为提高利用重力异常数据反演马里亚纳海沟地形的精度,基于残差深度神经网络(residual deep neural network,RDNN)方法和重力异常等数据反演了马里亚纳海沟的1′×1′海底地形,通过实测检核点水深对RDNN模型的精度进行评估,并和重力地质法(gravity-geology method,GGM)模型结果进行对比.结果表明RDNN较GGM对马里亚纳海沟地形反演更为精细,实测水深检核表明RDNN模型均方根误差为128.98 m,优于GGM模型的150.14 m,且RDNN与船测检核水深也有更好一致性,RDNN深度学习模型为利用重力异常数据反演高精度海底地形提供了参考和依据.
Inversion of mariana trench topography using residual deep neural networks
To improve the accuracy of using gravity data to invert mariana trench topography,this study presents a methodology for bathymetric inversion of the Mariana Trench utilizing Residual Deep Neural Network(RDNN)and gravity anomaly data.The accuracy of the RDNN model is evaluated by the in-situ check point depths,and compared with the Gravity-Geology Method(GGM)model.The results demonstrate that the RDNN provides a more detailed inversion of the Mariana Trench's bathymetry.The root mean square error(RMSE)of the RDNN model is 128.98 m,better than 150.14 m of GGM model,suggesting a better consistency with ship-measured check point depths.The RDNN deep learning model proposed in this study provides a reference for high-precision bathymetric inversion using gravity anomaly data.
gravity anomalyresidual deep neural networksmariana trenchshort-wave gravitational anomalyinversion of topographic features
国家科技期刊平台
NSTL
万方数据
