Reality Lab ID: PSLE-SCI-REALITY-0498
Wait, what? A wildlife map colours one square dark green and reports occupancy probability = 0.80. A student looks at the square and says, “Easy. The animal lives on 80% of the land inside this square.” Another student says, “No, it means scientists are 80% sure they saw the animal there.” Both readings sound plausible. Neither is a safe interpretation without reading what the model and the map actually define.
This PSLE Science Reality Lab is about a real scientific communication problem: maps and reports that show a probability of species occupancy or presence. Primary 5 and Primary 6 learners do not need to become wildlife statisticians. They do need to learn how to separate a modelled probability, a survey detection, a physical area and a direct observation before making a claim.
The 2026 PSLE Science assessment continues to include interpreting and analysing information, evaluating observations, information and methods, and communicating explanations and reasoning. The 2023 Primary Science syllabus also encourages healthy scepticism and openness to more than one possible explanation when evidence is incomplete. Occupancy maps are a powerful reality-lab object because a coloured cell can look like a simple fact even when it is the output of repeated surveys, imperfect detection and a model.
Quick Answer
No. An occupancy probability of 0.80 does not normally mean that 80% of the physical ground inside one map cell is occupied by the species. In occupancy modelling, the quantity usually concerns whether a defined sampling unit or site is occupied or has the species present during a stated period, while accounting for the fact that a species can be present but missed during a survey.
It is also not the same thing as a direct sighting. A model can assign a high probability of presence to a site even if the species was not detected on every visit, because the model uses repeated surveys, detection information, site characteristics and assumptions. The exact interpretation depends on the study, so read the model definition, spatial unit, time period and uncertainty before translating the number into words.
The Owned Learner Job
This article owns one evidence-transfer job: how to evaluate a scientific map or report that communicates species occupancy probability without confusing probability of presence with area covered, animals counted or direct observation.
It does not re-own biodiversity, habitats, sampling, probability, map reading or model limits as general science topics. Those have wider owners. For the underlying reasoning, route to How Scientific Evidence Works, How to Build a Simple Scientific Model From PSLE Science Evidence, and How to Decide Whether to Measure the Whole System or a Sample.
Field Notebook: The Marsh-Frog Survey
Imagine an original study of a fictional marsh frog. Scientists divide a wetland into six equal survey cells. During each visit, they listen for the frog’s call for a fixed time using the same protocol. A “1” means the frog was detected; a “0” means it was not detected.
| Cell | Visit 1 | Visit 2 | Visit 3 | Simple observation |
|---|---|---|---|---|
| A | 1 | 1 | 0 | Detected on two visits |
| B | 0 | 1 | 0 | Detected once |
| C | 0 | 0 | 0 | Never detected |
| D | 1 | 0 | 1 | Detected twice |
| E | 0 | 0 | 1 | Detected once |
| F | 0 | 0 | 0 | Never detected |
Cells C and F have no detections. Can we immediately conclude the frog was absent there?
No. Perhaps the frog was truly absent. But perhaps it was present and silent, too far from the recorder, hidden by wind noise, or simply missed. A zero in a detection table records no detection under that survey. It does not automatically prove absence.
The Crucial Split: Occupancy and Detection
Occupancy studies often separate two questions that beginners accidentally merge:
- Occupancy or presence: Is the species present in the defined site during the relevant period?
- Detection: If the species is present, what is the chance the survey method detects it on a visit?
The second question matters because animals are not laboratory switches. A bird can be present without singing. A bat can fly outside the detector’s range. A frog can remain quiet during one visit. A plant can be present but hidden by dense vegetation. A camera trap can face the wrong direction at the wrong moment.
USGS research on occupancy modelling was built around this problem: species are often detected imperfectly, so repeated detection/non-detection data can be used to estimate occupancy while allowing detection probability to be less than one.
Observed, Modelled, Claimed
Read an occupancy map in three layers.
- Observed: actual survey records such as detected/not detected, acoustic recordings, captures, environmental DNA results or another stated evidence type.
- Modelled: an estimate or predicted probability generated from those observations together with a statistical model, covariates and assumptions.
