ENV103 Chap.10 Spatial Analysis, Autocorrelation and Models
Spatial Analysis, Autocorrelation and Models
Spatial analysis turns layered data into explanation. This chapter defines it, works through Tobler's first law and spatial autocorrelation in its dispersed, random and clustered forms, explains why autocorrelation is not an error and why correlation is not causation, and follows the analysis workflow into the species distribution model of Aotearoa's burrowing seabirds.
A clinic page sets out the most common misreadings, and a worked counter-example shows how to test a tempting causal claim drawn from two matching maps.
What this chapter covers
- 01
Definitions of spatial analysis
- 02
Layering as the first step
- 03
Tobler's first law and distance decay
- 04
Dispersed, random and clustered patterns
- 05
Autocorrelation is not an error
- 06
Correlation versus causation
- 07
Models, the workflow and species distribution
Worked example · free
Classify three grid patterns by spatial autocorrelation
- +1Matching shares: 0 ÷ 60 = 0%, 32 ÷ 60 ≈ 53% and 54 ÷ 60 = 90%.
- +10% means neighbours always differ: a dispersed pattern with negative autocorrelation.
- +1About half matching is what chance gives with equal numbers of each colour: a random pattern, no autocorrelation.
- +190% matching means near cells are alike, a clustered pattern with positive autocorrelation, the situation Tobler's law describes.
Key terms
- Spatial analysis
- Spatial analysis is the process of turning spatial data into useful information for discovery or decisions.
- Spatial autocorrelation
- Spatial autocorrelation is the similarity or dissimilarity of values according to their locations.
- Distance decay
- Distance decay is the tendency for observations to become less alike as the distance between them grows.
- Moran's I
- Moran's I is a statistic that measures the degree of spatial autocorrelation in a dataset.
- Species distribution model
- A species distribution model combines presence records with environmental predictors to predict where a species can occur.
- Correlation
- Correlation is a statistical association between two variables, which on its own does not establish cause.
Spatial Analysis, Autocorrelation and Models FAQ
What is Tobler's first law of geography?
Everything is related to everything else, but near things are more related than distant things. It underpins spatial autocorrelation and explains why neighbouring areas often share characteristics, which the lecture illustrates with a map of the Index of Multiple Deprivation.
Is spatial autocorrelation a problem to remove?
Not in itself. It is a real property of the world and the reason maps of clusters are informative. It becomes a problem only when it violates the assumptions of a statistical model, leading to wrong inferences unless it is corrected for, and corrections depend on scale and resolution.
Why can two matching maps not prove a cause?
Two variables can share a pattern because both respond to a third factor, or because both are autocorrelated. Correlation shows association only. A causal claim needs a mechanism, a test that rules out alternatives and evidence beyond the shared spatial pattern.
How does a species distribution model work?
It combines presence records of a species with environmental predictor layers and fits a model that predicts where conditions suit the species. In the Week 5 seabird case, presence data from fossils, historic records and modern surveys were layered with predictors such as elevation, distance to the coast, slope, windspeed and summer rainfall to map a predicted breeding distribution.
Assessment move
Build this chapter around ideas you can explain without notes. Practise saying what spatial autocorrelation is, why Tobler’s law produces it, which direction a pattern shows and why it is not an error, then apply it to a map you have made and explain what it would do to a model. Sketch the three patterns, dispersed, random and clustered, and label each with its sign.
Next, learn the analysis workflow as a sequence you can recite: question, data collection, preparation, exploration, visualisation, model and interpretation. Apply it to the seabird case by naming the presence data, the environmental predictors and the output, and note why the presence points had to be reprojected before layering.
For correlation and causation, practise spotting causal leaps in your own map interpretations; every time you write “because” about a pattern, ask what evidence would rule out a third factor. Finally, watch the Week 8 interviews on how scientists use models and write one sentence for each on what the model simplifies.
The habit of naming what a model leaves out will strengthen the conclusion of your poster as well as your quiz answers.