Evaluation of soft classification algorithm for sub-pixel mapping of the shoreline

The main objective of this research is to study the potential of different soft classification algorithms to determine subpixel composition for shoreline mapping. Sub-pixel composition information is an important element in determining the accuracy of shoreline maps based on subpixel analysis. This allows the shoreline to be mapped within image pixels producing an accurate and realistic prediction of shoreline position. In this paper, the potential to map the shoreline at sub-pixel scale from different soft classification of relatively coarse spatial resolution satellite imagery were evaluated. The coarse satellite imagery used were at 5 m and 10 m spatial resolution. Three fuzzy functions were evaluated mainly, the Fuzzy sigmoidal, Fuzzy Linear and Fuzzy J-shaped and preliminary results showed that the different fuzzy functions gives varied accuracy. The fuzzy sigmoidal function produced the most accurate sub-pixel prediction with an r of 0.93 for 5 m pixel, while the Fuzzy J-shaped produced the less accurate prediction r of only 0.47. These predictions were later used to predict sub-pixel shoreline position based on a contouring technique. As expected the Fuzzy sigmoidal produced the most accurate shoreline with RMSE of 2.98 m from the actual shoreline.