. For typhoon conditions, both KMA and ECMWF reported a correlation of 0.92. A similar analysis was performed for wave height observations. The details of the numerical model used to compute the wave heights are presented in Section 4.1. It was observed that the performance of numerical model using wind input data from sources mentioned above varied between coasts and sources. The comparisons for wave heights were performed between numerical model output and records at observation gauges operated by KHOA (Korea Hydrographic and Oceanographic Agency), MOF (Ministry of Oceans and Fisheries), and KMA (Korea Meteorological Administration). Since this study focuses on wave overtopping for east coast, the comparison between RMSE for wave heights observed by different sources at east coast observation stations is shown in Figs. 4 and 5 for strong wind and typhoon conditions, respectively.

    This study investigated data sources’ reliability for computing maximum wave heights by numerical model. For non-typhoon cases the least and maximum RMSE of 0.80 m and 1.03 m was achieved using KMA and ECMWF observation data, respectively (results shown in Fig. 6). Similar correlation was observed between computed results and field observations for all cases. Fig. 7 shows the performance statistics for typhoon cases with KMA observations showing least RMSE of 1.61 m and a correlation of 0.82. Highest correlation was observed for JMA observations. This analysis established that KMA, JMA, and ECMWF meteorological observations can be used with confidence as numerical model input. This analysis shows significance of using reliable meteorological input data for modeling coastal flows.

    4. Overtopping Prediction Models

    This study selected Samcheok Port at the east coast of South Korea to investigate wave overtopping by numerical, empirical, and AI methods (both neural networks and deep learning based). The cross-section of a typical breakwater at the east coast of South Korea is shown in Fig. 8. The structure and wave parameters used to simulate the wave overtopping by models described in the following subsections are also shown in Fig. 8.

    4.1 Numerical Model

    Wave run-up and overtopping can be either modeled using in-house or commercial software based on non-linear shallow water equations (Rehman et al., 2018). This study used the dynamic coupling between open-source software including ADCIRC (Advanced Circulation) and SWAN (Simulating WAves Nearshore) to model wind-wave-current phenomenon. The coupled ADCIRC-SWAN model is run on unstructured computational domain that consisted of 1,888,531 nodes and 978,776 triangular elements as shown in Fig. 9. The coastline and the nearshore bathymetric data with resolution of 150 m was obtained from KHOA (Yoo et al., 2019). Far shore bathymetry, with a resolution of 500 m, was obtained from GEBCO (General Bathymetric Chart of the Oceans). The coupling time interval between SWAN and ADCIRC was set as 20 minutes. The SWAN wave model included 30 frequencies ranging from 0.031 Hz to 1.420 Hz with 72 directions (5° increase). For depth induced breaking the formulation by (Battjes and Janssen, 1978) was used. The exponential wind growth and whitecapping were modeled using (Janssen, 1991;Janssen, 1989). The readers can refer to the SWAN user manual for details and description on input conditions for the model. The coupled model was tested for typhoons and strong wind instances that occurred between January 1, 2013 and December 31, 2022. The wave heights and periods were computed for 32 wave observation gauges distributed around South Korean coasts as shown in Fig. 10. For analysis, the waves with significant wave heights and period of 3 m and 9 seconds or higher, respectively, were selected. The accuracy of the coupled model was tested by computing wave heights using wind input data from JMA, KMA, NCEP, and ECMWF for both TC and Non-TC conditions. The comparisons between wave heights obtained from numerical model and field observations for typhoon and non-typhoon conditions are shown for Samcheok port in Fig. 11 and 12, respectively. The difference between numerical model outputs using different wind input sources is not too significant. Consistent to the findings reported in Figs. 4-7, the model’s performance is best using ECMWF, JMA, and KMA. Further, it can also be observed that the proposed model is accurate in reproducing both extreme wave conditions such as those encountered during typhoons as well as strong wind conditions as shown in Figs. 11 and 12, respectively.

    4.2 Empirical equations based calculations

    Empirical equations based on physical model tests and field observations are available to estimate overtopping for common coastal defense structures. EurOtop manual (Van der Meer et al., 2018) is widely used by researchers and practitioners for estimating overtopping characteristics. The equations are based on more than 13,000 test cases in the EurOtop database that include both site and research structures. Since this study investigates overtopping over a sloped dike at Samcheok Port, we employ general equation for calculating mean and maximum overtopping discharge (Equations 1 and 2, respectively) over a sloped structures from EurOtop manual (Van der Meer et al., 2018).

