Lesson 4 of 10 · 20 min
Sampling an ecosystem
Choose a sampling method (random, systematic or stratified) and technique (quadrat, line or belt transect, capture–recapture) to suit a site, and reduce bias in how you sample.
What you will be able to do
- Describe random, systematic and stratified sampling, and choose the one that suits a site.
- Describe quadrats, line transects, belt transects and capture–recapture, and choose a technique for an organism.
- Explain strategies that minimise bias, including the size and number of samples, random-number generators, counting criteria, and calibrating equipment and noting its precision.
- Explain how habitats, microhabitats and distribution patterns affect the validity and reliability of a sampling design.
Nobody counts every grass plant in a paddock or every crab on a mudflat. Ecologists sample: they count a small, carefully chosen part of the area and use it to estimate the whole. The estimate is only as good as the sample. If the sample does not represent the site, the calculations that follow are wrong however carefully they are done.
A sampling design answers three questions: where to sample (the method), how to sample (the technique), and how to keep the results free of bias.
Key idea
Sampling methods
- Random sampling: sample positions are chosen by chance, using coordinates from a random-number generator on a grid laid over the site. Every point has the same chance of being chosen. Best for a fairly uniform site.
- Systematic sampling: samples are taken at regular intervals, such as every 5 m along a line. Best for showing how organisms change across an environmental gradient (from a creek bank inland, or from a forest edge inwards).
- Stratified sampling: the site is divided into distinct zones or layers (strata), and each is sampled separately, usually in proportion to its area. Best for a site with clearly different habitats, such as a dune system with foredune, swale and hind dune, or the layers of a forest.
A student wants to find out how the plant community changes from the edge of a salt pan to the mangroves fringing a creek, 60 m away. Which sampling method suits this best?
| Technique | How it works | Suits | Limitations |
|---|---|---|---|
| Quadrat | A square frame of known area (for example 1 m × 1 m for grasses, 10 m × 10 m for trees) is placed at each sample position; organisms inside are counted or their cover estimated. | Plants and slow or sessile animals (barnacles, snails, corals) | Misses mobile animals; time-consuming for large sites; needs counting criteria for edges. |
| Line transect | A tape is laid across the site and every organism touching it at set intervals (for example every 1 m) is recorded. | A quick picture of which species occur along a gradient | Records presence only, not abundance; easily misses species that don't touch the line. |
| Belt transect | Quadrats are placed along a line at set intervals, so a strip of known width is sampled. | How abundance (density, cover) changes along a gradient | Slower than a line transect; still one strip, so it may not represent the whole site unless repeated. |
| Capture–recapture | Animals are caught, marked and released; a second catch later shows what fraction are marked. | Mobile animals: fish, small mammals, insects, lizards | Relies on assumptions (no births, deaths or migration between samples; marks don't affect survival). |
Place quadrats at random, then systematically, then in strata. Which design gives an estimate closest to the true count, and which needs the fewest quadrats?
Minimising bias
Bias is a systematic error: something about the way you sample that pushes the result one way. A student who drops quadrats where the plants look interesting will overestimate them. The syllabus names five strategies to minimise it.
| Strategy | What to do | Why it helps |
|---|---|---|
| Size of samples | Choose a quadrat big enough to include most of the species present. A species–area curve (below) shows the size beyond which bigger quadrats add few new species. | Too small a quadrat misses species and gives a false picture of the community. |
| Number of samples | Take many samples, and keep adding them until a running mean of the counts stops changing much. | More samples reduce the effect of an unusual quadrat and make results more reliable. |
| Random-number generators | Use a calculator or app to generate the coordinates of each quadrat or the start of each transect. | Removes the observer's choice, so every position is equally likely. |
| Counting criteria | Decide before you start how to count organisms on the edge (for example: count a plant only if more than half of it is inside, or count edge plants on the top and left sides only), and what counts as one individual of a clumping plant. | Every observer counts the same way, so results can be compared and repeated. |
| Calibrating equipment and noting its precision | Check light meters, thermometers, pH and salinity probes against a standard before use, and record the precision (for example for a digital pH meter reading to 0.1). | Uncalibrated instruments give systematic errors; stating precision shows how much a reading can be trusted. |
A species–area curve for a grassland. Beyond about , larger quadrats add very few new species, so a quadrat of about that size is adequate.
Place quadrats by eye, then with the random-number generator. Then try the sample size, counting and calibration strategies: which one changes the estimate most?
Two students count the same quadrats of a grassland and get very different numbers of clover plants. What would most improve the agreement between their counts?
Habitats, distribution patterns and sampling
An ecosystem is not uniform. It is a patchwork of habitats (a dune swale, a rock platform, a stretch of forest floor) and smaller microhabitats within them (the shade under a log, a rock pool, the leaf litter at the base of a tree). Each has its own conditions, so species are not spread evenly. Their distribution pattern is usually one of three kinds.
