Economic uncertainty, volatile oil and gas prices, and frozen capital markets, combined with growing reservoir complexity, are having a significant effect on oil and gas exploration and development, causing a renewed focus on managing risk and reducing costs.
Restricted access to locations, environmental sensitivities, high drilling costs, and problematic production options are all driving the demand for new technologies that increase the probability of success in discovering, delineating, and developing oil and gas reservoirs. At the same time, companies must effectively reduce costs through efficient project execution.
One such technology that addresses this industry need is low frequency (LF) seismic, the spectral analysis of the natural seismic wavefield of the earth between 0.1 and 10 Hz. The methodology uses very sensitive broadband seismometers—not the standard 3C geophones—to directly acquire the earth's low frequency (<10 Hz) seismic background data (Fig. 1). Each instrument station includes a portable, ultrasensitive three-component broadband seismometer, battery pack, a GPS unit, and a hand-held controller.
The LF data are analyzed to study small lateral variations. Empirical observations suggest that multiphase fluids in hydrocarbon reservoirs directly affect these small variations and generate energy anomalies in the earth's ambient seismic wavefield.
In this article, innovations in extracting attributes from low frequency (<10Hz) seismic wavefields are examined. Also, potential applications for frontier exploration as well as in mature field development are considered.
Citing recent case study examples, some of the challenges and developments in LF seismic will also be reviewed with regard to the quantitative integration of LF data with the reservoir's rock properties. Significant advances in processing and interpreting the LF data have been made using classical statistical analysis and predictive noise filtering.
Background
Low frequency, passive seismic data have been acquired at several locations around the world.
In one such case in 2007, an extensive survey was carried out above a tight gas reservoir and an adjacent exploration area in Mexico. Data from several hundred stations with three-component broadband seismometers, distributed over 200 sq km, were used for the analysis.
The hydrocarbon, reservoir-related attributes were calculated, mapped, and compared to the known gas intervals, showing good agreement between the LF attributes and the known hydrocarbon locations. The adjacent exploration area was then mapped for potential hydrocarbon locations.
Wells drilled after the survey confirmed the predicted, high hydrocarbon potential in the exploration area. It is assumed that hydrocarbon reservoirs are partially saturated, whereas the surrounding rocks are fully saturated. Real data observations are consistent with this conceptual model.
Low frequency benefits
Low frequency (LF) seismic analysis produces attributes that describe the variation of the naturally occurring seismic wavefield below 10 Hz.
A growing number of surveys over different oil and gas fields throughout the world have established the presence of spectral anomalies in the earth's ambient seismic wavefield—microtremors—with a high degree of correlation to the location of hydrocarbon reservoirs.
Current hypotheses state that these anomalies may be directly related to the fluids inside the reservoir structure rather than to the reservoir structure itself. An analysis of the anomalies can therefore be used, together with other reservoir data prior to drilling, as an indicator for optimizing well placement during exploration, appraisal, development, and field extension, or reservoir management.
Spectraseis believes that a coherent explanation of the rock properties of the reservoir system can be developed using low frequency data, an uncommon measurement in traditional seismic surveys. Furthermore, it is expected that further analysis of reservoir data will explain the field observations that have been collected over the last decade around the world, linking reservoirs to low frequency seismic energy anomalies in the frequency domain.
The potential benefits to the oil and gas company from low frequency seismic are compelling. There is an increased probability of success in defining hydrocarbon zones prior to embarking on a drilling program, with fewer dry or nonproductive wells, lower drilling costs, and reduced exposure to health, safety, and environmental risk.
LF seismic also opens up opportunities that were previously considered "unexplorable." LF new technology developments allow the economic development and production of reserves that were thought to be uneconomic due to physical accessibility, field size, and-or geological complexity.
LF data provide information on the subsurface fluids in complex geological settings and can be economically acquired in relatively small areas that are uneconomic for traditional seismic and other surveys prior to drilling.
By integrating LF seismic results with other subsurface data, companies can better develop plays hampered by poor seismic imaging and target stratigraphic traps that can be difficult to map with traditional seismic alone. In this way, operators can segregate large areas according to high or low hydrocarbon potential, and manage their portfolios accordingly.
LF seismic also comes with a light environmental footprint with limited resource and HSE requirements. There is no need for external sources, such as explosives or vibrators, nor large infrastructure, such as cables and transportation.
Due to larger spacings between sensors and lighter equipment, smaller crews can be used to rapidly survey potential leads within a large concession. For example, LF seismic data can be acquired over 400 to 500 sq km in less than 20 days, with crews of less than 50 people in total. Such light equipment and limited manpower requirements can be equally valuable in remote and environmentally sensitive areas where costs and safety risks escalate rapidly.
Fig. 2 shows a typical survey design for a prospect-ranking application. In this particular case, the layout covers 310 km, to be acquired with 40 sensors over 12 days.