    KOSOMES-30-6-527_EQ1.gif
    (1)

    KOSOMES-30-6-527_EQ2.gif
    (2)

    where q is overtopping discharge per meter width of the structure, Hm0 is the wave height at the toe of dike, g is gravitational acceleration, ξm–1-0 is the breaker parameter characterizing whether the wave is breaking or non-breaking, α is the angle between dike slope and horizontal. γb, γf, γβ, γv, γ*, are the influence factors for a berm, roughness element on slope, oblique waves, wall at the end of slope, and non-breaking waves, respectively.

    4.3 Neural Network based prediction

    In the last two decades predictions based on NN models have seen applications in almost all coastal and river engineering phenomenon (Rehman et al., 2021;Rehman et al., 2022). Among the NN methods generally used for prediction of mean overtopping discharge, this study employed EurOtop Artificial Neural Network (ANN) (van Gent et al., 2007) and XGBoost or simply XGB (den Bieman et al., 2021). The EurOtop ANN method is applicable for overtopping prediction at several coastal defense structures including seawalls, flood embankments, breakwaters and other shoreline structures. The model was proposed after analyzing 13,000 test cases included in CLASH database (Steendam et al., 2005) selected around European and UK coasts. In comparisons with other NN models and empirical formulae (Van der Meer et al., 2018;van Gent et al., 2007), XGB was reported to show the least errors when applied for wave overtopping predictions (den Bieman et al., 2021). The readers interested in the details of NN and XGB models are referred to the relevant papers cited in this section.

    4.4 Real-time Image based detection

    To detect overtopping by real-time video monitoring, a CCTV camera was installed on a breakwater at Samcheok Port (location shown in Fig. 13) (Seong et al., 2022). The CCTV was capable of shooting a full high definition (FHD) video and saved a still image at 1 FPS (frames per second). The CCTV was able to shoot only during the daylight hours from 5 AM to 8 PM. The video was stored in sets of 15 minutes each. After storing 60 sets, the stills were processed to identify and separate the images with overtopping. Various deep learning models, having varying precision and complexity, can be used for object detection. Some of the most widely used computer vision models include AlexNet (Krizhevsky et al., 2017), EfficientNet (Tan and Le, 2019), Densenet (Huang et al., 2017), ResNet (He et al., 2016), Mask R-CNN, YOLO (Redmon et al., 2016) among others.

    For object detection, YOLO (You Only Look Once) algorithm was employed in this study. YOLO is based on convolutional neural network (CNN) and uses the convolutional, pooling, fully connected, and output layers to detect wave overtopping by predicting the bounding boxes for waves. As the waves are spread over several cells in the captured images, it is inevitable to have overlapping bounding boxes for breaking waves. A few samples of the predicted run-up heights captured in form of bounding boxes by YOLO are presented in Fig. 14 to demonstrate a qualitative comparison between run-up occurrence and YOLO’s ability to detect the overtopping. It is evident from Fig. 14 that the model can successfully capture run-up occurrences. It was observed that model cannot differentiate between spray rising from wave breaking (Fig. 14a) and actual run-up that can lead to false positives. This issue needs to be addressed either by enabling the model to differentiate between spray and run-up or through manual labeling or annotating of the images by domain experts. The run-up heights can be converted to overtopping heights appropriate reference level in the study area. The main limitation of the video monitoring facility was to analyze waves at night time due to which the records of run-up heights from 8 PM to 5 AM are missing each day. This limitation could be addressed by adding night vision capabilities to video monitoring facility, if necessary.

    5. Results and Discussion

    The quantitative analysis of the numerical model predictions for wave height and period is performed by calculating the RMSE (root mean square error) between observed and predicted values using meteorological data from KMA, JMA, NCEP, and ECMWF. Table 3 presents the collective RMSE for all the observations stations shown in Fig. 10. It is observed from RMSE in Table 3 that the numerical model based on KMA data outperforms other data sources except for wave period prediction in strong wind (non-TC) conditions where NCEP shows the least RMSE. However, the difference is almost insignificant. A noticeable difference was observed in numerical model's performance for typhoon and strong wind conditions with significantly lower RMSE observed for typhoon predictions compared to strong wind conditions. However, by observing the RMSE, it can be concluded that for both typhoon and strong wind conditions the numerical model's performance is excellent.

    The wave height output from numerical model was used to estimate overtopping discharge using EurOtop, NN, and XGB models described in Section 4. The models were tested for high wave intensity period that usually occurs during December each year at the Eastern Coast of South Korea.

