A Comparison of Global Seafloor Porosity Estimates: Past Attempts vs. New Techniques
Tuesday, 16 December 2014
Porosity is an important parameter in models of many seafloor and sub-seafloor processes active on a global scale, such as methane hydrate occurrence, sub-seafloor metabolic activities, and hydrothermal circulation of fluids. While global models require a global input grid, direct porosity measurements are sparse. We compare gridding techniques from previous modelers (nearest neighbor interpolation, kriging, etc.) to both physical and statistical prediction techniques including Bayesian networks and random decision forests (RDF). Sparse data (only a few thousand direct seafloor porosity measurements worldwide) and sampling bias (toward shallow water depths and low-middle latitudes) adversely affect the interpolated grids. Estimating porosity as a function of water depth alone is as effective in some regions. We have roughly doubled the number of available data points by calculating porosity from thermal conductivity measurements, but sampling bias and spatial data gaps remain problematic. In contrast, statistical methods such as RDF allow for more geologically intelligent predictions, decreasing the effect of sampling bias and improving predictions in regions with little or no data, while rigorously estimating the uncertainty of the result. Statistical prediction is based on empirical statistical relationships between porosity and other parameters for which we have compiled or generated more densely sampled estimates (e.g. water depth, sediment grain type, proximity to rivers or volcanoes). Our results demonstrate the utility of RDF in predicting properties for use in earth system models.