Non‐Gaussian data expansion in the Earth Sciences

ABSTRACT A formalism is proposed to generate alternative equiprobable images of an underlying population spatial distribution. The resulting images honour data values at their locations and reflect important characteristics of the data such as patterns of spatial connectivity of extreme‐values. The formalism capitalizes on a coding of all information available into bits (0‐l), which are then processed all together accounting for their patterns of correlation in space. Such common coding allows accounting for qualitative information, possibly of an interpretative nature, to complement the usually sparse hard data available in Earth Sciences applications. The approach proposed, although of a probabilistic nature, does not call for any Gaussian‐type modelling or hypothesis.