Sorry, you need to enable JavaScript to visit this website.
Skip to content Skip to navigation

Generalized sensitivity analysis

General Sensitivity Analysis Fig 1
General Sensitivity Analysis Fig 2
Basin and Petroleum System Modeling (BPSM) covers a large spatial and temporal interval. Many of the input parameters are highly uncertain. Previous studies have shown that the model responses can be impacted by uncertainty from various resources and scales. How to efficiently quantify these uncertainties (both parametric and spatial) remains challenging. Probabilistic approaches based on Monte Carlo simulations are often utilized to tackle these questions, however, a large number of models created from random sampling need to be constructed and simulated. Due to the intensive model construction and analysis process and high computational cost of simulating BPSM projects, we need to optimize the number of models while still covering the possible range of uncertainties. Thus, sensitivity analysis needs to be conducted to identify parameters that are sensitive to model responses and ultimately to improve the efficiency of uncertainty quantification workflow.
 
Graduate student Yao Tong and advisor Tapan Mukerji are exploring the application to BPSM studies of the Generalized Sensitivity Analysis (GSA) Method recently refined and successfully applied to reservoir modeling by Stanford Center for Reservoir Forecasting.  This method combines the Monte Carlo random sampling with advanced resampling statistical method to identify sensitive parameters. It is especially attractive for BPSM because:
  1. GSA method can tackle sensitivities resulting from various types of uncertain parameters (continuous, discrete, and even intrinsic scenario-based), and
     
  2. GSA method can identify not only the traditional single way parameter sensitivities but also multi-way interaction parameter sensitivities, which indicate more subtle correlations that may otherwise be ignored.
In the current preliminary study, we applied this GSA method to a 1D basin model and investigated both single way parameter sensitivities and multi-way interaction parameter sensitivities on source rock maturation responses.
 
The upper figure shows the Pareto plots of the standardized measure of sensitivity for each single parameter to the Vitrinite Reflectance (Ro). It indicates that Ro is most sensitive to heat flow value (HF) compared to source rock TOC and HI. The lower figure shows the Pareto plots of the standardized measure of sensitivity for multi-way interactions. It indicates a subtle level of sensitivity to Ro. Two main interactions are observed: Low HF-high TOC and high TOC-low HI are identified to be the most sensitive parameter to the Ro responses.