Uncertainty in reservoirs 1 - Classification: Internal 2010-05-27 Deepwater Horizon – Gulf of Mexico The slightly more mundane situation I consider: –We have a hydrocarbon reservoir –We have a model for the reservoir which will be used for future decisions. –The parameters in the model are uncertain. What do we do with the uncertainty? Operational uncertainties are unfortunately not a topic in this presentation. 2 - Classification: Internal 2010-05-27 Uncertainty in the petroleum industry Organisational issues: – Strong financial inertia to ”stick with the truth”. – Tradition for compartmentalized organisations where uncertainty information is not passed on. – ”What happens happens” – limited tradition to reevaluate uncertainty estimates. Current topics: – Choice of parameters – model selection. – Different scales. – Stochastic modelling. 3 - Classification: Internal 2010-05-27 $€£ $€£ $€£ $€£ $€£ Maybe the reservoir is larger? Producing fields: Or smaller? There is so much money, financial regulations e.t.c. in these questions that there is a strong organisational urge to just ignore the uncertainty. 4 - Classification: Internal 2010-05-27 An organisational challenge Structural model OK; I pass my best result on to Deborah! 5 - Classification: Internal 2010-05-27 I am working hard to interpret the seismic and build a structural model. Geological model I am doing flow simulations, and I pass my best management even wants I am creating a Welluncertainty - I’ll try outestimates these effort on to geological different values for a Phillip. days model. couple of parameters and see what happens. ”What happens happens” 1. We do our best to model and quantify uncertainty. 2. We make a decision to e.g. drill a well: – Estimated oil volume: A +/- B – found nothing! – Estimated gas volume: A +/ B – found both gas and oil. 3. The new information is used to infer that we were just wrong. Uncertainty estimates are not really challenged. 6 - Classification: Internal 2010-05-27 History matching – it is just plain stupid Traditionally History Matching is percieved as an optimization problem – a very problematic approach: –The problem is highly nonlinear, and severely underdetermined. –The observations we are comparing with can be highly uncertain. –The choice of parameterization is somewhat arbitrary – we will optimize in the wrong space anyway. 7 - Classification: Internal 2010-05-27 Geological concept Channel system Deep marine The choice of geological concept is an example of a choice which will have a Shallow marine profound effect on subsequent interpretations, and decisions. 8 - Classification: Internal 2010-05-27 Good agreement between model simulation and observation! Model/parameter selection II Two wrongs do not make a right – it is all to easy to get ”a match” for the wrong Maybe the ”real” reason was that the reasons: oil-water interface was shallower? Water rate Oil Water Time –Simulations show to little water. –Increase relative permeability of water 9 - Classification: Internal 2010-05-27 Different scales 4 Geo object 2 5 1 ~50 m ~0.25 m 10 - Classification: Internal 2010-05-27 ~10 m 3 Different scales II Pores Reservoir ~ 9 orders of magnitude 11 - Classification: Internal 2010-05-27 Different scales III: Upscaling 1 Permeability: 1<< 2<<3 1 Vertically: T 1 1 1 2 1 3 2 1 1 3 T 1 2 3 3 Horisontally: 1 2 3 12 - Classification: Internal 2010-05-27 ~3 ~ 1 Different porosity realisations Geostatistics It is quite common to sample properties like permeability and porosity stochastically – with various constraints/trend parameters: Spatial gradient Point measurements Correlation length 13 - Classification: Internal 2010-05-27 Modelling – the full loop Sample geostatistical parameters Ideal approach: Sample a geological realisation according to the parameters. Perform flow simulations and evaluate misfit. 14 - Classification: Internal 2010-05-27 Make all alterations Traditional approach: on geo parameters, and keep everything 1.syncronized. Cutting the link to geostatistical paramaters. 2. Direct updates of the properties of the realisation McMC and stochastic modelling – attempt 0 The geo modelling process is not a closed form PDF; it can only be observed from the created realizations. We have tried to update update geo parameters; initial attempts show some success! Uncertainty: Cdd C0 CC CS 15 - Classification: Internal 2010-05-27 Example – channel direction Oo Prior: θ~100 Posterior: θ~0 Conditioning the distribution P(θ|d) with McMC 16 - Classification: Internal 2010-05-27 Main challenges 1. Model selection – and how to handle the ”Uknown unknowns”. 2. Conditioning of coarse parameters like geostatistical trends. Thank you! 17 - Classification: Internal 2010-05-27