Numerical relationships between magnetic parameters measured in Quaternary sediments and global paleoclimatic proxies

The complexity of most geological and geophysical problems prompts sometimes the use of non linear mathematical methods to handle them. An adaptive neuro fuzzy inference system (ANFIS) that combines fuzzy logic with neural networks, is applied here to study a paleoclimate section from the Quaternary sedimentary fill of the Lake Mucubají (western Venezuela). The purpose of this work is to find a set of numerical relationships that could predict the possible connections between oxygen isotope (δ¹⁸O) values from two different locations in the northern hemisphere (Ammersee in southern Germany and an ice core from the Greenland Ice Core Project — GRIP) and rock-magnetic parameters measured in Mucubají samples (i.e. mass-specific magnetic susceptibility — χ, magnetic remanence S-ratio, mass-specific saturation isothermal remanent magnetization — SIRM and anhysteretic remanent magnetization — ARM). The best inferences in terms of coefficient of determionation R² and the Root Mean-Square Error (RMSE) are obtained using those magnetic data as input that include information about magnetite grain size distributions, e.g., SIRM and ARM in FIS structures 1χ, 4ARM] and 4ARM, 1SIRM]. A comparison between Ammersee and GRIP actual data, as well as their corresponding inferences for the FIS structure 4ARM, 1SIRM], reveals a reasonable good inference of global trends for both records, overlooking the regional and/or local paleoclimate forcings that might have affected Ammersee. A better correlation between global isotope paleoclimate records and magnetic proxies, is perhaps prevented by the role played by local and regional paleoclimate and tectonism in Mucubají. We also argue that the ratio of ARM over SIRM appears to be related in a complex way to the onset and to the end of the Younger Dryas. Our novel approach to the assessment of a specific paleoclimate case study shows the potential of the ANFIS technique in solving problems where traditional univariate and multivariate linear regression methods could prove inadequate.