A tracking map joins three dots with two straight lines. The lines total ten metres. Did the animal travel ten metres? That conclusion requires more than correct geometry: the dots must locate the animal accurately, and the unobserved movement between them must match the assumed path.
Short answer: the length of a line joining recorded fixes is a property of the recorded map. With exact fixes, it is a lower bound on travel between them. With location error, even that simple lower-bound interpretation can fail. A useful account separates recorded chord length, possible true positions at the fix times and the continuous path between those times.
This guide uses original fictional coordinates and explicitly supplied error bounds. It involves no animal handling, tag deployment or sensitive locations. Return to Wildlife for responsible observation and to Learning That Transfers into Life for the wider learning purpose. The task here is to make a distance claim that the evidence can actually support.
eduKateSG · Wildlife
Find your next learning step
Distinguish mapped line length from animal travel through original location-error bounds, coupled position examples, speed constraints and careful sampling checks.
ROUTE 1 · CHAPTERS 1–5
Define distance and observations
Separate recorded points, true fixes, continuous travel and location error.
ROUTE 2 · CHAPTERS 6–10
Build and tighten geometric bounds
Derive step bounds and solve a coupled ordered record.
ROUTE 3 · CHAPTERS 11–15
Add time and error models
Separate path length, speed constraints, incompatible records and competing errors.
ROUTE 4 · CHAPTERS 16–20
Preserve statistical and sampling scope
Check squared distance, accuracy descriptions, missingness, a new case and speed.
ROUTE 5 · CHAPTERS 21–25
Practise and continue
Compare sampling choices, write evidence and use the connected wildlife routes.
Chapter contents
Define distance and observations · Chapters 1–5
1. Name three different distances
2. Put the coordinates in a compatible metric frame
3. Reconstruct an exact-fix example
4. Explain the triangle inequality in this setting
5. Show how a stationary animal can acquire a mapped distance
Build and tighten geometric bounds · Chapters 6–10
6. Represent location error as a vector
7. Derive a bound for one true step
8. Sum segment bounds without calling them exact extrema
Add time and error models · Chapters 11–15
11. Keep true-fix length separate from continuous travel
12. Introduce a speed ceiling explicitly
13. Use incompatibility to question the model, not accuse the animal
Preserve statistical and sampling scope · Chapters 16–20
16. Check a squared-distance identity independently
17. Do not convert a reported accuracy statistic into a hard boundary
18. Keep screening and missingness visible
Practise and continue · Chapters 21–25
21. Compare nested samples without declaring an optimal interval
22. Build a distance evidence note
23. Choose repair, stabilisation or extension
CHAPTER 1 OF 25 · Define distance and observations
1. Name three different distances
The first distance is the recorded polyline length: add the straight-line distances between consecutive recorded coordinates. It can be reproduced from the file once the coordinate system, order and missing-data rules are known.
The second is the polyline through the animal’s true positions at those same fix times. Those positions may differ from the recorded ones because the instrument’s locations have error. A sum calculated exactly from the file need not equal this second quantity.
The third is the length of the animal’s actual path during the full observation window. Between fixes, the animal might turn, follow a curved route, remain still or return towards a previous point. A straight line between endpoints is an interpolation choice rather than a continuous observation.
Write these as S_recorded, S_true-fixes and L_path. The notation is less important than the separation. Asking whether a tracking distance is correct becomes a more useful question: correct as which of these quantities, over which window, under which assumptions? This prevents an apparently precise map statistic from quietly becoming an unsupported biological measurement.
CHAPTER 2 OF 25 · Define distance and observations
2. Put the coordinates in a compatible metric frame
Our teaching points use a flat Cartesian plane measured in metres. The distance between (x₁,y₁) and (x₂,y₂) is √[(x₂−x₁)² + (y₂−y₁)²]. The coordinate differences have length units, so the result does too.
Latitude and longitude in degrees cannot be inserted into that expression and reported as metres. Real geographic work needs a suitable coordinate reference and distance method, with attention to scale and distortion. A method appropriate to a small local analysis need not be appropriate to a continental migration.
