Assessment of Genetic Algorthim in Developing Bathymetry Using Multispectral Landsat Images

Abstract

Bathymetry is required for coastal zone management; hence, it is important to be evaluated properly. Also, bathymetry is highly dynamic in nearshore zone, so, it needs continuous monitoring. The conventional method for bathymetry retrieval is based on sounding that requires intensive time, cost and calm sea conditions. Recently, remote sensing is commonly used to map the shallow water bathymetry since it is frequently captured. Adding up, satellite images may be attained freely with quietly high resolution like Landsat images; 30 m spatial resolution. Consequently, there is intensive research directed to correlate the reflectance/radiance to the water depths. Variable models either linear or nonlinear were developed while in this research, a nonlinear technique, Genetic Algorithm (GA), is introduced. GA was applied on data from multispectral Landsat images. Landsat images were geo-referenced, radiometrically calibrated and atmospherically corrected to attain the reflectance of different bands. GA was utilized to derive the bathymetry for a coastal stretch along the Egyptian Northern Coast (NC). Several trials have been investigated using a reflectance of a single band and a combination of bands, i.e., blue, green and red bands of Landsat 8. Bathymetry measurements at the study site have been used to calibrate the different models/trials. 70% of the data has been assigned to the training of models and the rest has been utilized in the testing process. Comparison of linear model, ratio transform model and GA has been performed. GA showed a comparable performance for estimating shallow water depths; R-squared= 0.95 and RMSE=0.59 m; while enhancements of the derived bathymetry can be achieved by clustering water depths with different assigned GA equations.