- Claimed: a sentence someone writes from the model, such as “the species is likely to occur in this region”.
A strong learner does not treat all three layers as the same object. The map colour may be model output. The field observation may be a sound recording. The headline may be an interpretation. Each layer can be useful, but each supports a different kind of claim.
Why 0.80 Is Not “80% of the Ground”
Suppose a 5 km by 5 km cell has predicted occupancy probability 0.80. The model’s unit is the site or grid cell, not every square metre inside it. The value does not draw a hidden boundary around exactly 20 square kilometres of occupied habitat.
The animal might use only a small part of the cell. It might move across the cell. The cell may contain unsuitable and suitable habitat. The model may define occupancy as presence somewhere in the site during a season. A probability about the state of the site is not automatically a fraction of physical land area.
Why 0.80 Is Not “We Saw the Animal 80% of the Time”
That would be a different quantity. If an observer visited ten times and detected the species eight times, the raw detection frequency would be 8/10. An occupancy probability can be estimated from a model and need not equal the simple fraction of visits with detections.
Keep the denominator visible. “Eight detections out of ten visits” uses visits as the denominator. “Occupancy probability 0.80 for a site” describes a modelled state probability. Similar-looking decimals can answer different questions.
Why 0.80 Is Not a Direct Sighting
A direct sighting is an observation. An occupancy probability is a modelled quantity. A model may combine detections, non-detections, environmental features, known detection behaviour and information from many sites. That can make a probability scientifically valuable, but it does not turn the model into a camera photograph.
This distinction protects you from two opposite errors. Do not dismiss modelled evidence just because it is not a direct sighting. But do not describe model output as though someone physically observed the animal in that exact location at that exact moment.
Map Check: What Does One Cell Mean?
Before reading the colour, find the spatial unit. A cell might be 1 km², 25 km², a wetland, a forest plot, a listening station buffer or another defined site. The scientific claim should stay at that scale unless additional evidence justifies a finer one.
A probability estimated for a 5 km by 5 km cell does not tell you which tree, pond or path inside the cell contains the animal. The map can be useful for broad monitoring while remaining unable to answer a fine-scale location question.
Time Check: Occupied When?
Species move and seasons change. A probability can refer to a breeding season, summer period, survey year or multi-year average. It may not describe today.
USGS data products, for example, can publish predicted occupancy probabilities for defined seasonal windows and specified year ranges. If a map summarises 2017–2022 summer data, a learner should not write “the species is definitely there right now in September 2026.” The date range is part of the evidence.
Detection Check: What Could the Survey Miss?
Ask how the species is detected. Each method has a detection boundary.
- An acoustic recorder detects sound, not silent presence.
- A camera trap observes only its field of view and active time.
- A live-capture survey samples animals that enter or encounter the trap.
- Environmental DNA can provide evidence of biological material without directly observing a living animal at the exact sampling point.
- A human observer can miss small, hidden, distant or inactive organisms.
Detection probability asks a conditional question: if the species is there, how likely is the method to detect it under the study conditions? This is why repeated surveys can be more informative than a single visit.
Alternative Explanations for “No Detection”
- The species is truly absent.
- The species is present but inactive during the survey.
- The sensor or observer did not cover the right place.
- Weather reduced detectability.
- Background noise hid a sound signal.
- The organism was present at another time in the stated occupancy period but not during that visit.
- The observation was misclassified or of insufficient quality.
Notice the reasoning discipline: multiple plausible explanations can survive the same non-detection. The next step is to seek evidence that separates them, not to choose the most dramatic story.
Alternative Explanations for a Detection
Even a positive detection needs provenance. Was the sound correctly identified? Could another species produce a similar signal? Was the observation location accurate? Was the timestamp correct? Was the environmental DNA transported from somewhere else? Reliable studies build methods to reduce such errors.
This does not mean “never trust detections”. It means a scientific record becomes stronger when its method, quality checks and uncertainties are known.
Model Check: Which Variables Help Predict Occupancy?
Some occupancy models use site features such as elevation, vegetation, water availability or climate. If the model has learned that a species is more often found in certain conditions, it may assign a higher probability to similar sites.