Record whether the question concerns horizontal distance, three-dimensional travel or distance along a constrained route. A horizontal track can omit vertical movement. An altitude estimate brings its own error and datum questions. Adding another coordinate does not automatically make the result more accurate.
For the original examples below, none of those geographic choices remains hidden: the plane is fictional, units are metres and no real terrain is represented. That lets us isolate the effects of location error and unobserved movement without claiming to have solved map projection or species-specific movement modelling.
CHAPTER 3 OF 25 · Define distance and observations
3. Reconstruct an exact-fix example
Suppose an animal’s positions are supplied exactly at three times: A = (0,0), B = (3,4) and C = (6,0). The first straight distance is √(3²+4²) = 5 metres. The second is also 5 metres. The polyline length is therefore ten metres.
The straight distance from the start to the finish is six metres. It measures net displacement magnitude, not the length of the two-segment route. The route passes through B, four metres away from the start-to-finish line.
If the animal travelled straight from A to B and then straight from B to C, its path length is ten metres. If it took detours between the same points, its path is longer. The fix record does not show which history occurred unless additional path information is supplied.
The defensible exact-fix conclusion is L_path ≥ ten metres for that window. Equality requires an appropriate straight-between-fixes history. Calling ten the actual travel distance adds that path assumption. Calling six the total travel distance discards the known intermediate position. Both mistakes can arise even when every square root was evaluated correctly.
CHAPTER 4 OF 25 · Define distance and observations
4. Explain the triangle inequality in this setting
The shortest unconstrained route between two points in the supplied plane is their straight connecting segment. Any path through additional points is at least that long. This is the geometric reason an exact-fix polyline gives a lower bound on continuous travel.
For a third point B, distance(A,C) ≤ distance(A,B) + distance(B,C). Applying this relationship repeatedly shows that adding an exact intermediate point cannot shorten the polyline. It either leaves the length unchanged or reveals additional turning.
The conclusion concerns the same underlying path and nested exact observations. It does not say that any larger dataset from another deployment must have a longer distance, nor that noisy added locations improve an estimate. The words exact and nested do real work in the argument.
A useful learning exercise begins with A and C alone, giving six metres, then reveals B and raises the lower bound to ten. Ask what changed: the animal’s historical journey did not change, but the evidence about it did. This distinction helps students interpret revisions to a scientific estimate without treating every revision as a contradiction or a failure.
CHAPTER 5 OF 25 · Define distance and observations
5. Show how a stationary animal can acquire a mapped distance
Now consider an animal that stays at true position (0,0) throughout a short fictional record. The instrument reports (−0.5,0), then (0.5,0), then (−0.5,0), all in metres.
The recorded polyline has two one-metre segments, totalling two metres. The actual path length is zero. Every recorded position is within half a metre of the true position, so the record is compatible with a stipulated half-metre location-error bound.
This example directly rejects the claim that a recorded polyline must always be a lower bound on actual travel. That statement was justified for exact fixes; it is not automatically justified after the fixes acquire error.
It does not follow that every short segment is noise or that an analyst should delete all small movements. Another animal could genuinely move between nearby points. The same recorded scale can contain biological movement, instrument error or both. A filtering rule needs its own purpose, evidence and evaluation. The paper example exposes an ambiguity; it does not prescribe a universal distance threshold for real wildlife data.
CHAPTER 6 OF 25 · Build and tighten geometric bounds
6. Represent location error as a vector
Let yᵢ be the recorded position and xᵢ the unknown true position at fix i. Write yᵢ = xᵢ + eᵢ, where eᵢ is the location-error vector. Consecutive recorded displacement is then yᵢ₊₁−yᵢ = (xᵢ₊₁−xᵢ) + (eᵢ₊₁−eᵢ).
The difference between consecutive errors matters. A common offset applied to every fix changes absolute placement but cancels from every displacement. Alternating offsets can create large apparent steps even when the underlying animal is stationary.