In addition, the lower cost and logistical ease in acquiring LF seismic versus traditional seismic, can be particularly valuable for operators who are focusing on fields that are close to populated areas. California is one such example where many of the state's fields are too small for the major oil companies but with the extent of the residual reserves better defined might prove economic for independent oil companies prior to committing new investment.
Furthermore, some of the oil fields are unpredictable in terms of production and many of the state's fields are also close to towns, cities, and national parks: areas that would be unsuitable for traditional seismic techniques but where LF seismic is a viable option. Several other states have similar challenges.
LF developments
LF seismic technology, however, like any emerging technology, comes with its processing challenges.
For example, without strong sources like dynamite and Vibroseis units, can the relatively low signal-to-noise ratio from the passive acquisition of data overcome the natural and anthropogenic background noise?
The remainder of this article will examine how these challenges are being addressed by applying classical statistical methods and established noise filtering techniques. Techniques that will demonstrate how LF seismic is rapidly becoming a key technology in frontier exploration are reviewed.
Bayes methodology
Bayesian inference is a statistical inference in which evidence or observations are used to update or to infer the probability that a hypothesis may be true.
Based on the work of the British mathematician Thomas Bayes in the 18th century, Bayesian inference has applications in industrial quality control to discard faulty (vs. nonfaulty) products from a conveyor belt and is being used in both upstream and downstream oil and gas applications.
In LF seismic surveys, the Bayesian methodology captures basic empirical relationships between recorded LF seismic data and the subsurface properties, represented as statistical probability distributions, accounting for both uncertainty and variability (Fig. 3).
Hydrocarbon likelihood
As Fig. 3 demonstrates, an LF survey over a field in Texas was conducted to identify areas with increased prospectivity.
As a first step, a small set of receivers, representative for hydrocarbon (HC) and nonhydrocarbon (NHC) areas, is selected. The attributes of these sets are used to construct HC (green) and NHC (blue) Probability Density Functions (PDFs) over the two-dimensional space shown. In a Bayesian approach, receivers in new areas are classified as HC or NHC by comparing their LF attributes with the constructed PDFs.
The Bayes method then compares new LF seismic observations to the basic statistical relationships and decides what subsurface property best fits the empirical observations. Due to the statistical uncertainty in the empirical models, this decision on what subsurface property is involved comes with a degree of confidence in the form of a probability of hydrocarbon content. The models can then be modified to be consistent with the recorded data. The validity of the resulting models is directly related to the quality and quantity of the LF seismic data acquired.
Key aspects of the Bayesian methodology include the fact that it accounts for the uncertainty of assumed models; uses actual empirical data as well as theoretical (synthetic) models; bases decisions on several subsurface properties while giving easy-to-interpret results; and has the ability to integrate new evidence into existing models (Bayesian learning). Well data, prior knowledge about the geology, and reservoir production data can also be readily integrated.
The process utilized information provided by the statistical distribution of the energy attribute, as opposed to simply the average value of the energy attribute. This generated the results in quantitative, hydrocarbon probability maps that are easier to interpret and more accurate than previously produced hydrocarbon potential maps based on a single (average) attribute value.
Capturing data, negating noise
One of the primary challenges in analyzing low frequency seismic data is the separation of wavefield components that contain information about the subsurface, from surface-generated noise traveling predominantly as surface waves.
Most of the seismic energy that is measured at the surface propagates in the form of surface waves. This "ground roll" noise is also present and can be a challenge for traditional or "actively acquired" seismic data.
Over the past 50 years, the seismic industry has developed various techniques to suppress the unwanted noise from surface waves. Lately, interest has emerged to use the ground roll as a signal source. The detrimental role of surface-generated noise in industry-type passive seismic surveys has been described recently12 and highlights the need for advanced acquisition and processing methods for low frequency seismic data.
A recent case of an onshore project in continental Europe, demonstrates how these advanced acquisition and processing methods can be successfully applied. The survey took place around a small city. The layout of the survey consisted of two lines (a southern and northern line) with 25 stations, spaced 300 m apart, and a maximum line length of 7.5 km. Each station was equipped for continuous passive seismic recording with a buried three-component broadband seismometer, a digitizer, and a GPS unit. An oil reservoir was located approximately in the middle of both lines near the city center.
The LF survey was recorded in a 2-day period over a known oil field in the region as a test of concept for expanded exploration and development use. Careful identification and the removal of anthropogenic or man-made noise sources are necessary prerequisites for the analysis of LF seismic data, acquired passively at the surface.
In the context of anthropogenic noise contamination, the European survey presented a worst case scenario with various sources of noise criss-crossing the survey area. The northern line ran through an industrial quarter and crossed a much-used waterway, while the southern line ran across highways with high traffic volumes.
Fig. 4 shows the average daytime and nighttime noise spectra in the survey area in comparison to the New Global Noise Model. The arrow in Fig. 4 denotes machinery noise at 2.083 Hz.
