Geoscience internships

Two measurements can support different predictions between them

Fictional one-dimensional transect: a reading at coordinate 0 metres has value 10, and a reading at 10 metres has value 30. The values are arbitrary measurement units. Estimate the reading at 4 metres using only those two observations.

Nearest-neighbour estimation uses the closer observation at 0 metres, giving 10. Linear interpolation gives 10 + (4/10) × (30 − 10) = 18. Both exactly reproduce the two supplied endpoints, but they disagree at the unobserved point. Matching the data used to construct a model does not choose between them.

Use an observation that was not part of the fit

Now reveal a previously held-out measurement at 4 metres: 16. The nearest-neighbour absolute error is |10 − 16| = 6; the linear-interpolation error is |18 − 16| = 2. Linear interpolation performs better at this held-out point.

That one comparison does not prove the entire transect is linear. A boundary between sites, another physical process or measurement uncertainty could change the pattern. Test other withheld locations and show where no observations exist rather than hiding gaps under a smooth surface.

At 20 metres, extending the same line predicts 50. This is extrapolation beyond the observed 0–10-metre range, not interpolation within it. Label the extra assumption. For a mapped dataset, document coordinate reference and units before computing distance; numbers from different coordinate systems cannot simply be treated as points on this same metre-scale line.

• Training observations: (0 m, 10) and (10 m, 30).
• Held-out observation: (4 m, 16); do not use it to construct the estimates you are comparing.
• A claim about error across a region needs more evidence than this single held-out result.

Use the figure to explain where evidence is missing

An application note could say: Both methods fit the endpoints; their predictions differ between them. Linear interpolation had the smaller error at one held-out location, but the sample is too small to establish regional accuracy. I would compare additional withheld locations before choosing it for the full dataset.

Use your own project to explain which observations would be most informative next. For an interpretation role, discuss the physical alternatives; for a coding role, show reproducibility and checks. Ask whether the intern collects data, processes it, interprets it or builds tools for someone who does.

Before you apply

Eligibility

Match your degree and methods to the actual geology, geophysics, computing or engineering assignment. Eligibility for one technical route does not cover every geoscience employer.

Timing

SLB's early-career route is explicitly a future-matching application. Record a confirmed project and start date separately when they are supplied.

Location

Clarify whether the assignment involves field collection, laboratory work, interpretation or software. Confirm any field travel and training.

Pay

Obtain placement-specific compensation and travel terms. An academic field course and a paid employer internship are different arrangements.

Sources: SLB geoscience and petrotechnical internships · Chevron internships

Job titles to search

Geology or geoscience intern

Ask which observations and geological or environmental question you would interpret.

Geophysics intern

Ask which measurements, processing methods and physical assumptions the work uses.

Geoscience data or software intern

Ask whether the output is a checked dataset, processing tool or model used by specialists.

Where to look

Search routes, not confirmed vacancies.

Compare the work and try an example

Use a role comparison or worked exercise to choose your preparation before tailoring an application.

• Energy trading, scheduling and risk internships: distinguish the work
• Midstream commercial versus engineering internships
• Utility internships beyond engineering: finance, contracts and project work