how-sure · tool 04 · prospect volume as a range · v1.3
How much, and how sure?
A prospect's oil in place as a range, P90 to P10: where the range comes from, which inputs drive it, and which data could change the decision.
Maxim Balagurov, petroleum geologist. In my own work I ran probabilistic volumetrics (P90/P50/P10) using a Monte Carlo spreadsheet; this page shows the reasoning. Oil in place here is stock-tank oil initially in place (STOIIP). P90 is the low case, the volume exceeded with a 90 % chance, and P10 the high case. The example below is made up.
AThe low case is not every input at its low end
Each input's low is a value it beats nine times in ten. All five landing at that end together is far rarer, so their product sits well below the P90. The same holds at the high end.
BA few inputs drive the range
The variance is that of the volume on a logarithmic scale, each input taken alone. Sharpening an input that accounts for a few per cent of it leaves the range almost as wide.
CWhich data could change the decision
Each input is tested as if it could be known exactly. Real data only narrows a range, so this is the most that data on one input could do. Data can only be worth getting if its answer could move you across the line. Before the well, area is what seismic and depth conversion can sharpen; pay, porosity and saturation mostly wait for the well or rest on analogues. Tools 02 and 03 of this set ask the same of a survey and of a drilling campaign.
Try your own prospect
The numbers in full
Limits
The numbers are made up. No survey, well or company data went into the example.
The range is only as good as the input ranges. The page multiplies the ranges you give it; it cannot tell a careful range from a hopeful one. A look-back on past predictions can.
This is the volume if the prospect works. The chance that there is a trap, a reservoir, a seal and a charge at all is a separate number and is not in here.
More limits
Oil in place only: no recovery factor, no gas, no column height tied to a spill point. Area times average net pay stands in for a mapped rock volume.
The minimum volume is set on oil in place to keep recovery out of the page. In practice it is set on recoverable oil.
Two distributions, lognormal and normal, both set by a low and a high value. A skewed fraction or a range with hard ends is not represented. Values below zero count as zero and Sw above 1 as 1; a widened normal range can still reach values that make no physical sense, such as Bo below 1.
Only two pairs of inputs can be correlated, each through one number. The bars and the variance shares in finding B take inputs one at a time and ignore correlation.
Finding C tests one input at a time, as if it could be known exactly. Two inputs known together are not tested.
The decision here is a volume and a chance. Cost, price and timing are left out on purpose.
How it works
Oil in place = area × net pay × porosity × (1 − Sw) ÷ Bo, times 6.2898 to turn km²·m into million barrels.
The page takes 50,000 sets of the five inputs. A fixed pattern, with no random numbers, spreads them evenly over all five ranges at once, so the tails are covered and the same inputs always give the same answer.
The 50,000 volumes are sorted. P90 is the value 10 % of the way up, P50 the middle, P10 the value 90 % of the way up; the mean is their average, and the chance of the minimum volume is the share of volumes above it.
"One at a time" varies one input from its low to its high value with the others at their P50. An input's share of the variance is its swing squared, on a logarithmic scale, over the sum for all inputs.
"If it were known" holds one input at a value, takes the others again and reads the chance of the minimum volume. The value at which that chance equals the required one is found by halving the interval. If the input is correlated with a partner, the partner follows it.
Correlated inputs share part of their position in the range: with a correlation of 0.5, a quarter of one input's variance is explained by the other.
Widening multiplies each input's spread around its P50 and the page then works with the widened low and high values throughout.
Rules it uses
P90 is the low case. The volume exceeded with a 90 % chance, as in petroleum usage; other disciplines count the other way round.
Low and high inputs are P90 and P10 too. They span eight outcomes in ten, short of the extremes.
Lows do not multiply into a low case. Run the draws, or at least do not report the product of the lows as P90.
The mean sits above the P50 when the range leans towards large volumes, as it almost always does when sizes multiply. Swanson's rule, 0.3 × P90 + 0.4 × P50 + 0.3 × P10, gets close to it.
Independence is an assumption. If the large inputs rise together, the range is wider than independent draws say.
Checks:
The checks run in your browser each time the page opens, on fixed cases, whatever is in the form. "By hand" answers come from formulas and arithmetic written out beside them. "Independent calculation" answers come from volume_ref.py: plain Monte Carlo, 4,000,000 draws per case, sharing no code with this page.