Microseismic Monitoring: from descriptive to prescriptive

By Peter M. Duncan
Founder & Co-Chairman

It seems to me that microseismic monitoring of hydraulic fracture reservoir stimulations has topped out after growing for the last 10 years or so. I believe that today something much less than 10% of the wells frac’ed are being monitored. Monitoring is seen as “nice to have” rather than “need to have”. This is an issue of value proposition. Microseismic event location plots, pictures and movies have proven useful for detecting mechanical issues in the completion, for detecting the reactivation of pre-existing faults and joints and for mapping out the morphology of the stimulated rock volume (SRV), but analysis of the data has been largely qualitative. Issues related to event location uncertainty, event detectability and the complexity of the geomechanical processes that govern the frac’ing have led to inconsistent and sometimes contradictory results when the analyses have been made more quantitative. For example, microseismic derived SRV’s usually produce estimates that greatly exceed the actual production. If the technology is to gain wider acceptance and application, a more rigorous quantitative interpretation workflow must be developed, one that is validated with outside, independent data. Furthermore, we need to get beyond simply describing what the stimulation achieved and begin predicting how changing treatment parameters will change the treatment results and even prescribing what parameters should be used on any given well to achieve a better completion.

That all sounds pretty neat, but how do we judge “better”? In the past we have used such things as number of events or the areal or volumetric extent of the microseismic cloud as a measure of success, but of course the real measure of success is “more hydrocarbons for less dollars”. So I believe that we need to be able to predict production from the microseismic data, and we need the ability to predict how changing the treatment parameters will change that production so that we can prescribe the best set of parameters or at least demonstrate the sensitivity of the production to various treatment options.

At MSI we have been working on just such a workflow and have actually gotten to the point where we can estimate production from a model solely derived from the microseismic data. This workflow involves building a discrete fracture network (DFN) model from the microseismic event set, estimating what portion of the DFN gets propp’ed, estimating the permeability of frac’ed system by calibrating the model with historical production. Recently we built such a model from one well of a 4 well set and then successfully predicted the production on the 3 other wells on the same pad. Such a “blind test” is exactly the validation we need to establish the credibility of microseismic data in general and this workflow in particular.

The next step is to be able to predict how changing treatment parameters will change how the rocks respond. A simple way to do this is to calibrate the rock properties of a 2-D frac model with the microseismic data. That is, we adjust the rock properties until the modeled treatment overlays the microseismic cloud. Then we can run the calibrated model with new treatment options and catalogue the results. However, we, and most other people, recognize the shortcomings of the 2-D models for unconventional reservoir rocks. New, more geomechanically sophisticated models of fracturing are starting to be available. These too require some sort of local rock and stress property calibration. We are approaching such a calibration by searching for the range of parameters that causes the more complex model to match what the microseismic monitoring data observed at the treatment well. We can test the validity of the predicted DFN by doing a blind test like the one that validated the production prediction, but this time we will test the predicted DFN with the one actually observed at other wells nearby the calibration well. Once we are satisfied with the calibration, we permute the treatment parameters and catalogue how the stochastic model responds. The DFN model that results is submitted to a reservoir simulator and the production differences for different treatment options can be estimated. It’s early days for this workflow but the initial results are encouraging.

In the meantime we have made a huge step forward on getting a more useful and believable analysis product out of microseismic monitoring, namely an estimate of production, and we have moved a little closer to that vision of a world where every frac is monitored.