Jef Caers, Céline Scheidt and Pejman Tahmasebi Rapid updating of conceptual model, trend and reservoir facies model at the appraisal stage Jef Caers, Céline Scheidt and Pejman Tahmasebi
Rapid updating Updating at the exploration & appraisal stage New wells Improved/new seismic No production data yet What is critical ? Do this fast and transparent Uncertainty In this presentation: Updating of properties (not yet structure)
More than just reservoir model updating Model updating is not just about adjusting the reservoir model(s) to fit new data → this reduces artificially uncertainty Model updating requires Verifying whether our current prior assumptions still hold Updating the beliefs associated with interpretations Adjusting individual reservoir models
Example: facies models built from TIs and trend models 3 possible scenarios (deep channels, lobes and shallow channels) Deep channels Lobes Shallow channels 2 vertical trends (absence of trend, presence of trend) No Trend Trend Z
Overview dnew Inconsistent Interpretation Redo TIs/Trend All new Models Incomplete Interpretation Add TIs/Trend Add Models dnew Global insight Confirms Interpretation Update P(TI, Tr) to P(TI,Tr|dnew) Update Models Local update
Approach to Problem 30 models per interpretation 180 models available Before acquiring new data: existing reservoir models Data Interpretations Well Seismic 3 TI 2 Trends New well data: dnew Are the interpretations still valid? Can we “recycle” some of the models for updating?
Use of distances-based scenario modeling Overview Inconsistent Interpretation Redo TIs/Trend All new Models Incomplete Interpretation Add TIs/Trend Add Models dnew Global insight Confirms Interpretation Update P(TI, Tr) to P(TI,Tr|dnew) Update Models Use of distances-based scenario modeling Local update
Use of distances-based scenario modeling Approach to Problem Synthetic well values Existing models New well data Well location Extract well values from existing models Distance between wells Use of distances-based scenario modeling Update P(Tr, TI|dnew)
Probability of TI for a fixed trend and given well data Sequential Approach Use of a sequential approach to compute the probability Trend: provides global information TI: provides fine-scale details information P(Tr = ti, TI = sck |data) = P(Tr = ti |data) * P(TI= sck |Tr = ti,data) Probability of Trend given well data Probability of TI for a fixed trend and given well data Challenges: Definition of a distance to identify trends Definition of a distance to identify TIs
Distance for Trends Synthetic well values The distance should identify models with similar trends Must be a function of the facies locations in the well Use of a Manhattan distance Prior smoothing of the (indicator) well values is applied Smoothing mimics better the nature of the trend Use of moving average - dij = No Trend Trend No Trend Trend P(Tr|dnew) 1.00 0.00
Distance for TIs The distance should identify models with similar scenarios Patterns in the well can provide information about scenarios Use of Multiple Point Histogram (MPH) Analyze of the patterns in the well Use of JS divergence for the distance Method of measuring the similarity between two probability distributions
Distance for TIs No Trend (100%) Combination of probabilities Deep channels Lobes Shallow channels Deep Ch. Lobes Shallow Ch. P(TI|Tr = 1, dnew) 0.85 0.15 Combination of probabilities No Trend Trend Deep Lobes Shallow P(TI,Tr|dnew) 0.85 0.15
Renewed global insight Old global insight New global insight 30 30 5 No Trend 35 models for updating dnew Trend No new model is created Only “local” information at the new well is used Updating the beliefs associated with interpretations Adjusting individual reservoir models
Methodology Based on the previous study, 30 lobes and 5 channels models will be used for local model updating For each model, a region around the well location is updated (i.e. re-simulated) Use of CCSIM for updating the models (fast!) For each scenario, one of the “old models” is selected as the pattern database. The size of updated region is an important aspect which can affect the CPU time and uncertainty space We will study this aspect
Which region size around the well should be used? ? Simulation Grid Update Region Uncertainty space will increase when increasing region We study the impact on spatial uncertainty of the updated model due to choosing a certain region size We do not (yet) propose a method for choosing such region
Updated Lobes model CPU time=45(s) Model 1 Model 2 Old model Old model New model New model Model 1 Model 2
Updated Channel model CPU time=50(s) Model 1 Model 2 Old model New model New model Model 1 Model 2
Investigate update using ensemble average (EA) EA of Lobes Scenario Excellent conditioning Due to having larger objects in the channel models, the region size is larger than lobes model. EA of Channel Scenario No boundary artifacts
Different used update regions Larger regions increase the CPU time Do we need to re- simulate the entire grid? How does the uncertainty space change with different region sizes? Region 1 < Region 2 < Region 3 < Region 4 < Region 5
Combined MDS plot for different resolutions in ANODI analysis MDS plot representing between updated model variability Lobes Models large Region 5 Region 4 Region 3 Region 2 small Region 1 The uncertainty spaces are different for each scenario Uncertainty space for channel model is larger than lobes. Channel Models
Conclusions A rapid model updating in 3 stages Validation of existing interpretations (global) Updating of beliefs (global) Adjusting consist reservoir models (local) Initial results of a larger project (PROFIT/ENI) Including process models as TIs Updating of process model assumptions and parameters Extending to “new” seismic data