Landscape genetics investigates how the arrangement and history of environments relate to genetic variation and reproductive exchange among organisms. Its gold standard is a testable explanation connecting a landscape feature with a biological process, supported by suitable sampling and independent checks. A road on a map, a genetic difference between sites and a predicted corridor are three different observations; a dependable study explains how they are connected and which alternative explanations remain.
This guide connects Landscape Ecology, Population Genetics and Conservation Genetics. It develops original worked examples of routes, resistance, sampling and monitoring. Every numerical landscape below is fictional. Its values demonstrate a method, not a movement threshold or intervention prescription for a real species.
THE GOLD STANDARD · LANDSCAPE GENETICS
Test how landscapes shape reproductive connections
Choose a route through spatial models, genetic evidence, sampling and tests of connectivity.
ROUTE 1 · CHAPTERS 1–3
Frame the mechanism
Distinguish distance, movement and reproductive exchange.
ROUTE 2 · CHAPTERS 4–6
Represent the landscape
Choose scale, resistance and least-cost routes.
ROUTE 3 · CHAPTERS 7–10
Connect models and samples
Compare multiple paths and independent evidence.
ROUTE 4 · CHAPTERS 11–14
Test the explanation
Account for history, validation and fragile assumptions.
ROUTE 5 · CHAPTERS 15–17
Transfer and continue
Solve the route problem and build a learning plan.
Full chapter index · Try the transfer task · Gold Standard subject hub
Chapter index
Frame the mechanism · Chapters 1–3
Represent the landscape · Chapters 4–6
Connect models and samples · Chapters 7–10
Test the explanation · Chapters 11–14
Transfer and continue · Chapters 15–17
CHAPTER 1 OF 17 · Frame the mechanism
1. Ask which landscape process the genes could reveal
Imagine two woodland populations separated by a road. Their genetic samples differ. It is tempting to say the road caused the difference, especially when the road is visually prominent. But the populations might have been separated before the road existed. They might occupy different climates, have different founding histories or have been sampled unevenly. The road is a hypothesis, not an explanation completed by a map.
A useful question specifies the organism, landscape feature, genetic response and timescale. For example: does this road reduce successful reproductive exchange among these woodland populations over recent generations, beyond the effect of geographic distance? That wording identifies a comparison. It also prevents an analysis of ancient population separation from being presented as direct evidence about a recently built road.
The field’s foundational 2003 paper by Manel and colleagues connects spatial genetic patterns with environmental and geographic features. The central work is combining biological and spatial evidence, rather than simply adding coordinates to a genetic dataset.
Write down the predicted observations before examining the final results. If a road restricts exchange, pairs across it might differ more than comparable pairs on the same side, subject to the population’s history and the chosen genetic measure. If distance alone explains the pattern, the road may add little predictive value.
Also state what the study cannot answer. Genetic samples may help examine reproductive connectivity while saying little about collision mortality during one week. A camera survey may reveal crossings without establishing reproduction. Matching the evidence to the question is the first practical skill in landscape genetics.
CHAPTER 2 OF 17 · Frame the mechanism
2. Follow the sequence from dispersal to inheritance
An organism can leave its birthplace, cross a landscape and survive at a new site without contributing genes to the next generation. Gene flow requires reproduction to carry genetic material into the recipient population. The distinction is especially important when monitoring a newly created wildlife passage.
Consider a fictional sequence: 100 animals approach a crossing, 40 pass through, 20 establish themselves beyond it, and five eventually reproduce. These numbers describe successive outcomes. A count of 40 crossings does not establish 40 breeding migrants. If the same individual crosses repeatedly, even the number of unique movers is smaller than the number of recorded events.
Different evidence addresses each step. Tracking can describe paths and survival during movement. Repeated observations can help identify residents. Parentage or ancestry analyses can support reproductive exchange. Descendant survival and reproduction address whether that exchange contributes to population persistence. No single method automatically supplies every link.
For plants, pollen and seeds can follow different routes. Pollen may move between established plants without relocating those plants, while seed dispersal carries a developing offspring. Nuclear and organelle markers can therefore reflect different aspects of history, depending on the species’ inheritance system. The biological interpretation should precede the choice of marker.
This sequence also clarifies intervention claims. The existing specialist guide to Genetic Rescue, Outbreeding Risk and Assisted Gene Flow examines a further question: whether gene flow improves fitness. Landscape genetics can help locate restrictions or connections, but a predicted connection alone cannot establish a rescue benefit.