That creates an important evidence boundary. A predicted probability is partly supported by relationships learned from surveyed data and stated variables. It is not the same as new direct observation at every unsurveyed cell. The model can extend evidence, but the extension depends on assumptions and how similar the new location is to the data used to build the model.
Uncertainty Check: Is 0.80 the Whole Story?
Scientific probability maps may provide uncertainty intervals as well as a central estimate. Two cells can both show 0.80 but differ in how uncertain that estimate is. A cell with much less survey information may have a wider interval.
If the map gives only one colour, check whether the accompanying documentation provides uncertainty. A single displayed probability can hide important information about how strongly the data constrain the estimate.
Comparison Check: Same Probability, Different Evidence
Imagine two cells both have predicted occupancy probability 0.80.
| Cell | Survey evidence | Model context |
|---|---|---|
| G | Detected on 4 of 5 visits | Many nearby surveyed cells |
| H | No direct visits this year | Prediction from habitat and regional data |
The same displayed probability does not mean the evidence pathways are identical. One is closely supported by repeated local detections. The other may be a spatial prediction. A careful reader asks where the number came from before treating the cells as equivalent.
What Evidence Would Strengthen a High-Occupancy Claim?
- Repeated detections under a documented protocol.
- A detection model that fits the survey method and species behaviour.
- Independent observations from another method where appropriate.
- Similar predictions from nearby time periods or updated surveys.
- Clear spatial and temporal definitions.
- Uncertainty that is reasonably narrow for the decision being made.
- Checks for false positive identification where that risk matters.
What Would Weaken It?
- One brief survey with low detectability.
- Unknown survey effort.
- A model applied far outside the conditions represented by its data.
- Old data used to make a present-tense claim after major habitat change.
- A probability map with no explanation of the spatial unit.
- A colour legend treated as direct animal counts.
- False-positive risk ignored for easily confused species.
Worked Case 1: The 80% Forest Cell
A report says a 25 km² cell has occupancy probability 0.80 for Species K. A student writes, “Species K occupies 20 km² of the cell.”
Repair: The probability refers to the species being present in the defined cell under the model, not the fraction of cell area physically covered by the animal. No 20 km² area can be calculated from 0.80 without a separate area-use model.
Worked Case 2: Three Silent Nights
A bat detector records no calls on three nights. A student concludes the site is unoccupied.
Repair: Three non-detections reduce the evidence for presence only in relation to how likely the detector was to record the species if it was present. If detectability is low, absence remains uncertain. Survey conditions and method performance matter.
Worked Case 3: One Clear Recording
An expert-confirmed species call is recorded at Site J. The occupancy model gives the site 0.62.
Repair: The confirmed detection is direct evidence of occurrence during that observation period. If the model output seems lower, investigate timing, model version and definitions rather than assuming the recording or model must automatically be wrong. They may refer to different periods or quantities.
Worked Case 4: Probability Changed After New Surveys
A map changes from 0.35 to 0.72 after a new season of surveys. Did the animal population necessarily double?
No. A change in occupancy probability is not a direct population-size ratio. New detections, model updates, changed habitat variables or improved information can change the estimated probability without showing that animal abundance doubled.
Worked Case 5: The Darkest Colour
The darkest map colour represents probabilities from 0.75 to 1.00. A student points to a dark cell and says, “This cell is 100% certain.”
Repair: A colour class contains a range. Read the actual value if available. Even a model value near 1 is not a guarantee that every location and every moment inside the cell contains the species.
Worked Case 6: Old Probability, New Habitat
A 2021–2023 occupancy model assigns 0.85 to a wetland. In 2026 the wetland has been greatly altered. Can the old probability be used as current proof?
Repair: The old model is historical evidence. If important habitat conditions changed, current surveys or a model updated with current variables are needed before making a present-tense claim.
Tempting but Invalid Reasoning
- “0.80 means 80% of the land.” Site-state probability is not area fraction.
- “No detection means absence.” Detection can be imperfect.
- “One detection means the species permanently lives there.” A detection has a time and method scope.
- “Higher occupancy probability means more animals.” Occurrence and abundance are different quantities.
