This thesis intends to develop a method for spatio-temporal monitoring and analysis of coastline dynamics and its physical impact from 1984 to 2018.
The experiment is conducted with Landsat data from Bonny Island, Nigeria and its interest is to design a fast track process and of low cost.
The preparation of the time series classification is reported along with the theoretical aspects concerning the methods and concepts used in order to conduct Random Forest and Otsu’s thresholding to classify land and water. After the classification, the coastline is extracted by the boundary of the classes, and transects are deployed with DSAS tool. The transects are deployed and produce erosion and accretion rates that upon them, the analysis takes place.
The results show that about in 75,5% of the Island, accretion is dominant with a gain of 6.81m per year. The statistic outcomes of the transects rates show a strong agreement between the results of Otsu and Random forest respectively. A validation applied, calculated the area under erosion and accretion for 1984 to 2003, showed that the Otsu’s method has the highest accuracy. The measurements of the thesis are presented extensively in the conclusions.
Assets and downsides have been found for both Random Forests and Otsu and is suggested that the choice depends on what is intended with them. The free coarse resolution data is suggested in the literature that does not necessarily result to less accuracy in classification methods.