Nonlinear black-box approaches and data fusion for ocean bathymetry modeling in south Iran

Abstract The detailed knowledge of bathymetry pattern represents a key factor in the deep understanding of ocean processes, physical oceanography, biology, ecohydraulics, and marine geology. However, the accuracy of bathymetry modeling is still low from satellite altimetry, gravity model, and shipborne gravity data. In this paper, a novel scheme is proposed based on black-box theory for regional bathymetry modeling in the Persian Gulf and the Oman Sea via geodetic data sources such as satellite altimetry, gravity model, and shipborne gravity data. Multi-Layer Perceptron (MLP), Adaptive Neuro-Fuzzy Inference System (ANFIS), and Local Linear Model Tree (LOLIMOT) algorithms are used as nonlinear black-box tools to identify the basic mathematical model. The geoid height, gravity gradient, and gravity anomaly are used as inputs to these artificial intelligence models, with the GEBCO bathymetry model as the output. The derived basic model is further improved by assimilating with the shipborne bathymetry measurements using the 3D variational optimization method to determine the final bathymetry model. The model is validated by the shipborne bathymetry in control tracks of regions Chabahar, Genaveh, and Alamshah, and the results show high accuracy and reliability with root mean square errors (RMSEs) of about 4, 0.8, and 0.92 m, respectively. The proposed approach is valuable for various uses in marine science.