CHAPTER 3 OF 17 · Frame the mechanism
3. Separate distance, resistance and environmental differences
Isolation by distance describes a pattern in which genetic similarity tends to decline with geographic separation under relevant dispersal and demographic conditions. Isolation by resistance asks whether the difficulty of crossing the intervening landscape helps explain genetic relationships. Environmental differences can also be associated with genetic patterns through processes such as selection or habitat-associated reproduction.
These possibilities are not mutually exclusive. Nearby sites may be separated by a strong barrier. Distant sites may be connected by a continuous river corridor. Two habitats may differ environmentally while a historical population boundary runs between them. An analysis needs comparisons capable of separating these explanations.
Suppose every sampled pair across a motorway is also more than ten kilometres apart, while every same-side pair is less than two kilometres apart. If across-road pairs differ genetically, distance and road exposure are confounded. There is no clean comparison showing which accounts for the difference. Collecting more samples under exactly the same arrangement may sharpen the pattern without resolving its cause.
A better design could include similar-distance pairs with and without the road, where the landscape allows them, and sample across other relevant gradients. This does not make the field experiment perfectly controlled. It makes the alternatives more distinguishable.
Avoid treating the best-fitting model among a narrow set as the true mechanism. If all candidate models omit historical separation, the winning road model may absorb that missing explanation. A useful model comparison includes a defensible baseline, biologically motivated alternatives and an explicit account of what remains outside the analysis. The question is whether the landscape model adds supported understanding, not whether software can rank several formulas.
CHAPTER 4 OF 17 · Represent the landscape
4. Choose the spatial extent and resolution
A landscape is represented at a chosen extent, the overall area included, and a chosen grain, the size of its spatial units. These are analytical decisions. They can alter the routes and barriers visible to the model.
Imagine a narrow strip of suitable vegetation running through a developed area. A fine map may represent it as a continuous connection. A coarse cell containing vegetation and buildings may classify the entire mixture as developed land. The resulting break is partly a representation effect. Conversely, averaging a narrow road into a large green cell may erase a real obstacle.
Neither the finest available map nor the largest available study region is automatically best. Resolution should match positional accuracy, relevant movement processes and reliable environmental data. A one-metre grid does not create one-metre knowledge when sample coordinates are uncertain by hundreds of metres. Enlarging the extent can capture alternative routes, but it also increases demands on data and interpretation.
For an original classroom check, place a sampling point 15 metres from a mapped habitat boundary and assume its location uncertainty is a radius of 30 metres. The point could plausibly belong on either side. A model that assigns it confidently to one habitat class conceals an uncertainty larger than the distance to the boundary.
Repair the analysis by examining plausible alternative positions or aggregating to a scale supported by the data. Record the coordinate reference system and distance units as well. Geographic coordinates expressed in degrees should not be silently treated as distances in metres. Before debating biology, confirm that the locations and landscape layers actually align.
CHAPTER 5 OF 17 · Represent the landscape
5. Build a resistance surface as an explicit hypothesis
A resistance surface assigns values representing a hypothesised difficulty of movement or gene flow through places. Those values may be informed by observations, expert knowledge or statistical fitting. They remain model inputs with assumptions attached. A land-cover label is not itself a measured resistance value.
Suppose a fictional map assigns woodland resistance 1, grassland 3 and developed land 9. This says how the model compares those classes. It does not mean an animal is nine times more likely to die in developed land, nor does it imply a universal biological scale. Multiplying every value by ten may preserve some route rankings while changing numerical resistance distances.
The values must match the process. Habitat preferred for feeding is not necessarily the habitat used during dispersal. An animal may move quickly through an area it rarely occupies, or settle in a preferred patch without reproducing successfully. Roffler and colleagues’ research explicitly examines the relationship between habitat-selection models and landscape features influencing gene flow.
Start with a short set of biologically distinct candidate surfaces. State why each variable belongs and how its transformation changes the hypothesis. For example, a gradual increase with road density represents a different claim from treating any road as an absolute barrier.
Missing data need their own treatment. An unknown cell is not automatically a barrier, and a barrier is not merely an expensive cell. A model that excludes cells entirely can disconnect a network. A model with high but finite resistance retains possible passage. That distinction should be deliberate and documented before interpreting the output.
