Prediction of coastline evolution using remote sensing and deep learning approach; Case study of the Northwest of the Persian Gulf

Abstract

Considering the rapid population growth in coastal areas and the anthropogenic activities, it is important to have a forecast tool for monitoring and managing the coastal changes. In this manuscript the coastline dynamics are analyzed using remote sensing indices and deep learning approach. The studied area is approximately 212 km north of the Persian Gulf. This area is of great economic and political importance and includes several rivers, ports, estuaries, etc. The Landsat satellite images were collected from the USGS dataset over a period of 38 years. In the next step, two methods were used to detect the coastline. The first method was to use indices such as NDWI, and the second method was to combine the bands. To calculate the changes of the coastline and make predictions, several methods were employed including EPR, LRR, LMS, and SCE. The performance of these traditional methods were compared with the CNN method; especially in complex segments such as close to the mouth of the largest river in the study area. The results show that in the upper parts of the region, i.e. the area of Hendijan port, the situation of changes is relatively intense and on average it reaches ∼6 m/year of accretion. This amount reaches ∼1 m/year in the middle areas of the coast. In contrast, in the lower parts of the region, the coast is under erosion of ∼1 m/year. The predictions for the upper parts indicate an accretion of more than 80 m in the next 20 years, and the amount of erosion for the lower part is ∼13 m in this time period. Also, based on the results, CNN is able to predict the coastline dynamics successfully which is important because it can handle large amount of images and nonlinear interaction of coastline with physical phenomenon such as sea level rise and tidal effects.