Sampling Landscape soils and surface environments

Landscape soils and surface environments - Week 3 Workshop 1b

Raphael Viscarra Rossel, Lewis Walden

2026-03-03

Where do we sample?

We’ve learned WHY and HOW to measure. Now WHERE to sample, considering that:

  • Landscapes are spatially heterogeneous (Week 1)
  • Soil and vegetation vary with landscape position (Week 2)
  • Measurement is expensive and time-consuming (Last hour, Workshop 1a)

Important

We cannot (easily) measure everywhere | Sampling helps to determine what we learn about our environment

Why sampling design matters

Statistical goal:

  • Represent the whole landscape from limited samples
  • Estimate with known uncertainty
  • Detect real patterns, test hypotheses, etc.

Ecological goal:

  • Capture functional differences across landscape positions
  • Understand spatial controls on processes
  • Provide baselines, monitoring change, etc.

Caution

Poor design → biased estimates → wrong conclusions → poor management

What can go wrong: innacurate sampling (soil, vegetation, water…)

Sampling bias: Systematic over- or under-representation of part of the landscape, e.g.

  • Only sampling near roads → miss remote areas
  • Only sampling flat ground → miss slopes
  • Only sampling where soil is easy to dig → miss rocky/compacted areas

Result: Wrong conclusions, harmful management decisions, etc.

Important

Sampling design exists to minimise bias while working within practical constraints

Three main sampling approaches

1. Random sampling

All have equal probability | No spatial structure | Unbiased

2. Stratified sampling

Divide into meaningful units | Sample within strata | Uses prior knowledge

3. Systematic sampling

Regular spacing | Even spatial coverage | Good for interpolation | NOT unbiased

Approach 1: Random sampling

Every location has equal probability of selection (unbiased)

How it works:

  1. Define study area boundary
  2. Randomly generate X, Y coordinates
  3. Sample locations are independent
  4. Valid statistical inference (hypothesis tests, CI, etc.)

Note

No prior knowledge required

Random sampling: Trade-offs

Strengths:

  • Unbiased estimator of population mean
  • Valid statistical inference
  • Simple, transparent, defensible

Weaknesses:

  • May cluster by chance → miss variation
  • Inefficient for structured landscapes
  • Ignores spatial autocorrelation

Best when: Little prior knowledge or unbiased mean is the objective

Approach 2: Stratified sampling

Divide landscape, sample within each

How it works:

  1. Classify landscape into strata (GIS, remote sensing, field)
  2. Allocate samples to each stratum based on variability
  3. Sample randomly within strata

What defines strata?

Prior knowledge: landscape position (catena), vegetation type, soil type, land use, geomorphology, etc.

Allocating samples to strata

Once strata are defined — how many samples in each?

Strategy Approach Best for
Equal Same n per stratum Comparing between units
Proportional n reflects area of stratum Landscape-wide estimates
Optimal More n where variance is higher Maximising precision per $

Example equal: 5 catena positions × 10 samples each = 50 total

Example proportional: SCP covers 60% of area → 60% SCP, 40% DR

Stratified sampling: Trade-offs

Strengths:

  • All landscape units represented
  • More efficient than random (lower variance)
  • Can compare between strata
  • Uses landscape knowledge effectively

Weaknesses:

  • Requires prior knowledge or maps
  • Quality depends on stratification scheme
  • Strata boundaries may be fuzzy
  • May miss within-stratum variation

Best when: Clear landscape units exist (catenas), prior knowledge available, comparing between units, or budget is limited

Approach 3: Systematic sampling, e.g. grid

Regular sample spacing across the landscape.

How it works:

  1. Define grid spacing (e.g. 50m × 50m)
  2. Place sample points at grid intersections
  3. Optionally add random offset within cells
  4. Spacing should match scale of variation
  • Too coarse → miss spatial patterns Too fine → redundant samples (wasted budget)

Grid sampling: Trade-offs

Strengths:

  • Even spatial coverage
  • Good for interpolation (mapping)
  • Captures spatial patterns
  • Predictable field logistics
  • Repeatable

Weaknesses:

  • May align with landscape features (bias)
  • Fixed spacing may not suit variable landscapes
  • Can miss features between grid points
  • Assumes no periodicity in landscape

Best when: Objective is creating maps, modelling spatial structure, or no reliable stratification available

Which design for which objective?

Example objective Best design Why
Estimate landscape mean pH Random Unbiased, valid inference
Compare soil across vegetation types Stratified (by veg) Ensures each community sampled
Map soil clay content in a paddock Grid (30–50m) Supports interpolation
Map vegetation community boundaries Stratified + grid within Captures transitions
Baseline survey, no prior knowledge Random No assumptions needed


Often hybrid: stratified by landscape unit, soil or veg., random or grid within each stratum

Reducing soil micro-scale variability: Compositing (bulking)


Soils: Compositing or bulking

Collect multiple cores (3–10), mix into one sample

Averages out micro-scale variability

One lab analysis → cheaper

Trade-off: lose within-plot variability

Reducing vegetation micro-scale variability: Subplots or quadrats


Vegetation: Subplots/quadrats

Multiple small quadrats within a larger plot

Average cover, species counts across subplots

Captures patch-scale heterogeneity

Trade-off: miss rare species between quadrats

Same principle — replicate within location to get a representative measurement

How many samples?

Depends on:

  • Landscape variability — more heterogeneous = more samples
  • Desired precision — narrower confidence intervals = more samples
  • Budget and time — the practical constraint

Rules of thumb (not strict guidance):

Design Minimum Rationale
Any ~30 total Central limit theorem
Stratified ~20 per stratum Meaningful comparison
Grid Spacing ≤ half the scale of variation Capture spatial pattern

Indigenous knowledge and sampling design (prior knowledge)

Reading Country as stratification:

  • Identifies functionally distinct landscapes
  • Vegetation signals environment conditions
  • Fire history defines management zones
  • Water-associated areas vs upland areas

Two-way approach:

  • Western: DEM, soil maps, satellite imagery
  • Indigenous: landscape reading, seasonal indicators

Note

Indigenous knowledge may identify contextual variation that other methods may miss

Key takeaways

  • Spatial variability drives sampling needs

  • Three main approaches, each with trade-offs:

    • Random: Unbiased, simple, inefficient for structured landscapes
    • Stratified: Uses prior knowledge, efficient, requires classification
    • Systematic: Good for mapping, even coverage, may miss some patterns

  • Match design to objective

  • Use prior knowledge including Indigenous knowledge, etc.

  • Must also meet practical considerations

Integration with unit concepts

  • Spatial thinking - Informs scale of sampling, stratification approach
  • Soil formation - Topography controls properties - Stratify by position
  • Catenas - Systematic hillslope variation - Transect or stratified sampling
  • Indigenous knowledge - Alternative stratification schemes, functional landscape units
  • Measurement costs - Efficiency matters, design affects sample size feasible

Activity (20 min): Comparing sampling designs in the Scarp area

Objectives:

  1. Implement random, stratified, systematic sampling
  2. Apply to Perth-Darling Scarp landscape data
  3. Compare performance:
    • Which design best estimates true landscape mean?
    • Which is most efficient (precision per sample)?
  1. Visualise sampling designs and resulting estimates
  2. Discuss advantages and disadvantages of each design in this context
  3. Discuss when to use each approach

You’ll work with:

  • Soil clay content and organic carbon data
  • R code templates for each sampling design

Leave a comment or question