This is why a single statement about typical location accuracy is insufficient to determine distance error. The sizes, directions and dependence of successive errors all matter. A map can be displaced systematically while preserving local step lengths, or appear well centred overall while its zigzags inflate the polyline.
Keep absolute location and relative displacement questions separate. A shared offset may cancel for travel distance yet still place the animal on the wrong side of a habitat boundary. The existing habitat-use guides examine those classification and availability questions. A correction suitable for one target quantity should not silently be treated as a solution to every ecological interpretation.
CHAPTER 7 OF 25 · Build and tighten geometric bounds
7. Derive a bound for one true step
Suppose the teaching problem guarantees ||eᵢ|| ≤ rᵢ and ||eᵢ₊₁|| ≤ rᵢ₊₁. These are hard geometric bounds in the supplied model. Let dᵢ be the recorded straight distance between the fixes.
The error difference has magnitude at most rᵢ+rᵢ₊₁. Applying the triangle inequality in both directions gives a true-step lower bound of max(0,dᵢ−rᵢ−rᵢ₊₁) and an upper bound of dᵢ+rᵢ+rᵢ₊₁.
For a recorded five-metre segment with half-metre bounds at both ends, the true endpoint separation lies between four and six metres. The bounds describe positions at the fix times. They do not limit how far the animal may have travelled between those positions.
The zero in the lower bound matters. A negative distance is not an informative biological result; it means the endpoint uncertainty regions may overlap. The bounds can be attained for an isolated pair when the relevant points can move along their connecting line without other constraints. In a whole trajectory, the shared middle points couple neighbouring segments, which changes what can be attained simultaneously.
CHAPTER 8 OF 25 · Build and tighten geometric bounds
8. Sum segment bounds without calling them exact extrema
Adding valid lower bounds gives a valid lower bound on the true-fix polyline. Adding valid upper bounds gives a valid upper bound on that polyline. The result can be conservative because the same unknown fix participates in two neighbouring segments.
Return to the recorded triangle (0,0), (3,4), (6,0), now with half-metre error bounds at each point. Each recorded segment is five metres, so the pairwise bounds give 8 ≤ S_true-fixes ≤ 12 metres.
Every compatible set of true fix positions respects that interval. We have not proved that both endpoints can be achieved by one set of three positions. The lower bound for the first segment might demand a middle position different from the one needed to minimise the second.
Report the result as a conservative interval unless a joint optimisation or explicit endpoint construction establishes sharpness. A valid inequality is useful without being the narrowest possible inequality. Distinguishing valid from sharp prevents a learner from declaring an uncertainty calculation exact merely because each individual step used a correct formula.
CHAPTER 9 OF 25 · Build and tighten geometric bounds
9. Solve a coupled one-dimensional record exactly
Use recorded positions 0, 4 and 8 metres on a line. Each true position is guaranteed to lie within one metre of its matching record. Thus x₀ lies in [−1,1], x₁ in [3,5] and x₂ in [7,9].
Every admissible sequence is ordered x₀ < x₁ < x₂. Therefore its true-fix polyline is (x₁−x₀)+(x₂−x₁) = x₂−x₀. The middle coordinate cancels. The smallest possible value is 7−1 = 6 metres, and the largest is 9−(−1) = 10 metres.
Both endpoints are attainable. For six metres, choose positions 1,4,7. For ten, choose −1,4,9. Every chosen position respects its bound. This proves the exact interval 6 ≤ S_true-fixes ≤ 10.
| Quantity in the ordered three-fix example | Supported interval |
|---|---|
| Recorded polyline | 8 m exactly |
| Sum of separate true-step bounds | 4–12 m, conservative |
| Joint true-fix polyline | 6–10 m, sharp |
The separate segment bounds would have given [4,12], because each recorded four-metre step individually permits [2,6]. That interval is valid but wider. To minimise both steps separately, the middle position would need to be simultaneously near opposite ends of its allowed interval. One animal has one middle position; it cannot be assigned independently to each calculation.