Uniform, random and clumped distributions, each with 16 individuals.
| Pattern | Usual causes | Effect on sampling |
|---|---|---|
| Uniform | Competition or territorial behaviour keeps individuals evenly spaced (nesting seabirds, some desert shrubs whose roots take all the water nearby). | Easy to sample: most quadrats give similar counts. A systematic grid with the same spacing as the plants can give a biased result. |
| Random | Resources are spread evenly and individuals neither attract nor repel each other (wind-dispersed seedlings). | Random quadrats give a fair estimate with a moderate number of samples. |
| Clumped | Resources or microhabitats are patchy (moist hollows, shade, rock crevices), offspring stay near the parent, or animals live in groups. | Counts vary hugely between quadrats: many zeros and a few very high counts. Many quadrats are needed for a reliable mean, and a few randomly placed ones can miss the clumps or land on them by chance. |
Definition
Reliability and validity in fieldwork
- Reliability is about consistency: would repeating the sampling give similar results? It improves with more samples, agreed counting criteria and calibrated equipment.
- Validity is about whether the method measures what the investigation set out to measure, so the conclusion is justified. A survey of a whole dune system is not valid if every quadrat fell in the swale, and a comparison of two sites is not valid if they were sampled with different quadrat sizes.
Worked example
Design a survey of mat-rush in a reserve
A council wants to estimate the population of spiny-headed mat-rush (Lomandra longifolia) in a 2.0 ha bushland reserve. The reserve has two habitats: 1.5 ha of dry eucalypt open forest on a ridge, and 0.5 ha of moist gully along a creek, where mat-rush grows in dense clumps. A team has time to sample 40 quadrats. Propose and justify a sampling design, including the method, technique and how bias will be minimised.
Step 1: Choose the method
Stratified random sampling. The reserve has two distinct habitats, and mat-rush is likely to be much more abundant in the gully. Sampling each stratum separately stops chance from putting most quadrats in one habitat.
A purely random design could, by chance, put few quadrats in the small gully and badly misjudge the clumped population there.
Step 2: Allocate quadrats in proportion to area
The ridge is of the reserve and the gully , so 30 quadrats go on the ridge and 10 in the gully.
Step 3: Choose the technique
Quadrats, because mat-rush is a plant that stays put. A quadrat suits a tussock this size; a species–area curve from a pilot survey would confirm it.
Step 4: Minimise bias
- Place each quadrat at coordinates from a random-number generator on a grid over each stratum.
- Counting criteria: count a tussock only if more than half its base is inside the quadrat; treat leaves from one base as one plant.
- Record the GPS precision of each position and use the same tape and quadrat for every sample.
- If the running mean in the gully is still changing after 10 quadrats, add more there: clumped plants need more samples.
Step 5: Explain the calculation
Find the mean density in each stratum, multiply by that stratum's area in m² ( and ), and add the two to estimate the total population.
Try it yourself first, then check the working one step at a time.
Answer
Stratified random sampling with 1 m² quadrats: 30 on the ridge and 10 in the gully (in proportion to area), positions from a random-number generator, agreed counting criteria, more gully quadrats if the running mean is unstable; population = sum over strata of mean density × stratum area.
Method
Name the method and justify it from the site (distinct zones → stratified; gradient → systematic; uniform → random).
Name the technique and justify it from the organism (plants → quadrats; mobile → capture–recapture).
Allocate samples in proportion to area.
List specific bias controls: random numbers, counting criteria, sample size and number, calibration.
Watch out for
Writing "random sampling" and then choosing spots by eye.
Calculating a single mean over both strata, which over-weights whichever stratum got more quadrats.
Stratified sampling in a forest
Strata can be zones across the ground or layers above it. A rainforest has an emergent layer, a canopy, a sub-canopy, a shrub layer and a ground layer. Sampling each stratum separately (with binoculars, a canopy crane or ladders above, quadrats and pitfall traps below) lets you:
- identify the habitats in the forest, one for each layer;
- measure abiotic factors in each stratum: light intensity can fall from full sun above the canopy to around 1–2 % of it at the forest floor, while humidity rises and temperature varies less;
- compare community composition: epiphytes such as staghorn ferns in the canopy, shade-tolerant seedlings and fungi on the floor;
- infer interactions within and between strata: canopy trees shade the ground layer and drop the leaf litter that decomposers break down;
- classify the ecosystem, using the height and cover of the tallest stratum (the lesson on classifying ecosystems).
A coastal dune system is 70 % hind dune woodland, 20 % swale and 10 % foredune. A student has time for 40 quadrats. Using stratified sampling in proportion to area, how many quadrats should go in the swale?
Exam tip
Justifying a sampling design
A full answer names the method, the technique and why each suits the site and organism, then says how bias is reduced. For example: "Stratified random sampling, because the dune has three distinct zones; 1 m² quadrats placed with a random-number generator within each zone, in proportion to its area; plants counted only if rooted inside; at least 10 quadrats per zone so the clumped spinifex does not distort the mean."
Summary
Random sampling suits uniform sites; systematic sampling suits gradients; stratified sampling suits sites with distinct zones or layers.
Quadrats sample plants and slow animals; line transects show presence along a gradient; belt transects show abundance along it; capture–recapture estimates mobile animals.
Minimise bias with adequate quadrat size (species–area curve) and number of samples, random-number generators, agreed counting criteria, and calibrated equipment with its precision recorded.
Habitats and microhabitats make distributions uniform, random or clumped; clumped distributions need many samples for a reliable result.
Reliability is consistency on repeating; validity is whether the design measures what the question asks.
Finished the lesson?
Try the 3 quick checks above and the lesson completes itself, or mark it complete now.