4D Seismic History Matching

At present, production is declining from many oil fields on the Norwegian continental shelf (NCS) and the fields are scheduled to be abandoned. The Norwegian Petroleum Directorate (NPD) estimates that more than half of the oil that was located in these fields at the time of discovery will be left behind when the fields are abandoned unless technologies for recovering the oil are improved. One key to recovering the remaining oil is an improved understanding of the location of the remaining oil, and the internal barriers that have prevented it from being produced. Seismic surveys, in which sound waves are sent from the surface to the reservoir and back, provide useful information on movement of oil and gas in regions that are otherwise inaccessible to measurements. Extracting as much useful information as possible from repeated seismic surveys is the subject of this project. Decisions on how best to produce oil from a reservoir (where to drill wells, where to inject water, how much water to inject) are generally made using computer models of the reservoir. To make optimal decisions, the computer models must be calibrated to the observed seismic survey data, as well as available production data. The models should also be capable of predicting reservoir behavior in the future, when conditions have changed or new wells have been drilled. In this project, we are working on methods for calibrating computer models to seismic data so that predictions from the models are more reliable and can be used to recover more of the remaining oil. We have now developed methods and work flows that allow the reservoir models to be calibrated efficiently even when large amounts of data must be assimilated. The methods have been tested on various problems, including data from the Norne Field.

Outcomes: There has been a large improvement in the ability to handle large history matching problems with large amounts of data as a result of this project. Software developed in the project allow users to efficiently apply the new history matching work flows and apply them to seismic data. Forecasts are more robust, and estimates of uncertainty are improved over the previous methodologies. The collaborative relationship with the time lapse research group at Heriot-Watt University is much stronger than it was before the project. Impacts: Industry partners have implemented many of the methodologies developed in the project and have reported that they are using model diagnostics and the prior predictive distribution to evaluate their models of uncertainty before history matching on real field cases. It is expected that these will result in improvements in reservoir recovery and management.

At present, many fields on the Norwegian continental shelf (NCS) are in a tail production phase with facilities scheduled for decommissioning in a few years. The Norwegian Petroleum Directorate (NPD) estimates that 54% of the original oil-in-place in these fields will be left behind at abandonment under current technologies. Some of this oil is still mobile and located in unswept regions of the reservoir, isolated by structural or sedimentary complexity. Repeat seismic surveys have been extensively used in the North Sea to provide information on changes in reservoir regions that are inaccessible from wells. Extracting as much information as possible on reservoir behavior from the repeated seismic data is the subject of this project. Reservoir management and development decisions are generally based on reservoir models. To obtain optimal decisions, it is necessary that the reservoir models are both well calibrated and properly quantify the uncertainty in predictions. For complex reservoirs, in which geologic heterogeneity and compartmentalization control the location of pockets of undrained mobile oil, the models on which decisions are currently made suffer from shortcomings. (i) It is extremely difficult to calibrate models to large amounts of data coming from 4D seismic and production. (ii) The relationship between the data and the model parameters is indirect, so the calibration process is complex. (iii) Important information from geologic models based on cores, logs, and outcrop analogs are often destroyed in the process of calibration to production and geophysical data, losing potential vital geological information on heterogeneity in the process.