CHAPTER 10 OF 25 · Build and tighten geometric bounds
10. Check what removing the middle fix changes
With only the first and last recorded positions in the ordered example, the true endpoint separation still lies in [6,10] metres. Adding the middle fix did not tighten those bounds on S_true-fixes because its entire allowed interval already lies between every admissible start and finish.
This is an informative special case, not a rule that intermediate fixes are useless. A middle observation away from the start-to-finish route can add a stronger geometric requirement. An observation can also matter to timing, habitat classification or quality review even when it does not change this particular distance interval.
Ask the learner to predict which observation could tighten the bound before calculating. If the new allowed region forces a detour away from all shortest start-to-finish paths, it can raise the minimum. If it sits along already feasible straight routes, it may add no distance information.
This is a practical way to connect data collection with the target question. More records are not automatically more informative about every quantity. Their value depends on what uncertainty they resolve, how accurately they locate the animal and whether their sampling arrangement can be compared with the rest of the record.
CHAPTER 11 OF 25 · Add time and error models
11. Keep true-fix length separate from continuous travel
Even the exact interval [6,10] in the ordered example concerns the polyline through true fix positions. It is not an upper bound of ten metres on the actual path. The animal could move away and return between observations while passing through the same positions at the fix times.
Without a supplied speed or path constraint, finitely many point observations do not create a finite mathematical upper bound on continuous path length. One can insert progressively longer excursions between the same endpoint times if arbitrarily high speeds remain allowed in the model.
That is a statement about missing constraints, not a claim that a real animal has unlimited speed. Real movement is limited, but the analysis must identify appropriate evidence before imposing a numerical limit. A guessed plausible speed is an added assumption, not a value measured by the sparse record.
A careful report can therefore combine a lower bound on L_path with a bounded interval for S_true-fixes while leaving the path upper bound unresolved. Different levels of certainty for different quantities are an accurate result. Forcing all of them into one neat distance estimate would conceal the distinction the data actually supports.
CHAPTER 12 OF 25 · Add time and error models
12. Introduce a speed ceiling explicitly
Suppose a new classroom model supplies a hard speed ceiling u throughout the observation window. If the window lasts T seconds, L_path ≤ uT. This follows because total distance is the time integral of speed, and speed never exceeds u under the declared assumption.
For fix times 0, 3 and 6 seconds with u = 2 metres per second, the whole-window upper bound is twelve metres. Combined with the ordered position constraints, the earlier minimum of six metres remains possible: the positions 1,4,7 can be joined at one metre per second.
The speed ceiling must also be checked interval by interval. A long enough total window cannot rescue a pair of successive true positions whose separation requires more travel than the matching interval permits. Feasibility is a joint position-and-time question.
Do not read the fictional ceiling as a species fact or an instruction for filtering real tracking points. Here it is a supplied mathematical condition used to demonstrate how additional information changes an inference. A biological model would need a justified ceiling, the relevant behaviour and terrain, and an account of uncertainty in both locations and timestamps.
CHAPTER 13 OF 25 · Add time and error models
13. Use incompatibility to question the model, not accuse the animal
Imagine two recorded locations twenty metres apart, with one-metre hard error bounds at each end, one second between fixes and a stipulated speed ceiling of two metres per second. Their true separation is at least eighteen metres, but the model allows at most two metres of travel.
No trajectory satisfies all those conditions. The contradiction identifies an incompatible set of inputs. It does not establish which input is wrong. A timestamp, coordinate, unit, error bound or speed assumption may need review.
A useful evidence note preserves the original record and states the inequality: minimum required displacement exceeds maximum permitted travel. It then lists the conditions used in the calculation. This is more informative than silently deleting the point and presenting a smooth replacement path as observed movement.
The same principle applies to a fix flagged by a software quality screen. A flag is a prompt for an appropriately defined review, not independent proof of a biological impossibility. Distinguish a diagnostic condition from a correction rule and from the decision about whether a record belongs in a particular analysis.