A least-cost path finds the route with the smallest accumulated cost under the stated network or raster rules. It is a modelled optimum. It does not prove that an organism knows or follows that route.
Use a simple network with two routes between sites A and B. Route One contains three links with costs 2, 3 and 4. Its total cost is 9. Route Two contains two links with costs 5 and 5, giving total cost 10. The least-cost route is Route One, even though it contains more links. Link count and total cost are different quantities.
Now imagine restoration reduces one of Route Two’s costs from 5 to 2. Its total becomes 7, and the least-cost route changes. The model has identified a potential improvement under its assumptions. It has not measured a corresponding increase in migrants, offspring or survival.
In a spatial raster, movement length and transitions between cells need a defined rule. A diagonal step covers a different physical distance from an orthogonal step on a square grid. Whether diagonals are permitted can change which cells connect. These settings can matter near narrow barriers or gaps, so they belong in the methods rather than being left as unexplained defaults.
The original network is intentionally small enough to check by hand. Before trusting a large computation, reproduce such a toy case. Confirm that the algorithm returns 9 before restoration and 7 afterwards. Then deliberately block a link and check whether the resulting route matches the intended topology. These checks diagnose implementation mistakes separately from uncertainties in the biological resistance values.
Circuit-based models represent connectivity through multiple available paths. In a simple resistor network, resistances along one series path add, while alternative paths in parallel can reduce effective resistance. This mathematical structure differs from selecting only the cheapest route.
Take two independent parallel routes from A to B with total resistances 6 and 12 model units. Their effective resistance is 1 divided by (1/6 + 1/12), which equals 4. If the second route is removed, the resistance becomes 6. The best single route still costs 6 in both situations, but the parallel network has lost an alternative.
Add a third independent route with resistance 12 to the original network. Effective resistance becomes 1 divided by (1/6 + 1/12 + 1/12), or 3. This toy calculation shows why redundancy can affect a circuit metric without changing the best path. Real spatial networks often share edges, so complete routes cannot always be treated as independent parallel resistors.
McRae’s 2006 isolation-by-resistance paper develops the connection between circuit theory and genetic patterns. The Circuitscape documentation explains the software’s resistance inputs and connectivity outputs.
The electrical analogy is a model, not a claim that organisms behave like electrons. Effective resistance and current maps require interpretation in relation to the biological process and source configuration. A bright current concentration does not directly count animals. It can identify a modelled bottleneck or a place worth investigating, while field and genetic evidence test whether that interpretation applies.
CHAPTER 8 OF 17 · Connect models and samples
8. Read genetic responses at the right level
Landscape genetic studies may compare individuals, populations or generations. An individual relatedness measure, a population differentiation statistic and a parentage assignment do not measure the same thing. Choosing the response should follow the biological question.
If the question concerns recent successful dispersal, parentage or assignment methods may provide relevant evidence, subject to sampling and model limitations. If the question concerns accumulated population structure, differentiation measures may be useful. Historical demographic analyses address longer processes. Their conclusions should not be exchanged merely because all use genetic data.
Consider a fictional pair of sites with similar allele frequencies today. Similarity is compatible with ongoing exchange, but also with recent separation from a common population. If the barrier arose only last year in a long-lived species, much of the sampled generation may predate it. A small genetic difference cannot automatically certify present-day connectivity.
Conversely, a strong difference may persist after a barrier is removed. Reproductive exchange and demographic turnover take time to alter inherited patterns. A restoration project should not be judged solely by whether old structure disappears immediately.
Use several observations when the decision requires them, while keeping their meanings distinct. A camera record, a parentage result and a differentiation estimate can complement one another. Their disagreement may reveal timing, sex-specific movement or sampling issues. It should prompt a revised explanation rather than selective reporting of the result that best supports the preferred corridor proposal.
CHAPTER 9 OF 17 · Connect models and samples
9. Sample across the contrasts you need
Sampling determines which landscape explanations can be tested. A large collection of samples concentrated beside one accessible track may represent that track well while leaving the wider landscape largely unknown. The number of genotypes does not repair a missing spatial comparison.
Map the proposed samples before collecting them. Check which combinations of distance, habitat and barrier exposure are represented. If all high-elevation samples come from one region, elevation and region may be difficult to distinguish. If one site contains many close relatives, family structure may dominate a comparison intended to describe landscape effects.
