The world is too large to measure everywhere, so Geography must decide where to look.
We cannot place a rain gauge on every roof, test soil in every square metre, count every tree every day or measure air quality on every street continuously. Every field survey, sensor network and monitoring programme therefore makes a sampling decision.
Spatial sampling is the design problem of deciding which locations, times and units to observe so that the resulting evidence can represent a larger geographic area as faithfully as possible.
Before a map can be biased by analysis, it can already be biased by where we chose to measure.
Quick Read: The Sampling Mechanism
QUESTION → TARGET POPULATION / LANDSCAPE → SAMPLE FRAME → DESIGN → LOCATIONS → MEASUREMENTS → WEIGHTS / INFERENCE → UNCERTAINTY
1. Sampling Begins With the Question
A design suitable for estimating average soil moisture may be poor for detecting rare contamination hotspots. A network built to track long-term climate may be poor for capturing street-level heat. Sampling quality depends on the job.
2. The Sample Frame Defines What Could Be Chosen
If some places are absent from the frame, they cannot enter the sample. Inaccessible slopes, private land, unsafe areas or missing addresses can therefore create geographic blind spots before data collection begins.
3. Convenience Sampling Is Easy and Dangerous
Measuring only beside roads, schools or accessible buildings can make fieldwork efficient while systematically missing remote or unusual environments. Convenience is not randomness.
4. Random Sampling Protects Against Some Bias
Randomly selecting locations gives eligible sites a known chance of inclusion. This supports statistical inference, but pure randomness can leave important subregions sparsely sampled by chance.
5. Systematic Sampling Creates Coverage
A regular grid or fixed interval spreads observations across space and can produce excellent coverage. Yet systematic designs can interact badly with periodic spatial patterns if the sampling interval accidentally aligns with them.
6. Stratified Sampling Protects Important Subgroups
If a study area contains coast, forest, industry and housing, researchers may sample separately within each stratum. This prevents a large common land type from overwhelming smaller but important environments.
7. Cluster Sampling Can Reduce Field Cost
Researchers may select a smaller number of areas and collect many observations inside each. This reduces travel, but nearby observations may be similar, so the effective amount of independent information can be lower than the raw sample count suggests.
8. Adaptive Sampling Follows What the Data Reveal
If early measurements reveal an unexpected hotspot, later sampling can concentrate nearby to map it better. Adaptive designs are powerful when rare or clustered phenomena matter, but the changing selection process must remain explicit.
9. Sample Density Must Match Spatial Variation
A slowly varying temperature field may need fewer points than a highly fragmented pollution landscape. Sampling must be dense enough to resolve the spatial scale of the process under study.
10. Spatial Autocorrelation Changes Sample Efficiency
If neighbouring observations are highly similar, placing sensors too close together can produce redundant information. Spatial Autocorrelation therefore affects how far apart samples should be.
11. Time Is Part of Spatial Sampling
A site measured once at noon may not represent nighttime conditions. Rainfall, traffic, heat, pollution and human activity change through time. A spatial design that ignores temporal variation can misrepresent the geography.
12. Missingness Can Be Geographic
Sensors fail more often in harsh environments. Surveys may receive lower response in some communities. Mobile-phone data represent device users better than non-users. Missing data are dangerous when the missingness itself has spatial structure.
13. More Samples Are Not Automatically More Representative
One thousand measurements from the same accessible corridor may represent the wider region worse than one hundred well-distributed observations. Sample size and sample design are different dimensions of evidence quality.
14. Spatial Sampling Is Not the Same as Spatial Data
Spatial Data owns the broader question of how geographic observations become datasets and map layers. Spatial sampling owns the narrower design decision: which observations enter the dataset in the first place.
15. Spatial Sampling Comes Before Interpolation
Spatial Interpolation estimates between measured points. If the sampling pattern is poor, even a sophisticated interpolator begins with weak evidence.
16. Primary Geography: Where Should We Put Five Thermometers?
Give children a playground map and only five thermometers. If all five go under the same shelter, they learn little about the whole playground. Spread them across sun, shade, grass, concrete and buildings and the design becomes more informative.
17. Secondary Geography: Design a Fair Fieldwork Sample
Students can compare random, systematic and stratified ways of sampling pedestrian counts, noise or vegetation. The key question is not which method is universally best, but which method matches the inquiry.
18. Advanced Geography: Probability Designs and Inclusion Weights
When locations have known probabilities of selection, analysts can use inclusion probabilities and weights to estimate wider population characteristics. This is one reason probability sampling remains so valuable for defensible inference.
19. Singapore Example: Urban Heat Monitoring
A heat-monitoring network concentrated in open parks would underrepresent dense street canyons. One concentrated along major roads could overstate traffic influence. A defensible design should capture relevant urban forms, land covers, coastal influence and time periods.
20. Singapore Example: Pedestrian Counts
Counting pedestrians only during weekday lunch near office towers cannot represent weekend movement in residential or recreational areas. Where and when the count occurs defines the population of movement being observed.
21. Environmental Monitoring Example
USGS guidance on monitoring design emphasises defining questions, objectives, reporting units, sample frames, design type, stratification, allocation and sampling effort. The sequence matters because field measurement should follow the decision problem rather than precede it.
22. Hostile Test: “We Collected Thousands of Data Points”
From where? At what times? Under which inclusion rule? Thousands of observations can still be biased if they come from a narrow, self-selected or easily accessible slice of the landscape.
23. Where Spatial-Sampling Reasoning Breaks
- Large-sample worship: assuming quantity guarantees representativeness.
- Roadside bias: sampling where fieldworkers can easily reach.
- Temporal blindness: treating one time period as universal.
- Frame blindness: ignoring places that were never eligible to be selected.
- Cluster redundancy: counting nearby similar observations as fully independent information.
- Method-before-question: choosing a familiar sampling design before defining the inquiry.
24. Ten Questions for Spatial Sampling
- What geographic question are we answering?
- What landscape or population should the sample represent?
- What is the sample frame?
- Which places are excluded?
- Should selection be random, systematic, stratified, clustered or adaptive?
- How dense must observations be?
- How does spatial autocorrelation affect spacing?
- Which times need to be sampled?
- Where might missingness be spatially biased?
- What claims can this design legitimately support?
25. Research Anchor
U.S. Geological Survey sampling-design guidance for landscape monitoring treats question definition, reporting units, sample frame, design type, stratification, allocation and sampling effort as explicit parts of the design workflow. That is the operational discipline behind good spatial sampling.
26. Where This Fits
Spatial Data owns geographic data broadly. Spatial Interpolation owns prediction between measured sites. This article owns the prior decision of where, when and how observations are selected.
The Idea to Keep
A map can only know the world through the places someone decided were worth measuring.