    The comparisons between overtopping discharge estimated by different models are presented in Fig. 15. An excellent agreement is observed between models in capturing the exact time of maximum overtopping discharge. However, a difference exists between models for estimation of maximum discharge quantity and total volume. The underestimation of maximum discharge and total overtopping volume by the EurOtop model can be attribute to the fact that the model is based on test cases performed in Europe that follow different standards for breakwaters and have different wave conditions than those typically encountered on east coast of South Korea. The XGB model shows rather unphysical rising and falling limbs for discharge hydrograph by having too many fluctuations compared to the rest of the models. This is a known issue for XGB based predictions with non-smooth behavior reported in previous studies as well and is attributed to their inherent traits of gradient boosting methods (den Bieman et al., 2020). Considering the units of overtopping discharge, the difference between discharge predictions by investigated models is not too significant. The run-up height predictions based deep learning method using CCTV observations are particularly encouraging in context of real-time monitoring, warning, and safety of individuals and property. However, in context of video monitoring further improvements are required to enable the model to differentiate between different types of run-ups.

    6. Conclusions

    Coastal flooding by sea-level rise, hurricanes, storm surges, and swells is a major hazard faced by coastal communities around the world. The increased frequency and intensity of extreme coastal events is challenging the conventional coastal defense systems. There is a need to adapt the coastal management system to climate change threats through better utilization of advanced computational and machine/deep learning methods that can offer a real-time quantitative assessment of coastal hazards. This study focused on application of various methods for prediction of wave overtopping with the objective to offer real-time predictions for overtopping characteristics for public and property safety.

    A detailed analysis of meteoroidal parameters obtained from different sources was conducted to establish input data reliability that is essential to confirm model accuracy in wave overtopping prediction. A coupled ADCIRC+SWAN numerical model was used to compute wave heights and period along South Korean east coast. Considering the vulnerability of east coast of South Korea to flooding by wave overtopping, Samcheok port that is located along the east coast, was selected as a study area. The numerical model’s performance was validated against field observations for both typhoon and strong wind conditions using meteorological data from KMA, JMA, ECMWF, and NCEP. It was observed that models based on KMA, JMA, and ECMWF data consistently performed better under all conditions. Both qualitative and quantitative assessments were carried out to gauge the numerical model's performance in simulating selected typhoon and strong wind scenarios. The qualitative assessment showed that numerical model's predictions based on data from various sources showed excellent agreement with observed wave heights at Samcheok port. Additionally, the least RMSE between numerical predictions and observations for wave heights was found to be 0.60 m and 0.44 m for typhoon and strong wind conditions, respectively. For wave period, a minimum RMSE of 1.68 m and 1.84 m was observed for typhoon and strong wind conditions, respectively. Overall, the model's performance is observed to be excellent as evident by qualitative and quantitative assessments.

    A CCTV was installed at Samcheok port breakwater as a pilot test to investigate real-time wave monitoring abilities through video capturing. From the captured films, image data was collected and analyzed using an advanced deep learning method based on computer vision technique, which demonstrated the model’s (YOLO) ability in identifying run-up events; however, there is a need to improve the model further to enable wider usage for variety of run-up cases. Conventional empirical and NN methods were also analyzed for their performance on wave overtopping estimates. The results showed satisfactory comparisons between empirical, NN, and XGB models. The integrated numerical and real-time video monitoring method provides overtopping predictions based on both physical principles as well through utilizing big data collected through various sources such as observation gauges, video monitoring, and image capturing, which consequently offers better hazard assessment and mitigation measures.

    Acknowledgments

    This research was conducted with the support of the Korea Institute of Marine Science and Technology Promotion (Project No. 20220180, Development of Wave-Overtopping Quantitative Observation Technology) from the Ministry of Oceans and Fisheries in 2024.

    Figure

    KOSOMES-30-6-527_F1.gif

    Number of wave overtopping occurrences around South Korea from 2000 to 2021. Data collected from Internet news and site surveying.

    KOSOMES-30-6-527_F2.gif

    Coastal flooding damage caused by typhoon Hinnamnor along east coast of South Korea. Snapshots from site survey conducted on September 7, 2022.

    KOSOMES-30-6-527_F3.gif

    RMSE for strong wind conditions (Non-TC) and typhoon conditions (TC) for selected meteorological agencies.

    KOSOMES-30-6-527_F4.gif

    RMSE for wave heights (comparison between numerical model output and wave height observation operated by KHOA, MOF, and KMA) for strong wind conditions (Non-TC) using wind data from ECMWF, KMA, NCEP and JMA along east coast stations.

    KOSOMES-30-6-527_F5.gif

    RMSE for wave heights (comparison between numerical model output and wave height observation operated by KHOA, MOF, and KMA) for typhoon conditions (TC) using wind data from ECMWF, KMA, NCEP and JMA along east coast stations.

