Geostatistical Predictive Analytics is a crucial part of the Multi-Measurement Interpretation (MMI) methodology. The term Predictive Analytics itself is a bit of a buzz term that can be used to describe any approach to data mining to predict trends or behaviors in the effort to identify risks and opportunities.
Among the various approaches to Geostatistical Predictive Analytics are Static Uncertainty Analysis and Correlative Predictive Analytics (CPA) – each deals with a multitude of input data where a toolkit of techniques is applied to produce useful outputs, such as sweet spot maps. At NEOS, we execute these two approaches to mathematically and objectively identify the most relevant geo-measurements that align with the most productive wells or with the location of known wells. In essence, it is this data that provides that key additional insight to the customer, allowing them to make the all-important big decisions.
But what specifically are Static Uncertainty Analysis and CPA? How do they differ and when does NEOS apply each? We’ll start with Uncertainty – stay tuned to Sweet Spots blog for more on CPA.
Quantitative Risk Analysis
We are each faced with uncertainty every day; by definition this means we encounter more than one possibility in many different situations. Who will win the World Cup? While uncertainty is everywhere, only those states of uncertainty where there are personal stakes involved include risk (such as a potential loss or catastrophe). Should I buy this stock? It is this risk that drives individual need to make an educated decision.
To better understand this concept, consider weather. Weather is uncertain. Each day quantitative risk analyses (aka – weather forecasts) are generated for you to make various decisions about how to approach your day. Quantitative risk analysis can be performed a couple of different ways. One source might predict the forecasted temperature for September 15th in Pittsburgh deterministically (with single-point estimates) while another might do so probabilistically (with a range of possible values). A deterministic outcome would give a single value, say 63F with no rain. A probabilistic outcome would give a distribution of temperatures, with a bell curve peaking at 63F and percentage probability (30%) of rain.
We know that the deterministic answer (no rain) is possible, but we understand that there is uncertainty in the forecast and we are accustomed to hearing terms such as “chance of rain” or “probability of precipitation”. While a deterministic answer is useful, a probabilistic answer is necessary to make an informed decision.
In this example, there is uncertainty in the estimate, but unless we have a personal stake and decision to make (we will be in Pittsburgh on September 15th and we need to decide whether to bring a raincoat) that uncertainty does not translate into a risk.
Uncertainty exists in oil & gas exploration, where there are high stakes and [often] high risk, typically as a result of limited available data and information. This perfect storm scenario makes the uncertainty analysis necessary for operators to make those all too important decisions of where to explore, lease or drill.
“When data is sparse and uncertainty if high, that’s the best time to model things probabilistically.” – Patrick Leach
NEOS takes both a deterministic and probabilistic approach to decision making in the presence of uncertainty, beginning with a single common base case conceptual model and ultimately creating a uncertainty assessment that addresses questions, intended for integrated team discussion, like what is the total volume of oil, what are the areas of interest for oil and what may contribute to uncertainty in the area.
As an example, the Oil in Place (STOIIP) model below defines how oil volume is calculated for a formation. While by no means the only important consideration, in place volume estimates and uncertainty models are the foundation of many decisions one has to make in the life of an oil or gas field. For each input parameter in STOIIP, we define an uncertainty model through a set of probability distributions or more complex spatial uncertainty models. Through Monte Carlo Simulation, the input uncertainties are combined into a model of uncertainty for STOIIP.
Global uncertainty in STOIIP can be represented by its distribution, typically normalized per unit area (square miles in this case).
Local uncertainty can be represented through different maps, such as standard deviation maps, P10/P50/P90 maps or probability of being above a given cutoff.
Finally, it can be useful to look at the contributions of the different sources of uncertainty with tornado charts.
As with our weather example, while a deterministic approach (STOIIP) is useful, including a probabilistic approach is necessary to make an informed decision, which can range from leasing an area, selling it, drilling exploration wells, designing data acquisition surveys, to development strategies.
This means that a good understanding of our clients’ objectives and in particular the decisions they are facing is crucial in order for NEOS to deliver information that helps make those all-too-important informed decisions.