There is little doubt that an increase in subsurface understanding is the key to unlocking the estimated one trillion barrels of additional resources on currently producing assets.
Capturing the knowledge found in all available data – when the data arrives – is central if we want to achieve this task.
Traditionally, reservoir modelling is seen as the answer to this challenge.
Unfortunately, when looking at the historical performance of reservoir model predictions over the past 20 years, we should critically question the value of the current approach to reservoir modeling in oil and gas companies.
Hence, it is understandable that the appetite for alternative approaches to traditional reservoir modelling that potentially can help increase the subsurface understanding is steadily increasing – where machine learning algorithms play a central part these days.
A challenge with these contemporary data driven approaches is, however, their robustness in fields where data is sparse, and/or are collected with low quality/high uncertainty – which is typically the situation in offshore fields.