    KOSOMES-30-6-527_F6.gif

    Distribution of maximum wave height computed by numerical model at east coast for non-typhoon cases using meteorological data from NCEP, JMA, KMA, and ECMWF.

    KOSOMES-30-6-527_F7.gif

    Distribution of maximum wave height computed by numerical model at east coast for typhoon cases using meteorological data from NCEP, JMA, KMA, and ECMWF.

    KOSOMES-30-6-527_F8.gif

    Cross-sectional view of a breakwater at Samcheok Port. The wave and structure parameters shown in the figure are used as input in models described in Section 4.

    KOSOMES-30-6-527_F9.gif

    Computational mesh for numerical model based on coupled ADCIRC/SWAN. (top) Overall domain for wave prediction model; (bottom) domain around Korean Peninsula.

    KOSOMES-30-6-527_F10.gif

    Meteorological and wave buoy network along South Korea coastline with stations operated by KMA, KHOA, and MOF. Samcheok Port on east coast highlighted as study area.

    KOSOMES-30-6-527_F11.gif

    Wave height comparisons between numerical model output and field observations using wind input data from JMA, KMA, NCEP, and ECMWF for 8 most severe typhoons observed at Samcheok Port between 2013 and 2022.

    KOSOMES-30-6-527_F12.gif

    Wave height comparisons between numerical model output and field observations using wind input data from JMA, KMA, NCEP, and ECMWF for 8 most severe wave conditions observed at Samcheok Port between 2015 and 2022.

    KOSOMES-30-6-527_F13.gif

    CCTV location at Samcheok breakwater; top view (left), side view (right).

    KOSOMES-30-6-527_F14.gif

    Sample wave overtopping prediction at different time instants by YOLO at Samcheok Port using CCTV footage.

    KOSOMES-30-6-527_F15.gif

    Comparisons between predicted overtopping wave discharge by different models.

    Table

    Monthly distribution of wave overtopping damage occurrence from 2000 to 2021 reported by media

    Property damages types due to wave overtopping recorded between 2000 and 2021 reported by media

    Summary of RMSE for numerical model predictions and observed wave characteristics for all the wave buoys around South Korea using meteorological data from KMA, JMA, NCEP, and ECMWF