CHAPTER 14 OF 25 · Add time and error models
14. Compare under-sampling with noisy over-counting
Exact sparse fixes can miss turns and make a connecting polyline shorter than the actual path. Location error can add apparent turns and make a recorded polyline longer than the true-fix polyline. These mechanisms can act at the same time.
Consequently, there is no universal sign for the error in one observed total relative to the animal’s actual travel. A record might miss large detours while adding small noisy zigzags. The two effects could partly cancel numerically without either becoming harmless.
A convenient agreement with another distance is therefore weak evidence unless the measurement and sampling mechanisms are understood. Two wrong contributions can produce a plausible total. Compare changes under a declared sampling interval and an appropriate error model rather than choosing the version whose number looks most reasonable.
The 2015 research paper linked at the end separates measurement error from interpolation error in GPS movement data. Our original examples make the distinction inspectable without importing the paper’s datasets or assuming that every animal record shares one error distribution. Keep each claim attached to the conditions that make it true.
CHAPTER 15 OF 25 · Add time and error models
15. Understand what zero-mean error does and does not imply
Suppose an error model has zero mean, so the average recorded position equals the true position in repeated comparable trials. That does not imply that the average calculated distance equals the true distance. Taking a vector’s length is a nonlinear operation.
For a stationary point with two-dimensional errors pointing in different directions, positive apparent distances can accumulate even though the position errors balance around zero. Averaging signed coordinate errors and averaging nonnegative lengths answer different questions.
But avoid the stronger claim that every unbiased error distribution must strictly inflate every expected step length. In a one-dimensional example with true displacement four metres and an error difference of either −2 or +2 with equal probability, the measured step is two or six metres. Its mean remains four.
The example shows why strictness and dependence require care. A statistical expectation is also not a guarantee about one realised record. Use the probability model to state an average behaviour under its assumptions, and use deterministic geometric bounds only when hard bounds have actually been supplied. Neither language should be substituted for the other simply because both mention uncertainty.
CHAPTER 16 OF 25 · Preserve statistical and sampling scope
16. Check a squared-distance identity independently
Let d be the true displacement vector and δ the difference between successive location errors. Then ||d+δ||² = ||d||² + 2d·δ + ||δ||². If the model gives E[δ] = 0 and finite second moments, averaging removes the middle term.
Thus E[recorded squared step] = ||d||² + E[||δ||²]. The additional term is nonnegative and vanishes when the error difference is zero with certainty. This explains why shared offsets behave differently from changing errors.
Do not take a square root of both sides and call it the expected recorded distance. In general E[√Z] is not √E[Z]. In the one-dimensional two-or-six example, the mean squared step is (4+36)/2 = 20 square metres, while the mean step is four metres. The square root of twenty is neither of those quantities.
This is a useful transfer check because the arithmetic is small and the conceptual mistake is common. State whether an analysis reports distance, squared distance, mean distance or a root-mean-square quantity. Their units and averaging order reveal that they answer different questions.
CHAPTER 17 OF 25 · Preserve statistical and sampling scope
17. Do not convert a reported accuracy statistic into a hard boundary
A receiver’s stated error statistic may be a percentile, a standard deviation, a quality category or a result under particular test conditions. Those descriptions are not interchangeable with a guarantee that every true position lies inside a supplied circle.
Our geometric intervals depend on hard bounds. If a circle contains a true location with a stated probability instead, a deterministic whole-track conclusion requires different reasoning. Many fixes also create a joint-coverage question; a per-fix probability is not automatically a guarantee for every point simultaneously.
Keep the actual definition in the record. If only an indicative accuracy measure is available, report it as such and avoid treating all outside positions as impossible. A precise-looking error circle on a map may be a visual summary rather than the full statistical model.
The best learning task is to give students two differently defined accuracy cards and ask which calculations each supports. One card may permit an exact geometric bound; another may support only a model-based probability statement. Choosing the correct interpretation before calculating is more valuable than adding extra decimal places to an unsupported interval.