In their 2012 simulation study, Oyler-McCance and colleagues compared sampling designs under specified landscape and genetic conditions. Designs differed in their ability to recover the generating process. The lesson is to test sampling adequacy for the problem; the study’s particular sample sizes are not universal requirements.
An original planning exercise makes this concrete. A budget allows 60 samples. Design A collects all 60 at one convenient pond. Design B spreads them over several ponds chosen to separate distance from barrier exposure. Design B is better aligned with a between-pond landscape question, but may leave too few individuals per pond for some population-level estimates. The choice depends on the response, not a blanket preference for spreading samples as widely as possible.
Simulate plausible data under competing explanations and ask whether the proposed design can distinguish them. Also plan for failed samples, uncertain locations and missing seasons. A design that works only if every specimen yields perfect data is fragile. Record deviations during fieldwork so the final analysis describes the actual sample rather than the intended one.
CHAPTER 10 OF 17 · Connect models and samples
10. Do not count every pair as an independent replicate
Pairwise datasets grow quickly. With n sampled sites, there are n(n − 1)/2 unordered site pairs. Ten sites produce 45 pairs; 20 sites produce 190. This arithmetic can make a modest field survey look like a much larger independent experiment than it is.
Pairs sharing a site also share information. The comparisons A–B and A–C both involve A. If A has an unusual history, poor-quality samples or a distinct environment, it can affect both comparisons. Treating all pairwise observations as independent can misrepresent uncertainty.
A simple diagnostic is to identify how many conclusions depend on one location. Imagine nine pairs involving a single northern site account for most of the apparent road effect. Removing that site changes the result dramatically. That does not prove the effect is false, but it shows that replication across independent landscape contrasts is weaker than the raw pair count suggests.
Statistical methods should account for the dependence structure and the form of the response. There is no universal instruction to apply an ordinary correlation test to two distance matrices and accept its significance as proof. USGS research on spatially structured network models develops approaches that explicitly represent landscape-genetic relationships as networks.
For readers evaluating a study, ask where the independent information comes from, how shared locations are handled, and whether conclusions survive reasonable changes in the sample. For learners, draw ten sites and label several pairs containing the same site. The shared endpoint makes the dependence visible before any statistical terminology is introduced.
Genetic patterns integrate reproduction over time. A current land-cover map is a snapshot. Matching the two without considering history can attribute an old genetic pattern to a recent environmental feature.
Imagine a woodland species with a generation interval of about ten years. A road opened two years ago. Sampling mature adults now may provide little direct evidence of the road’s effect on offspring produced after construction. A study can still establish a baseline and investigate movement, but its claim about accumulated genetic separation needs a realistic timescale.
Landguth and colleagues’ 2010 simulation research on barrier-detection lags showed that response times depend on the genetic statistic and dispersal conditions. Historical structure can persist, and different measures can respond differently. The reported simulation timings should not be transferred as a fixed waiting period for every restoration project.
Construct a timeline containing landscape changes, specimen dates, life stages and plausible generations. Historical maps or archived samples may help, provided their spatial accuracy and laboratory quality are understood. If the maps have incompatible classifications, apparent change may partly reflect different mapping systems.
The practical consequence is to separate early and later indicators. Early monitoring might establish use of a passage and successful breeding by migrants. Later monitoring may examine whether population structure or diversity changes. Neither stage replaces survival and reproduction records. A project can be progressing biologically before a slow statistic responds, while a rapid statistic can change without demonstrating durable population recovery.
CHAPTER 12 OF 17 · Test the explanation
12. Validate predictions beyond the data used to fit them
A flexible resistance model can fit the observations used to choose its parameters while predicting poorly elsewhere. This is especially likely when many transformations, variables and combinations are tried and only the strongest result is reported. Fitting and testing need distinct roles.
Begin with a simple baseline, such as a model using geographic distance under clearly stated assumptions. Ask whether the proposed landscape variables improve prediction on appropriately held-out information. The relevant validation design should match the intended use. Predicting a new region is a more demanding task than predicting another nearby pair among familiar sites.
Randomly splitting pairwise rows can allow the same individuals or sites to appear in both training and test sets. The test then contains familiar endpoints and may overstate performance for genuinely new locations. Holding out sites or spatial blocks can provide a more relevant challenge, while still requiring attention to sample size and the remaining dependence.