    Reference

    1. Altomare, C., X. Gironella, and A. J. C. Crespo ( 2021), Simulation of Random Wave Overtopping by a WCSPH Model, Applied Ocean Research 116:102888.
    2. Battjes, J. A. and J. P. F. M. Janssen ( 1978), Energy Loss and Set-up Due to Breaking of Random Waves, In Proceedings of the Coastal Engineering Conference. Vol. 1.
    3. Bieman, J. P. den, M. R. A. van Gent, and H. F. P. van den Boogaard ( 2021), Wave Overtopping Predictions Using an Advanced Machine Learning Technique, Coastal Engineering 166 (November 2020): 103830.
    4. Bieman, J. P. den, J. M. Wilms, H. F. P. van den Boogaard, and M. R. A. van Gent ( 2020), Prediction of Mean Wave Overtopping Discharge Using Gradient Boosting Decision Trees, Water (Switzerland) 12 (6).
    5. Fortes, C. J. E. M., M. T. Reis, L. Pinheiro, P. Poseiro, V. Serrazina, A. Mendonça, N. Smithers, M. I. Santos, J. Barateiro, E. B. Azevedo, M. Salvador, and F. V. Reis ( 2020), The HIDRALERTA System: Application to the Ports of Madalena Do Pico and S. Roque Do Pico, Azores, Aquatic Ecosystem Health & Management 23 (4), pp. 398-406.
    6. Gent, M. R. A. van, H. F. P. van den Boogaard, B. Pozueta, and J. R. Medina ( 2007), Neural Network Modelling of Wave Overtopping at Coastal Structures, Vol. 54, No. 8, Coastal Engineering, pp. 586-593.
    7. He, K., X. Zhang, S. Ren, and J. Sun ( 2016), Deep Residual Learning for Image Recognition. In Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Vol. 2016-December
    8. Huang, G., Z. Liu, L. van der Maaten, and K. Q. Weinberger ( 2017), Densely Connected Convolutional Networks, In Proceedings - 30th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2017. Vol. 2017-January.
    9. Krizhevsky, A., I. Sutskever, and G. E. Hinton ( 2017), ImageNet Classification with Deep Convolutional Neural Networks, Communications of the ACM 60 (6), pp. 84-90.
    10. Lee, H., K. Calvin, D. Dasgupta, G. Krinner, A. Mukherji, P. Thorne, C. Trisos, et al. ( 2023), IPCC, 2023: Climate Change 2023: Synthesis Report, Summary for Policymakers. Contribution of Working Groups I, II and III to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change [Core Writing Team, H. Lee and J. Romero (Eds.)], IPCC, Geneva, Switzerland., pp. 1-34.
    11. Janssen, P. A. E. M. ( 1991), Quasi-Linear Theory of Wind- Wave Generation Applied to Wave Forecasting, J. PHYSICAL OCEANOGRAPHY 21 (11), pp. 1631-1642.
    12. Janssen, P. A. E. M. ( 1989), Wave-Induced Stress and the Drag of Air Flow over Sea Waves, Journal of Physical Oceanography 19 (6), pp. 745-754.
    13. Jo, J., S. Kim, N. Mori, and H. Mase ( 2024), Combined Storm Surge and Wave Overtopping Inundation Based on Fully Coupled Storm Surge-Wave-Tide Model, Coastal Engineering 189 (October 2023): 104448.
    14. Koosheh, A., A. Etemad-Shahidi, N. Cartwright, R. Tomlinson, and M. R. A. van Gent ( 2021), Individual Wave Overtopping at Coastal Structures: A Critical Review and the Existing Challenges, Applied Ocean Research 106 (December 2020): 102476.
    15. Meer, J. W. Van der, N. W. H. Allsop, T. Bruce, J. de Rouck, A. Kortenhaus, T. Pullen, H. Schüttrumpf, P. Troch, and B. Zanuttigh ( 2018), EurOtop: Manual on Wave Overtopping of Sea Defences and Related Structures. Report.
    16. Redmon, J., S. Divvala, R. Girshick, and A. Farhadi ( 2016), You Only Look Once: Unified, Real-Time Object Detection, In Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Vol. 2016-Decem.
    17. Rehman, K., H. Khan, Y. -S. Cho, and S. H. Hong ( 2021), Incident Wave Run-up Prediction Using the Response Surface Methodology and Neural Networks, Stochastic Environmental Research and Risk Assessment 0123456789:16.
    18. Rehman, K., K. -Y. Park, and Y. -S. Cho ( 2018), Experimental and Numerical Investigation of Solitary Wave Run-up Reduction, Journal of Coastal Research 85, pp. 1111-1115.
    19. Rehman, K., Y. -C. Wang, and M. Waseem ( 2022), Tree-Based Machine Learning Models for Prediction of Bed Elevation around Bridge Piers Tree-Based Machine Learning Models for Prediction of Bed Elevation around Bridge Piers, Physics of Fluids 085105 (August): 13.
    20. Seong, B. -R., W. -H. Cho, J. -Y. Moon, and K. -H. Lee ( 2022), Real-Time Wave Overtopping Detection and Measuring Wave Run-up Heights Based on Convolutional Neural, Port Res 46 (3), pp. 243-250.
    21. Steendam, G. J., J. W. van der Meer, H. Verhaeghe, P. Besley, L. Franco, and M. R. A. van Gent ( 2005), The International Database on Wave Overtopping, Coastal Engineering 2004, pp. 4301-4313.
    22. Stokes, K., T. Poate, G. Masselink, E. King, A. Saulter, and N. Ely ( 2021), Forecasting Coastal Overtopping at Engineered and Naturally Defended Coastlines, Coastal Engineering 164 (December 2020), 103827.
    23. Tan, M. and Q. V. Le ( 2019), EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks, In 36th International Conference on Machine Learning, ICML 2019. Vol. 2019-June.
    24. Tuan, T. Q. and H. Oumeraci ( 2010), A Numerical Model of Wave Overtopping on Seadikes, Coastal Engineering 57 (8), pp. 757-772.
    25. Vanneste, D. F. A., C. Altomare, T. Suzuki, P. Troch, and T. Verwaest ( 2014), Comparison of Numerical Models for Wave Overtopping and Impact on a Sea Wall, Coastal Engineering Proceedings 1 (34): 5.
    26. Yoo, S. C., J. Y. Mun, W. Park, G. H. Seo, S. J. Gwon, and R. Heo ( 2019), Development of Bathymetric Data for Ocean Numerical Model Using Sea-Floor Topography Data: BADA Ver.1, Journal of Korean Society of Coastal and Ocean Engineers 31 (3), pp. 146-157.
    Copyright © The Korean Society of Marine Environment and safety. All rights reserved.
    (530-729)Mokpo National Maritime Univ., Jukgyo-dong, Mokpo-si, Jeollanam-do, Korea