The resolution of past heat flow in sedimentary basins from non-linear inversion of geochemical data: the smoothest model approach, with synthetic examples

Abstract

SUMMARY An inversion technique to reconstruct the heat flow history of a sedimentary basin from downhole geochemical (or thermal indicator) data is presented. The method has been successfully applied to other geophysical inverse problems and attempts to bound a property of the model. This contrasts with the more common approach of merely finding a model which can predict the data, which is less meaningful for underdetermined problems. In this particular application we seek the smoothest model that can predict the observed data to within a given misfit value. This stabilizes the highly non-linear inversion problem and suppresses the generation of complexities in the heat flow history which are unwarranted by the data. Both first and second derivative smoothing constraints are considered, and the differences between the resulting models allows an assessment of the resolution of the heat flow history. Examples are given using synthetic vitrinite reflectance, sterane and hopane isomerization and sterane aromatization data. Our synthetic inversions indicate that for models with accurate thermal parameters, burial history, and thermal indicator predictive models, the heat flow generally cannot be well resolved back past the timing of maximum temperatures, which in many cases is likely to be the present day. The ability of a particular data type to resolve heat flow back in time depends on the effective kinetic parameters which control the rate of reaction as a function of temperature. When realistic uncertainties in the burial history, present-day heat flow and kinetic parameters are considered, false structure may be introduced into the heat flow history and the inversion generated heat flow models can show significant differences from different data types. The algorithm has the benefit of highlighting the degree of non-uniqueness in the problem and provides an efficient way of generating heat flow models which contain the minimum amount of variation necessary to satisfy the observations, thereby reducing the risk of overinterpreting the data.