Computational Science

An Empty Species Record Is Not Necessarily an Empty Habitat

2026-09-10

A blank location in a biodiversity map may indicate no observation rather than no species. A computational investigation should establish whether survey effort and repeat visits allow detection to be separated from occupancy. Otherwise, a convincing habitat model may mostly learn where people looked.

The computational starting point

Start with existing survey-event records rather than assuming occurrence-only points contain absences. GBIF's survey-data guidance describes effort and event structure. Inspect repeat visits, taxonomic consistency and geographic precision. Respect sensitive-species restrictions and avoid publishing precise locations that could create risk for vulnerable populations.

Sources: GBIF survey-data publishing guide.

Where intuition enters

An intuition that a habitat boundary has a hidden ecological cause becomes a prediction about occupancy after accounting for observation opportunity. The rival is a survey boundary, such as access, observer preference or reporting changes. This is a different question from simply maximizing classification accuracy on a map.

A test that can disagree

Define eligible survey units and distinguish explicit non-detections from missing records. Where repeat surveys exist, compare observation-process and habitat components separately. Candidate vegetation covariates can come from documented MODIS products, but their resolution and meaning must match the ecological question rather than supply a decorative environmental layer.

Hold out geographic blocks and survey organizations where possible. Compare the proposed habitat signal with an effort-only baseline. Simulate imperfect detection to see whether the analysis invents ecological absence. If records lack the structure needed to identify detection, label the result as occurrence-pattern modeling rather than occupancy estimation.

Sources: NASA MODIS vegetation indices.

An illustrative decision

Suppose a model predicts absence beyond a road network, but survey effort also drops sharply there. The computational result should not declare a biological boundary. It may instead show that the available archive cannot distinguish access bias from habitat preference, saving the buyer from a misleading geographic interpretation.

What the research would deliver

The deliverable would be an observation-process audit, a justified target variable and a transfer assessment. A conservation analytics buyer can use it to choose a defensible research question using existing data. The work does not require field expeditions, new animal studies or disclosure of sensitive species coordinates.

Questions this raises

Can occurrence-only data still be useful?

Yes, but the estimand and assumptions differ. They should not silently become a dataset of confirmed presences and absences.

Does a high map accuracy resolve sampling bias?

No. A model can predict the observation process extremely well while misrepresenting the ecological process of interest.

Sources and their limits

Prepared with AI assistance. The linked sources support the specified technical points; they do not validate applied psionics as a whole or guarantee a result for a client.

Read the editorial and evidence standard.

Continue reading

Explore Scientific Oracle consultingfor a scoped review of an existing-data research decision. Start with a non-confidential outline of the question, available evidence and the decision it needs to inform.