- “A modelled map is fake because it is not direct observation.” Models can be evidence-based; read their inputs and assumptions.
- “A probability near 1 is certainty.” Scientific estimates remain conditional on data, model and definitions.
PSLE-Style Transfer Case
Researchers survey five ponds for a frog. At Pond R the frog is not heard during two visits. A model that accounts for survey conditions gives Pond R an occupancy probability of 0.65. A student says, “The frog must occupy exactly 65% of Pond R.” Evaluate the statement.
Strong answer: The statement is incorrect. The 0.65 value is a modelled probability that the frog is present in the defined pond/site under the study conditions. It does not describe the fraction of the pond’s area occupied by the frog. The two non-detections also do not prove absence because the frog might be present but not detected. The survey method, detection probability and model uncertainty should be checked before drawing a stronger conclusion.
Practice Set
1. A map cell has occupancy probability 0.90. Can you say nine out of every ten square metres contain the species?
No. The probability refers to the defined site state under the model, not physical area fraction.
2. A species was not detected once. Is occupancy probability zero?
Not necessarily. The species may have been present but missed.
3. Why repeat a survey?
Repeated visits provide information about how often a present species is detected and help separate non-detection from true absence.
4. Does high occupancy probability prove a large population?
No. Presence probability and abundance are different scientific quantities.
5. What date information should you check?
The survey period, model period and data-update date, because an older probability may not describe current conditions.
6. Why does grid size matter?
A probability for a large cell does not locate the species precisely inside it; claim resolution should not exceed the model’s spatial unit.
7. If two cells both show 0.80, must their evidence be identical?
No. One may have direct repeated surveys while another is largely predicted from broader data and site variables.
8. What would make a non-detection more informative?
A survey method with high known detectability under suitable conditions, repeated sufficient effort and clear protocol.
Delayed Independent Return
After a day, draw three boxes labelled site state, survey detection and map probability. Write one sentence under each explaining what it means. Then draw arrows showing how repeated survey detections can inform a modelled probability without becoming identical to it.
On a second return, invent a site with no detections but high-quality habitat. List two explanations that could survive the evidence. Then state what new observation would help separate them. That is the deeper inquiry habit: not just naming uncertainty, but asking what evidence could reduce it.
Parent and Tutor Teaching Guide
Use coins or hidden cards instead of real animals. Place a card marked “present” or “absent” under each of several cups. On each survey round, allow the learner only a limited chance to reveal a present card. The child will quickly see that “not found” does not always equal “not there”.
Then show a fictional map cell labelled 0.80. Ask the learner to generate three wrong interpretations before writing the strongest one. This reversal is useful because students become alert to denominator and scope errors: 80% of area, 80% of visits, 80% of animals and 0.80 probability of site occupancy are different statements.
Finish by asking the learner to state the spatial unit and time period before interpreting any probability map. If those two pieces are missing, the learner should say what cannot yet be concluded. That is healthy scepticism used constructively, not cynicism.
Routes to Existing PSLE Science Owners
- How Scientific Evidence Works — the wider observation-to-claim owner.
- How to Read “No Evidence” Without Concluding “No Effect” — for the general absence-of-evidence distinction.
- How to Decide Whether to Measure the Whole System or a Sample — for sampling ownership.
- How to Read “Most Likely” and “Best Supported” — for probability-language scope.
- How to Build a Simple Scientific Model — for model construction and limits.
Authoritative Sources and Further Reading
- Singapore Examinations and Assessment Board — 2026 PSLE Science syllabus.
- Ministry of Education, Singapore — 2023 Primary Science syllabus.
- U.S. Geological Survey — estimating site occupancy when species are detected imperfectly.
- U.S. Geological Survey — occupancy models and imperfect detection.
- U.S. Geological Survey — example predicted occupancy-probability data product.
The Quiet Return
A coloured probability map is neither “just a guess” nor a direct photograph of reality. It is an evidence-based representation with a defined unit, method, time and set of assumptions.
When you see 0.80, do not rush to attach “80%” to the nearest visible thing. First ask: probability of what, for which unit, during what time, from what observations, under which model? That question is the real scientific skill.
