How Well Do We Predict Depth?

With more than 3000 hydrocarbon wells drilled, the Netherlands ranks as a highly mature petroleum province. Virtually all well planning relies on 3D seismic which is often of high quality. Velocity model building and time-depth conversion is a key step in all depth predictions. EBN, the Dutch state oil company, conducted an extensive review of the technical performance of the different operators in the country. Interesting observations can be made in the area of predicting reservoir depth. The analysis of 253 recent wells disclosed that at least one third of those wells with disappointing results (i.e. lower well rates, smaller volumes proven) suffer from poor depth prognosis. The depth errors at reservoir level range from − 219 m to +130m. The standard deviation amounts to 38 m which equals 1.2% of the (average) depth. Interestingly, there is a clear bias of 10 m in the depth errors towards being deep to prognosis. The number of wells being deep (64%) is almost double the number of wells (36%) being shallow to prognosis. A possible explanation for the bias is given by a mechanism that can be referred to as selection bias. It is important to realize that we do not have a precise knowledge of the subsurface. Our evaluations, including the depth maps, are the result of seismic interpretation and velocity assumptions which do contain noise. If we would conduct random drilling on these maps and compare actuals versus prognosis, -most likely- no bias would show up. However, in reality we do not drill randomly, moreover we put a lot of effort in selecting our targets carefully. In many cases an important selection criterion is structural height. In those cases where modest hydrocarbon columns are probable (or where the contact is already pinned down) often the planned well is aiming for a crestal position. The depth map available, with its inherent uncertainty, is a key factor in guiding the location picking. Due to the uncertainty, the crestal areas, as expressed on the depth map, are partly genuine highs, partly spurious highs. Selection bias acts equivalent to Darwin’s principle. The ranking of the drillable targets in a portfolio is analogous to the survival of the fittest. Whether the selected location was really crestal (fit) or only perceived crestal; that will show only after drilling. This effect will show up as a statistical bias in the depth prognosis errors.