A good quantification of geological uncertainty is demanded by almost every hydrocarbon development project. Uncertainty is normally presented as part of key project parameters, such as original hydrocarbons in place, recovery factor, and net present value. Petrophysical properties such as porosity, fluid saturation, and permeability are directly used in calculations of these economic indicators. Geology — in terms of facies proportions, geobody size, and geological trends — may not be directly involved in the calculations but is implied in the distribution of petrophysical properties. Therefore, geological uncertainty, which is the uncertainty in facies proportions and so on, should align closely with the uncertainty calculated from petrophysical property models. And conversely, our geological model — lithofacies distribution in particular — should not only be geologically sound but also be adequately described by its petrophysical properties.
Quantifying geological uncertainty requires both a good geological understanding and advanced geostatistical methods. Our geological knowledge and experiences help us considerably in data interpretation and constructing conceptual depositional models. With sufficient wells and seismic information, we normally have good confidence in determining sedimentary environment, facies distribution, and sequence stratigraphic evolution. However, when we transfer our geological understanding from vertical wells to three dimensions, a lot of variability enters our interpretation. Between wells we have to deal with great uncertainty. Facies proportions, size and shape of geobodies, and geological trends are significantly influenced by different geological scenarios. How can we deal with all these sources and types of uncertainty? This is where advanced geostatistical methods come in.
Deterministic models — whose results are uniquely defined by their inputs — cannot reflect our knowledge about where and how much we are uncertain, so we must use probabilistic models. With these practical tools we can construct reservoir models by statistical inference from available data. Multiple sources of data, including core and log data, seismic attributes, regional geological trends, and production data, can be integrated by using multivariate geostatistical methods. The resulting heterogeneous reservoir model will honour input well data and their statistics, especially their spatial correlations. The uncertainty is presented in multiple realizations or equally likely versions of the output.
To obtain a full characterization of geological uncertainty, there are three key sources of uncertainty you should know about:
1. Commercial geomodelling software packages such as RMS, Petrel, and GOCAD take care of well-data distributions, spatial correlations, trends, and random simulation paths, and provide geological uncertainty from the models. Most people use this uncertainty range to represent the full space of geological uncertainty; however, there are two big pieces missing.
2. The second key source is the uncertainty in our geological scenarios. Different scenarios significantly affect our gridding and geomodelling methodology and therefore the final uncertainty range.
3. The third is the uncertainty in input statistics such as the global mean values or percentages of different facies over our area of interest. These are generally treated as constant values in geomodelling, but observations show they change when adding new wells. Varying the mean values can greatly extend the uncertainty range.
Only a hierarchical modelling approach that accounts for all of these sources of uncertainty should be used to obtain a reasonable geological uncertainty range. Otherwise you’re much less certain than you think you are.