Uncertainty in Data
- How uncertainty arises
- Modelling and quantifying uncertainty
- Why geostatistics differs from statistics
- How geostatistics describes uncertainty
- Separating what we know from what we do not know
- Distinguishing between uncertainty and measurement
Participants will learn to appreciate the origins of uncertainty and how simple geological concepts can combine to create apparently intractable problems. Then, we see how accurate analysis can impose order again and, paradoxically, we can learn to predict the unpredictable. Through simple examples and simulations we will see that this requires us to make clear distinctions between what we know and what we do not know, and between what we have seen or measured and what we have not yet seen.
Uncertainty in Prediction
- Variability in one measurement affecting accuracy in prediction of another
- Variance within a single data set
- Placing bounds on what we don’t know
- Using these bounds to make better predictions
- Using these bounds to quantify uncertainty in predictions
Simple tools – histograms, bivariate regression – are re-visited from the perspective of uncertainty, and we see how much more useful information is contained in even the everyday tools with which we are already familiar. Then we look at the specific geostatistical tools, starting with variograms, and see how these make the relationship explicit between known and unknown.
- The role of geomodelling
- Modelling the predictable and modelling the unpredictable
- The history of geomodelling tools and the search for balance between different interpretations of ‘best’ predictions
Looking at the entire history of geostatistics, through Petrel, we see how the search for better characterization of uncertainty has driven the development of new tools. We look at cases in which each tool may be considered to give the best prediction. Understanding why each tool has been developed, and under which circumstances each is best, allows us to choose the most appropriate tool for each task.
Using Large and Complex Data Sets
- Multivariate analysis
- Log suites and seismic attributes
The need for better input to geomodelling has resulted in developments in seismic attributes and log interpretation which rely heavily on statistical and geostatistical tools . We look at these developments and try to understand their limitations, then see how statistical and geostatistical tools can place quantified limits on their accuracy and applicability.
Monte Carlo Modelling and Production Statistics
- How Monte Carlo methods work
- Finding the best inputs to a Monte Carlo simulation
- Optimizing simulations with limited data or limited time
- Monte Carlo analysis
We see how Monte Carlo methods work using simple Excel examples before using Petrel to demonstrate how Monte Carlo analysis combines with geostatistics to give insights unimaginable 20 years ago. Finally, the power of Petrel for geomodeling is combined with the understanding of uncertainty developed during the week to create statistics which embody the state of the art in understanding the relationship between geology and geomodeling.
Modeling and quantifying uncertainty
Tools - Variance, regression, histograms
Using wireline logs and seismic attributes
Monte Carlo modeling and production statistics
Geologists, petroleum engineers and geophysicists having input to the building of models in Petrel®.
Competence with Petrel - participant should preferably have taken at a minimum Petrel Fundamentals training course; Basic understanding of statistics.