CHAPTER 18 OF 25 · Preserve statistical and sampling scope
18. Keep screening and missingness visible
Removing locations can change more than a distance total. It changes which places, behaviours or conditions remain represented. A quality rule may reject more fixes under dense vegetation or difficult reception, making the retained record a different sample.
USGS research on GPS error screening examined this tradeoff under its tested conditions. Its findings are a reason to evaluate the consequences of a screen, not a universal instruction to retain or delete every particular class of fix.
For a fictional comparison, preserve the original timestamps and mark excluded observations rather than closing the gaps silently. A ten-minute gap created by screening is not equivalent to a ten-minute continuous observation of no movement. If a new line is drawn across that gap, label it as an interpolation.
A reproducible distance account therefore names the inclusion rule, the remaining time window and any changes to observation spacing. Otherwise two analysts can produce different totals from the same file while appearing to apply the same distance formula. The disagreement may lie in the record selection rather than in the geometry.
CHAPTER 19 OF 25 · Preserve statistical and sampling scope
19. Reconstruct a changed example before reading its answer
A new one-dimensional record gives positions 2, 7 and 12 metres. The guaranteed error radius is half a metre at every fix. Times are 0, 4 and 8 seconds. First find the recorded polyline, then a conservative segment interval, then the exact true-fix interval using the ordered positions.
The recorded steps are five metres each, so S_recorded = ten metres. Each true step is individually between four and six metres, giving the conservative interval [8,12]. However, all admissible positions remain ordered: the first lies in [1.5,2.5], the middle in [6.5,7.5] and the last in [11.5,12.5].
Their polyline is the last position minus the first. Its exact interval is therefore [9,11] metres. Choose positions 2.5,7,11.5 for the lower endpoint and 1.5,7,12.5 for the upper endpoint to verify attainability.
If the problem additionally supplies a speed ceiling of two metres per second throughout, the continuous-path upper bound is sixteen metres over eight seconds. The true-fix upper bound of eleven does not replace it. Explain the distinction before comparing any of these values with an animal’s actual travel.
CHAPTER 20 OF 25 · Preserve statistical and sampling scope
20. Keep an average-speed claim attached to its distance
Dividing a distance by a duration does not repair an uncertain distance. It carries that uncertainty into a rate. State whether the numerator is a recorded polyline, a true-fix lower bound or a modelled continuous path before reporting an average speed.
In the changed eight-second example, the recorded polyline gives 10/8 = 1.25 metres per second as a map-derived rate. The true-fix interval [9,11] gives a corresponding interval [1.125,1.375] for that sampled-position quantity divided by eight. It is not an interval containing every possible actual path-average speed.
With the separately stipulated two-metres-per-second ceiling, actual path length lies above its supported lower bound and at most sixteen metres. Its whole-window average speed is consequently at least 9/8 and at most two under those combined assumptions. No inference about instantaneous speed follows from the map-derived 1.25 alone.
If observation times are uncertain, the denominator also needs an account. Do not mix a precisely logged file timestamp with an unsupported claim about the physical capture instant. A reproducible rate sentence names the interval and the quantity divided by it. The extra wording prevents a correct division from silently changing the target of the analysis.
CHAPTER 21 OF 25 · Practise and continue
21. Compare nested samples without declaring an optimal interval
Take a stationary fictional animal and an error sequence that alternates between −0.5 and +0.5 metres. Retaining every location creates a one-metre segment at each step. Retaining every second location can leave repeated identical coordinates and a recorded distance of zero.
That numerical change does not prove that the coarser interval is generally more accurate. In this specially constructed record, it happens to cancel the alternating pattern. For an exact moving path with turns between those retained times, the same thinning operation could hide genuine travel.
The example makes an evaluation principle visible: compare sampling choices against an appropriate reference and a stated measurement model, rather than selecting whichever total is smaller or smoother. Different animals, behaviours and error dependence can change the result.
