Stratigraphic correlation: Old concepts inform new approaches

Speaker: Zoltan Sylvester, Ph. D., Research Scientist, Bureau of Economic Geology, Jackson School of Geosciences, The University of Texas at Austin
While automated stratigraphic correlation techniques are widely used in workflows for interpreting seismic reflection data, geophysical well logs and other time series data are still largely correlated manually. This is a time consuming and painstaking process, despite the fact that datasets with a dense well coverage are prime targets for automation using machine learning approaches. In this talk, I first want to revisit and visualize the basic ideas about how time is recorded - or not recorded - in sedimentary deposits and then show how chronostratigraphic diagrams can be used as key elements of an automated workflow for correlating geophysical well logs. We have written a software package called ‘ChronoLog’ to apply this workflow to datasets with relatively simple stratigraphy; in such settings, the results tend to match or outperform manual interpretations. Further work will focus on improving correlation results when major unconformities or onlap surfaces are present.

[@Belttonne1]

I am extremely enthusiastic to see such progress in the advances of sequence stratigraphy. You have really showed how important the stratigraphy is. It is purely science. I could not imagine the amount of time you invest normalizing the well log in open hole candidates selected in your scope of work, in this case the gamma rays (GR). I just noted in some of the slides you mentioned the use of GR, it depicts shale volume (VSH) If you did not consider the normalization of the GRs but you calculate VSHs one by one it is valid to affirm they are normalized. Similarly, the time spent in this calculation should be worthy. Once it is done, I could not image to use all wells to reveal the stratigraphy similarly to your work. Good question for you is: how do you deal with the measurement of such GRs coming from different source of measurements? for example wireline vs logging while drilling. who contributed to your work with such filter making sure than similarities or differences in correlating are preserved between both, and anomalies are removed to avoid any artifact in a form of systematic error might ended in not geological significance, for example in term of thickness variability? Avoiding any systematic error. Incredibly good starting point, I like the cross section along your field of reference. It is necessarily inductive, and this inductive form of correlating is just nothing if it is not merged with the deductive one plus certainly the rest of information linked to the core observation since usually, it should not talk about fluvial channels or carbonate deposits at just purely looking at the GRs but once it has interpreted the existing cored wells from the appraisal wells campaigns certainly before such any full field development. I think you did not mention the well spacing, could be interested to know, however, at the first glance it looks the field in the sub-basin location has too many wells and some areas with still not such population. Be careful with simplification in similarities with just only the well logs correlation while generating synthetic because it needs to be couple with the 3D seismic observation and interpretation to avoid local discontinuities that shows for example erosion, shortening or stretching of thickness (particularly in single reservoir units) case of sedimentary section near to basements or major faults. If it belongs to passive margin deposits, it is a call of be careful in active margins where tectonic pulses might induce drastic changes in those GRs in other regions- Basins. Outstanding explanation from the Barrel and Wheeler works on the Jurassic tank experiment you domain. It looks interesting and technically a challenge. What is the software you have used? Could you elaborate a little bit more on it, in detail? Regarding your call for automation because certainly you believe you have many wells, once again your automation in supporting a well picks or correlation you fill with some colors necessarily need a combination with geological creativity particularly while mapping. For example, in a fluvial system you does not have only channels or floodplains you might have other features or sub paleo depositional environments cannot be mapped out only with your nice correlation approach that again for me look like more inductive than interpretative with the geological creativity of realistic geological concepts might applicable in that area of interest. You do not need to drill a well in an unexpected sub environment where not commercial number of hydrocarbon spots can be found or where there is a disruption in the sedimentary section. Right? If you can predict it, you do not need to drill wells and then learnt from the bit specifically at such mature field development stage where you might have giant field with well separation of a kilometer. You can save drilling cost instead! Every kilometer distance separation you might find unexpected surprises particularly in carbonate rocks where I invite you to not generalize. Yes 60 km distance between brownfield or green field it is not the same than just one, but the geology offers you always surprises at shorter distance too as you might know after interpreting different scales of observation Eventually, It is impressive work, keep going, and best wishes in your endeavors. Hope you can share all specific steps you do in the open-source free softwares you are using because I am interested to learn. Thank you for educating me. Excellent lecture.