For a fictional error comparison, suppose a distance-only model makes absolute prediction errors of 0.03, 0.05, 0.04 and 0.08 on four withheld cases. Its mean absolute error is 0.05. A landscape model gives 0.02, 0.03, 0.05 and 0.06, averaging 0.04. The improvement is 0.01 in the response’s units. Four cases provide limited evidence; the difference is not automatically practically important or statistically secure.
Inspect where the model fails. If its largest errors occur across rivers, an omitted process may matter. If errors concentrate in low-quality samples, measurement needs review. A failure map can guide the next observation more effectively than celebrating a single summary of overall fit.
CHAPTER 13 OF 17 · Test the explanation
13. Test sensitivity and repair fragile explanations
Sensitivity analysis asks whether a conclusion changes under reasonable alternative assumptions. In landscape genetics, useful changes include resistance values, spatial grain, study boundaries, treatment of missing cells, genetic filters and the set of sampled locations.
Suppose a proposed corridor ranks first when developed land has resistance 10 but falls to fourth when the value is 20. If both values are biologically plausible, the ranking is uncertain. Reporting only the preferred value hides a decision-relevant dependence. Investigate which data could narrow that uncertainty before presenting the corridor as uniquely best.
| Warning sign | Possible explanation | Useful repair |
|---|---|---|
| A corridor vanishes after small grid changes | Resolution or alignment dominates | Compare justified grains and check narrow features |
| One site controls the relationship | Unusual history, family structure or error | Inspect that site and seek replicated contrasts |
| A habitat model predicts gene flow poorly | Residency and dispersal differ | Revisit the biological process and response |
| Current barriers do not match genetic structure | Historical signal or delayed response | Add a timeline and suitable temporal evidence |
| Excellent fitted performance collapses elsewhere | Overfitting or spatial transfer failure | Use independent spatial validation and simpler alternatives |
Repair should follow the identified problem. More markers will not necessarily solve a spatial design that confounds road exposure with distance. A finer raster will not solve uncertain specimen identity. More elaborate modelling will not turn unknown cells into observed habitat.
Keep rejected explanations and sensitivity results available. A reader should be able to see which choices matter and why the final interpretation was retained. This record makes later updating easier when new samples or landscape changes challenge the original conclusion.
CHAPTER 14 OF 17 · Test the explanation
14. Turn connectivity evidence into a bounded decision
A model can identify candidate bottlenecks, compare scenarios or guide monitoring. Choosing an intervention also requires ecological feasibility, the source and recipient populations, competing threats and the effects of the action itself. Connectivity is one part of persistence.
Consider two fictional proposals. Proposal A lowers modelled resistance between patches that both support successful reproduction. Proposal B produces a larger numerical reduction but connects to a patch where juveniles consistently fail to survive. The larger connectivity improvement does not automatically make B the better conservation action. Destination quality and the intended outcome belong in the comparison.
Similarly, a connection may facilitate unwanted processes, such as movement of an invasive organism or a pathogen. Whether that matters depends on the species and setting. The appropriate response is an explicit ecological assessment, not an assumption that every connection is either beneficial or dangerous.
Before intervention, define separate observations for construction, use, reproductive exchange and population outcomes. Completing a passage is a delivery milestone. Detecting individuals using it is a biological observation. Demonstrating descendants across the former barrier is stronger evidence for gene flow. Population recovery requires further support.
The Conservation Genetics root provides the wider framework for genetic evidence and persistence. Use the Landscape Ecology guide when the question concerns habitat arrangement, edges or the surrounding environment. Linking these perspectives prevents the genetic analysis from becoming detached from the habitat that must support the resulting population.
CHAPTER 15 OF 17 · Transfer and continue
15. Transfer task: the cheapest path stays the same
Two habitat patches are connected in a fictional network by independent routes with resistances 8 and 24. There are no shared edges between routes. A team proposes closing the second route because the least-cost route already uses the first. They claim that connectivity will therefore be unchanged.
Calculate the least-cost value and effective resistance before and after closure. Then consider a different proposal: retain both routes but reduce the second route’s resistance from 24 to 8. What happens to the two metrics?
Finally change the biological evidence. Tracking confirms occasional movement along both routes, but no parentage data are available and the genetic samples were collected before either proposal. State what the mathematical comparison can support and what would be required to claim improved reproductive gene flow.
Answer and reasoning
Initially the least-cost value is 8. Effective resistance is 1/(1/8 + 1/24) = 6. Closing the second route leaves a single route with resistance 8. The least-cost value is unchanged, but effective resistance increases from 6 to 8, a one-third increase relative to 6. The claim that connectivity is unchanged is therefore metric-dependent and incomplete.