A classroom project can construct both records on the same timeline: one stationary with alternating errors, another genuinely moving with exact locations. Apply the identical thinning rule and explain why its effect has different meanings. Preserve the original observations and label the reduced record as a deliberate selection. The exercise supports critical method comparison without prescribing a field protocol or suggesting that a preferred-looking map proves the underlying movement history.
Start with the biological question and the observation window. State the coordinate reference, units, ordered fixes, missing-data rules and whether the distance is horizontal or three-dimensional. Identify which quantities are recorded, assumed and calculated.
Next name the target: recorded polyline, true-fix polyline or actual continuous travel. If location uncertainty is represented, state whether it is a hard bound or a probabilistic description and whether successive errors are linked. Preserve the source and scope of any speed or path constraint.
Show enough intermediate values to rebuild the result. A segment table can list recorded separation, endpoint bounds and the resulting conservative interval. Add any stronger joint argument, such as ordered positions, rather than presenting it as if it came from independent segment minima.
Finish with a proportionate conclusion and one unresolved question. For example: the supplied ordered-coordinate model bounds the true-fix polyline between six and ten metres, while the actual path remains longer than or equal to six and has no supplied finite upper bound. That sentence is less dramatic than a single travel total, but it tells another analyst exactly what has been established.
CHAPTER 23 OF 25 · Practise and continue
23. Choose repair, stabilisation or extension
For repair, use exact positions and distinguish net displacement from route length. Draw the three-four-five triangle example and ask why the route is ten metres while its start-to-finish separation is six. Then introduce the stationary noisy example to show which earlier assumption changed.
For stabilisation, derive the one-segment bound with hard endpoint radii. Work the ordered one-dimensional record and construct the endpoint histories. Require the learner to explain why independently attainable segment minima need not be jointly attainable in one trajectory.
For extension, add a declared speed ceiling, test interval feasibility and compare a common offset with alternating errors. Keep deterministic bounds, probability models and biological interpretation in separate sentences. A more advanced method is useful only when the learner can still say what quantity it estimates.
A completed project can use fictional data on paper or in a spreadsheet. No wildlife disturbance is needed. The strongest evidence of learning is a changed case explained independently: what became more certain, which assumption did the work and what remains outside the result. That is a practical scientific habit, not merely a polished map.
CHAPTER 24 OF 25 · Practise and continue
24. Questions about dots, errors and animal travel
Is the recorded track length always a minimum travel distance? It is a lower bound under exact fixes and appropriate geometry. Location error can invalidate that simple interpretation.
Does a longer noisy track prove more movement? No. Changing location errors can add apparent steps, including for a stationary animal.
Can I subtract the error radius from every segment? An error description is not a known correction vector. Pairwise bounds may also be conservative because neighbouring segments share a fix.
Do more fixes always improve the distance estimate? Exact nested fixes tighten a geometric lower bound. Noisy fixes introduce an additional measurement mechanism that needs evaluation.
Does a bounded true-fix polyline bound the full path above? No. Unobserved detours require a supplied speed or path constraint before a finite upper bound follows.
Can this guide establish migration success or habitat preference? No. Those are separate ecological questions with different evidence needs. A distance calculation should not quietly become a conservation outcome claim.
Check measurement and sampling mechanisms
Ranacher and colleagues: Why GPS Makes Distances Bigger Than They Are examines measurement and interpolation error in movement records. USGS: Is GPS Telemetry Location Error Screening Beneficial? reports a particular study of screening and retained information. These primary research routes were checked on 23 October 2026. Our coordinates, hard bounds and constructed histories are original teaching examples, not a reanalysis of their data.
Keep the ecological question intact
Wildlife Tracking, Telemetry and Migration Evidence provides the wider measurement route. Bounding Habitat Residence Time Between Tracking Fixes concerns time and crossing assumptions. Matching Habitat Use with Each Animal's Available Environment develops a different comparison target.
Continue responsibly
Return to Wildlife and Wildlife Conservation for population and intervention context. Continue through How X Works and Learning That Transfers into Life. A useful final check is to explain why the exact interval for sampled positions and the upper bound for continuous travel can have different endpoints without contradicting each other.