A global prediction of seafloor sediment porosity using machine learning
A global prediction of seafloor sediment porosity using machine learning
复制标题
DOI：
10.1002/2015gl065279
复制
发表时间：
2015-12-28
影响因子：
5.2
通讯作者：
Becker, Joseph J.
中科院分区：
文献类型：
--
作者：
Martin, Kylara M.;Wood, Warren T.;Becker, Joseph J.
孔隙率（孔隙比）是声学传播、承载强度和许多其他海底现象模型中的一个关键参数。然而，像许多海底现象一样，直接测量既昂贵又稀少。我们在这里展示了如何使用机器学习技术（特别是随机森林）来估计海底各处的孔隙度。这种技术使用稀疏获取的直接样本和其他参数的密集网格，在缺乏直接测量的情况下产生统计上的最佳估计。我们的孔隙度估算在质量上比插值方法得到的结果更符合地质原理，在数量上比插值或回归方法得到的结果更准确。在这里，我们提出了一个海底孔隙度估计5弧分，像素注册网格，使用广泛可用的，密集采样网格的其他海底属性。这些技术是估计难以到达的海底区域（如北极）海底特性的唯一实用手段。
Porosity (void ratio) is a critical parameter in models of acoustic propagation, bearing strength, and many other seafloor phenomena. However, like many seafloor phenomena, direct measurements are expensive and sparse. We show here how porosity everywhere at the seafloor can be estimated using a machine learning technique (specifically, Random Forests). Such techniques use sparsely acquired direct samples and dense grids of other parameters to produce a statistically optimal estimate where direct measurements are lacking. Our porosity estimate is both qualitatively more consistent with geologic principles than the results produced by interpolation and quantitatively more accurate than results produced by interpolation or regression methods. We present here a seafloor porosity estimate on a 5 arc min, pixel registered grid, produced using widely available, densely sampled grids of other seafloor properties. These techniques represent the only practical means of estimating seafloor properties in inaccessible regions of the seafloor (e.g., the Arctic).