Introduction
Shoreline erosion is a very serious problem that we face as we try to protect the environment. It is caused by the increase in sea level due to global warming. The typical consequences of shoreline erosion are the receding of coastlines and decreases in landmass. There are many island nations in the middle of the ocean; for example, the Maldives are suffering from this problem. Ultimately, if severe enough, loss of landmass due to shoreline erosion will completely take away islands and cause the loss of ecological habitats. According to a data source online, between 1984-2016, about 24 percent of beaches underwent erosion at a rate of .5m/y.
Thus, it is necessary for us to prevent further erosion from taking place. In order to do this, we must understand how the shoreline has changed throughout the past years as well as predict their future erosion.
Although efforts are currently underway to measure shoreline erosion, most of the methods employed by scientists are not accurate or efficient enough.
Dataset
An example of an image that is not clear enough.
An example of an expectable data.
An example of how the model will be influenced by the image quality.
Method
I used Google Earth (an easily accessible and public online data source) to collect raw data of satellite images of coastal areas. Google Earth Engine contains satellite images of all areas within the world. It has been widely used by researchers and scientists for detecting changes around the world.
There were several problems during this process. Since it was not always sunny, there were clouds that interfered with the shoreline when we tried to detect it. To ensure accurate measurement of the shoreline, I used the “check_detection” model within CoastSat which allowed me to manually check image quality and keep only high-quality images.
This is an example of high-quality data.
I implemented the machine learning model within Coastsat on the raw data from Google Earth. First, I imported all packages from the conda environment for further usage. Then, I set up a couple of improvements beyond the original function of Coastsat, including hyperparameter tunings, batched training to save data storage, beaching classification, and so on. I used the code “coastsat_retrieve_data” to get the image of the selected area. Then, I input the collected data into the machine learning model. In this process, the data goes through a neural network classifier with many layers. The model starts by classifying the images into many different colors and different features with different intensity. It then labels the images with different classes: sand, water, white-water, and other land features. The second step of the model (the most crucial one) is the MNDW1 model which was used to detect the boundary between sand and water.
SWIR1 is the shore-wave infrared band (the output value will be between -1 and 1). A histogram was produced to show the value of different features. The water had a negative value while the sand had a positive value. The model then imposed the Otsu threshold algorithm to find a value that maximized the variance between sand and water. The final algorithm of the model was to use the iso-valued “sand/water” threshold with the Marching Square algorithm to detect a contour as the shoreline.
The post process of the algorithm also required some work, tidal correction being a major part of post-processing. Since the images were not guaranteed to be taken at the same time of the day, they were influenced by the tide and thus led to variation in the data collected.
I put together all the detected shorelines into a graph with matplotlib (a python package) and manually plot 5 transections on the graph to visualize the difference within the same shoreline between different time periods. Ideally, the transects on the graph would be perpendicular to the detected shoreline. I labeled the position of the shoreline relative to others on another graph to visualize the change of the shoreline. The position of each shoreline relative to others would be reflected through a graph with the x-axis as time and the y-axis as position. The graph showed all the shoreline’s position compared to other shorelines and provided a better visual for the shift within the shoreline.
The last step was to use Python to find the mean and median value of the position of the shoreline so I could evaluate whether or not there was any change in shoreline with the calculated results. This step eliminated small errors in the data that were caused by weather or tide.
Conclusion
From the data collected and the research I have committed, I conclude that there was indeed a change in the shoreline position during the time period I modeled, and that the shift is not very significant. However, due to data scarcity during the early time period and the short expanse of data, it was difficult to reach a conclusion with certainty. Despite this, machine learning has proven to be a better and more accurate model compared to the traditional method of shoreline erosion detection.
BIBLIOGRAPHY
Bennett, M. K., Younes, N., & Joyce, K. (2020, August 28). Automating drone image processing to map coral reef substrates using Google Earth engine. MDPI. Retrieved February 17, 2022, from https://www.mdpi.com/2504-446X/4/3/50/htm
Choung, Y.-J., & Jo, M.-H. (2017, May 14). Comparison between a machine-learning-based method and a water-index-based method for shoreline mapping using a high-resolution satellite image acquired in Hwado Island, South Korea. Journal of Sensors. Retrieved February 17, 2022, from https://www.hindawi.com/journals/js/2017/8245204/
Dwarakish, G. S., & Nithyapriya, B. (2016, July 9). Application of soft computing techniques in coastal study – A Review. Journal of Ocean Engineering and Science. Retrieved February 17, 2022, from https://www.sciencedirect.com/science/article/pii/S2468013316300109
Goldstein, E. B., Coco, G., & Plant, N. G. (2019, April 23). A review of machine learning applications to coastal sediment transport and Morphodynamics. Earth-Science Reviews. Retrieved February 17, 2022, from https://www.sciencedirect.com/science/article/abs/pii/S001282521830391X
Kumar, L., Afzal, M. M., & Afzal, M. S. (n.d.). Mapping shoreline change using machine learning: a case study from the eastern Indian coast. Retrieved February 17, 2022, from https://www.researchgate.net/publication/342563133_Mapping_shoreline_change_using_machine_learning_a_case_study_from_the_eastern_Indian_coast
Kupilik, M., Witmer, F., MacLeod, E.-A., Wang, C., & Ravens, T. (2017, December 4). Gaussian process regression for Arctic coastal erosion forecasting. arXiv.org. Retrieved February 17, 2022, from https://arxiv.org/abs/1712.00867
Luijendijk, Arjen, et al. "The state of the world’s beaches." Scientific reports 8.1 (2018): 1-11.
Luijendijk, Arjen, et al. "The state of the world’s beaches." Scientific reports 8.1 (2018): 1-11.
Ye Yincan et al, in Marine Geo-Hazards in China, 2017
Rachid Amara, ... Baghdad Ouddane, in World Seas: an Environmental Evaluation (Second Edition), 2019
Comments