Research

My current research interests include unconventional reservoirs (shale gas and shale oil), DTS/DAS microseismic studies, induced seismicities, and Big Data analytics (machine learning with Hadoop, Spark, and the graph database Neo4j).

Unconventional reservoirs are hydrocarbon reservoirs that have low permeability and porosity and so are difficult to produce. Often enhanced recovery techniques, such as fracture stimulation or steam injection etc., must be performed, making the process more difficult than a conventional play. You've probably heard the "Shale Revolution" these years. Shale gas and shale oil reservoirs are the major unconventional reservoirs in the United States, which are abundant of the shale resources (see the image upper right). To produce gas or oil from shales economically, hydraulic fracturing, or "fracking", must be applied (see the picture below). During a fracking completion, fluids are pumped into the target formation, or pay zone, to create fractures through which oil or gas flow back during the production. Proppants are also pumped down into the fractures to support them open for achieving a good conductivity. Although this technology has been advancing fast during the past decade, the petroleum industry is still facing great challenges in understanding some fundamentals of the fracturing mechanism and their relation to the production. Our approach is to combine modeling with the data collected from the laboratory and field (microseismic, completion, well logs, production, and lab experiments) in a comprehensive analysis.


Distributed Acoustic Sensing (DAS) and distributed temperature sensing (DTS) are fast developing new technologies in oil and gas exploration and production, especially for unconventional reservoirs. These technologies originated from fiber optical communication and the basic principles are to shoot laser pulses through fiber optical cables while detecting the Rayleigh (DAS) and Raman (DTS) scattering from each sections along the cable. The advantage of DAS over conventional geophones are mainly the far better spatial coverage (almost continuous) so the better imaging ability, the stabler frequency response, and the lower operation cost. DTS also provides a new dimention in monitoring the completion and production. However, there leaves much R&D to be done with these two technologies to fully benefit the exploration and production. We are working closely with instrument manuafcturers as well as oil companies in bringing better and more practical solutions with these two technologies.


Microseismic events are tiny seismic events (in general magnitude 0 to -4) detected during a hydraulic fracturing completion. So far, microseismic monitoring is the only somehow direct measurement of the fracture growth. Poineers in microseismic monitoring have put in great effort processing the data and making basic and advanced interpretation regarding the fracture growth, mainly by applying the theories and techniques of the modern earthquake seismology to the microseismic data. However, microseismic events have their own unique problems that must be solved before any further advanced studies could make sense. These problems include the instrument response and calibration, velocity structure calibration, magnitude determination, absolute and relative location, and source mechanism, etc. We have published a paper on the method of determining the correct seismic moment and magnitude for microseismic events. Currently, we are focusing on many other fundamental problems mentioned above.


Induced seismicity are seismic events caused by human activities, including water dams, geothermal energy development, waste water disposal, and oil and gas production. In 2012 National Academy of Science released a white paper in which some major induced seismic events in the past are listed (see the figure on the right). Since then, more and more seismicity around Oklahoma and Texas caused attention and are suspected to be at least parially related to the oil and gas production. In general, it has been a concensus that pumping large quantity of waste water into underground through disposal wells may induce seismic events, especially in active tectonic areas; while the relative small quantity of fluids pumped down during a hydraulic fracturing completion would not be able to produce seismic events larger than microseismic level. However, recently some small earthquakes of magnitude up to 2 have been detected during and around some fracking jobs. Investigating the mechanisms of microseismic events and the induced larger structural events is very important in controling the possible seismic hazard caused by petroleum production. We are working closely with the industry by conducting broadband seismic experiments around the completion sites and applying the modern earthquake seismological analysis, as well as modern earthquake enginering technology, such as probabilistic seismoic hazard analysis (PSHA), to these events.


Data set size is exploding almost everywhere these days, partly due to the availability of cheap sensing devices working continuously, such as power consumption meters that monitor each major piece of home appliances every some seconds at each of millions of customer houses, or due to the development of new sensing technology such as DAS recording covering the full length of horizontal wells. These Big Data sets become a huge challenge to conventional data processing and analysis systems, such as relational databases. Hadoop has become the most popular Big Data system for handling huge data sets, which is extremely efficient in Big Data ETL (extract, transform, and load). However, mining the Big Data for various business purposes requires some extra tools and a deep understanding in modern statistical theories. Spark, which has a machine learning library MLlib, is so far the best tool for machine learning on Big Data. Moreover, there is also support for S[arl tp call out to external programs in Matlab and R, which enables scientists to tackle problems with data sizes many orders larger than they could before with tools like R or Matlab. We are combining Hadoop, Spark, and the graph database Neo4j to provide practical solutions in geophysical data analysis, optimization of design, and business analysis.


We are conducting a series interdisciplinary studies on the source mechanisms of a broad spectrum of seismic sources, from perf shots, microseismic events, to shallow and deep earthquakes, and to mining and nuclear explosions. Another focus is to improve structure image and seismic location by seismic interferometry and double-difference method.