Intensive data analysis was required to identify and separate various types of anthropogenic noise from the records in order to isolate the signals due to the uncontaminated seismic background wave field. For LF seismic signal analysis, anthropogenic noise can generally be categorized as two source types: (i) broadband transient signals, created by traffic, fauna, explosions, or falling objects, and (ii) stationary sources of narrow bandwidth, created by machinery, running water, or the structural resonances of buildings or bridges.
Transient noise was the most abundant in the European data and was filtered out by a statistical approach. Spectraseis calculated spectrograms using 40-sec time windows with a 20-sec overlap. Fig. 5 demonstrates the removal of transient cultural noise, and Fig. 5a shows the nighttime spectrogram of station 17 before (top) and after (bottom) the muting of transient noise.
For each spectrum representing a 40-sec period, the average power spectral density (PSD) was calculated over a specified frequency band. For each frequency band, the power spectral densities of all periods were then arranged in a histogram. In Fig. 5b, the spectral variance is dominated by transient noise. After data conditioning (bottom), the variance is reduced and the stationary background noise emerges as the lower end member of the spectral variance. Colors denote frequency of occurrence of the respective PSD level.
A bimodal distribution was observed in the histogram, with the higher mode attributed to time periods of transient noise contamination, and the lower mode representing the desired, uncontaminated background.
Transient noise was then removed by "muting" the data in all time periods above a fixed, threshold level in the histogram. For quality control, the spectral variance was examined over the selected time period (Fig. 5c). Fig. 5c is a histogram of average spectral levels between 0.1 and 10 Hz showing a bimodal distribution caused by transient noise contamination. Removing the higher mode of the distribution effectively mutes transients in the data (as compared with Fig. 5a).
An overall reduction in spectral variance was observed and convergence was obtained for the average to the lower level in the histogram, which represented the natural background level, plus stationary noise.
In addition, a frequency-domain despiking algorithm was developed that removes narrow-banded peaks created by stationary noise.
Two attributes were calculated from the clean, despiked data with transients removed. These attributes were used for the quantification of: (i) integration of the Power Spectral Density (PSD) spectrum of the vertical component, over a data driven frequency band, and (ii) integration of the spectral ratio of the vertical and horizontal components (called V/H). The V/H attribute is more robust with respect to transient noise contamination and was the attribute of choice for this survey due to the urban setting.
Spectraseis observed a statistically significant increase in the spectral ratio of V/H, between 1.5 and 3.5 Hz, over the reservoir. Because the lateral variation of attribute values in the anomalous region is larger than the standard deviation, it can be concluded that the observed anomaly is statistically significant. Furthermore, a check of the near surface statics revealed that the observed, anomalous V/H ratio could not be attributed to site effects or noise in the shallow subsurface. It was therefore concluded that observations made from the LF seismic data are in fact, an expression of the earth's modified, background seismic wavefield that is directly related to the fluids in the reservoir structure.
In this survey, careful identification and removal of anthropogenic noise sources from the LF seismic data were achieved.
Hiking exploration confidence
LF seismic technology remains a work in progress.
Spectraseis, in close collaboration with the Swiss Federal Institute of Technology (ETH Zurich), is spearheading the industry's biggest research program focused on the theory, methods, and applications of low frequency spectral analysis.
Through new innovations such as the introduction of Bayesian statistics, the ability to reduce man-made noise, developments in time reverse modeling (TRM) for depth imaging, and the capability to carry out surveys in environmentally sensitive locations, LF seismic is becoming a powerful and cost-effective derisking tool. LF seismic is today a tool with applications across the entire reservoir lifecycle, from exploration bid rounds, to exploration and delineation, through to prospect ranking and development.
Conventional oil and gas has become harder to find and produce economically, leading to an increase in exploration for, and development of unconventional reserves. Due to the geological complexity and expensive production methods typically found with unconventional reserve plays, the growth of independents, the number of "less giant" but potentially economic fields, and operators' relentless focus on reducing costs while managing risks, the emergence of LF seismic technology could not have been more timely.
Acknowledgments
Thanks to Spectraseis for permission to publish, and to Alex Goertz, Nima Riahi, Brice Bouffard, and Rob Habiger of Spectraseis for reviewing this article. Their comments and suggestions were invaluable, although the author accepts full responsibility for any remaining errors. Established in 2003 in partnership with leading European universities, Spectraseis is the principal technology and service provider in the fast-emerging field of low frequency seismic technology. Customers include Petrobras, StatoilHydro, Pemex, other majors and independent operators in the Middle East and North America.
References
1. Hanssen, P., and Bussat, S., "Pitfalls in the analysis of low frequency passive seismic data," First Break, Vol. 26, 2008, pp. 111-119.
2. Nguyen, T., Lambert, M-A., Saenger, E.H., Artman, B., and Schmalholz, S.M., "Reduction of noise effects on low frequency passive seismic," EAGE expanded abstracts, 2009.
The author
More Oil & Gas Journal Current Issue Articles
More Oil & Gas Journal Archives Issue Articles
