Recovering Bathymetry of the Gulf of Guinea Using Altimetry-Derived Gravity Field Products Combined via Convolutional Neural Network

Global Predicted Bathymetry Using Neural Networks

“…Upward continuation of gravity anomalies limits the resolution of gravity from satellite altimetry to a length scale of about π times the regional depth (e.g., Smith & Sandwell, 2004), so it is not possible to realistically predict depth from only gravity (and its derivatives) at such scales. An approach using convolutional neural networks, as demonstrated by Annan and Wan (2022), may successfully learn from higher resolution bathymetry in regional settings.…”

“…There has been recent interest in using modern methods from machine learning to improve upon the prediction of bathymetry. For example, Annan and Wan (2022) and Wan et al (2023) have used neural networks with various architectures to predict absolute depth from gravity and gravity-related quantities (e.g., deflections of the vertical, gravity gradients). These models have been limited to a particular study area for training, and the predictions are thus limited to the selected area.…”

Altimetry‐Derived Gravity Gradients Using Spectral Method and Their Performance in Bathymetry Inversion Using Back‐Propagation Neural Network

“…A proven criteria was not followed in selecting these study regions; however, we were specific about the Atlantic and Arctic regions. This Atlantic test region was selected because it is a resource‐rich region; however, it has minimal research coverage in terms of bathymetric studies as compared to another important regions like the South China Sea (Annan & Wan, 2020, 2022). In addition, the Atlantic region helps to assess the performance of the presented approach in a tropical, equatorial region.…”

“…Moreover, there are different varieties of ANN; however, BPNN is the most widely applied variant of ANN in several disciplines; and hence is regarded as the basis for comparison with other ML techniques. Evidence of its efficiency has been documented in diverse geospatial disciplines, such as in geodetic coordinate transformation (Tierra et al., 2008; Ziggah et al., 2016), earth orientation (Schuh et al., 2002), natural terrain classification (Akwensi et al., 2021), surface energy balance (Alemohammad et al., 2017), mass movement (Mandal & Mondal, 2019) and bathymetry inversion (Annan & Wan, 2022).…”

“…Though BPNN is seen as a benchmark ML technique, its capability is yet to be extensively tested for bathymetry inversion. Last but not least, this study's application of BPNN helps to bypass some of the delicate technicalities involved in geodetic bathymetry inversion, such as: isostatic compensation, coherence analysis between bathymetry and gravity field signal, density contrast determination, downward continuation and filter design, to mention a few (Annan & Wan, 2022).…”