If both routes instead have resistance 8, the least-cost value remains 8 and effective resistance becomes 4. Relative to the original 6, that is a one-third reduction. These calculations show how parallel alternatives affect the circuit model. They do not establish matching percentage changes in animal movements or reproductive success.
The tracking observations support use of both routes by the monitored animals. They do not prove successful reproduction after movement. The old genetic samples can provide baseline information but cannot measure the consequences of proposals that have not yet occurred. A relevant follow-up design would connect individual movement with reproduction and later population observations, allowing for the species’ generation time.
Now remove the assumption of independent routes. If both pass through the same narrow bridge, that shared segment can dominate connectivity. You must represent the shared edge explicitly rather than apply the parallel formula to complete route totals. This changed topology is a different problem even if the labels on the map look similar.
Is landscape genetics the same as landscape genomics?
The fields overlap. Landscape genetics often focuses on spatial genetic patterns and processes such as gene flow. Landscape genomics commonly uses genome-scale information and may place greater emphasis on environmental associations and adaptation. Terminology varies, so read the actual question and methods rather than infer the entire study from its label.
Does a genetic boundary prove that a road is a barrier?
No. The boundary may reflect history, geographic distance, environmental differences or sampling. A convincing road explanation needs relevant contrasts and consideration of those alternatives. The date of construction also matters in relation to the generations represented by the samples.
Can a resistance map show exactly where animals travel?
It predicts connectivity under assumptions. Tracking and direct observations address actual movement. Even a good match does not mean every individual follows the same route. A map’s purpose and resolution determine what kind of interpretation is reasonable.
Is a current-density hotspot a count of migrants?
No. It is a model output conditional on the resistance network and source configuration. It may help identify concentrated pathways, but converting it into an animal count requires a separately supported relationship. The map legend should not imply that such calibration exists when it does not.
Does more genomic data solve poor sampling?
It can improve some genetic estimates, but it does not create missing landscape contrasts or correct an unrepresentative spatial design. A larger marker set and a better field design solve different problems. Both should be justified by the question.
How can students learn this without collecting wildlife samples?
Use invented networks, simulated genotype tables and published datasets with appropriate permissions. Calculate route metrics, test alternative explanations and examine how conclusions change when locations or assumptions change. These activities develop the core reasoning without disturbing organisms or disclosing sensitive locations.
CHAPTER 17 OF 17 · Transfer and continue
17. Build a practical learning and research route
Begin by drawing a network with two patches and two routes. Calculate the least-cost value, then calculate effective resistance for genuinely independent parallel routes. Add a shared bottleneck and explain why the earlier shortcut no longer applies. This establishes the relationship between a map’s topology and its numerical outputs.
Next design a hypothetical survey. Choose a focal organism and state whether you want to understand movement, successful reproduction or accumulated genetic structure. Mark a set of sampling locations that distinguishes distance from one proposed barrier. Identify an important combination that is missing. Explain what additional location would improve the design and why.
For a research-reading exercise, compare the sampling-design and time-lag studies below. Write four sentences for each: what was simulated or observed, what assumptions defined the study, what the result supports, and what cannot be transferred automatically to a new species. This keeps a published numerical result attached to its conditions.
The practical skills span field ecology, genetics, mapping, statistics and reproducible analysis. A useful project record includes sample identity, coordinate accuracy, map dates, resistance transformations, model settings and validation results. Preserve uncertain information explicitly. If another person cannot reconstruct which landscape and which organisms were analysed, a polished map is difficult to evaluate.
Primary research and official technical references
- Manel and colleagues, 2003: Landscape genetics, for the connection between geographic features and spatial genetic patterns.
- McRae, 2006: Isolation by resistance, for the mathematical connection between circuit networks and genetic differentiation.
- Oyler-McCance and colleagues, 2012: Sample design effects, for a simulation-based comparison of sampling regimes.
- Landguth and colleagues, 2010: Barrier-detection lags, for the interaction of genetic measures, dispersal and time.
- Circuitscape user documentation, for the software’s modelling workflow and interpretation requirements.
Continue through The Gold Standard master or How X Works to connect these methods with other subjects. The Education parent guide explains the learning principle behind the transfer task: a method becomes useful when you can recognise which assumptions survive a changed situation, and which must be rebuilt.