Sensitivity analysis of prospectivity modeling to evidence maps: Enhancing success of targeting for epithermal gold, Takab district, NW Iran

Abstract

The Takab metallogenic province in the northwestern Iran hosts a number of gold resources including carline-like and epithermal type mineralization. These occurrences are hosted by tectonized country rocks of Precambrian basement metamorphic rocks and Tertiary volcanic and sedimentary units which are intruded by the intrusion of probable age of Eocene and the sub-volcanic domes of Miocene age. Plausible mineralized zones in this high-potential area can be delimited based on mineral prospectivity maps during the regional exploration surveys. However, the success of mineral prospectivity mapping (MPM) is always plagued by different types of uncertainties (stochastic and systemic). The main aim of this study is to reduce those uncertainties resulting from (a) the imprecise/incorrect selection of targeting criteria, (b) adverse effects of using inefficient or least efficient evidence maps, and (c) employing an integration methodology which suffers from user interference. In this regard, the first case was modulated through the precise identification and selection of controls on mineralization as well as ore-forming processes; the second case was declined by the sensitivity analysis of input geochemical, geological and structural parameters to MPM using the area under the receiver operating characteristics (ROC) curve; and the third case was controlled by incorporation of genetic algorithm (GA) and artificial neural networks (here, MLP-ANN) as a suitable data-driven MPM method. Through sensitivity analysis and using MLP-GA-ANN and MLP-ANN algorithms, various prospectivity models were generated and compared. The results of prediction-rate curves demonstrated that the exclusion of poor evidence maps and that employing a self-adapted data-driven approach (e.g., MLP-GA-ANN) can significantly reduce the systematic uncertainty in mapping of mineral prospectivity. Accordingly, the superior model was then selected for delineating efficient exploration targets.