Moran's I: Do Your Values Cluster, or Are They Random?
Before you map hot spots or model spatial data, one question comes first: is there any spatial pattern at all? Moran's I gives a single, testable answer — whether similar values sit together, avoid each other, or fall where they may.
How it reads the surface
Moran's I correlates each value with the values of its neighbours, weighted by a spatial-weights matrix, and scales the result to roughly −1 to +1. Near +1 means strong clustering (highs by highs), near −1 means dispersion (a checkerboard), and around the expected value (≈ 0) means spatial randomness — judged against a z-score and p-value.
How it is calculated
What the numbers mean
→ +1 clustered · ≈ 0 (technically −1/(n−1)) random · → −1 dispersed / checkerboard. Judge against a z-score / p-value.
What it runs on
| Operates on | A point or polygon layer with a numeric attribute — not tied to specific spectral bands. |
Where it is used
It is the standard first test for spatial autocorrelation in ecology, economics and epidemiology, and before any regression on spatial data, where ignoring autocorrelation invalidates ordinary statistics.
Limitations to know
Global Moran's I is a single summary — it can report "no pattern" while strong local clusters cancel out. Follow it with a local indicator (LISA / local Moran's I), and remember the spatial-weights choice drives the result.
Compute Moran's I on your own study area
Skip the code. Draw or upload a boundary and Spatial Research Suite runs Moran's I on live data — with cloud masking, exports and citations built in.
Run this analysis in GISforus →Frequently asked
What does Moran's I actually measure?
Spatial autocorrelation — whether similar values cluster (near +1), disperse (near −1), or are randomly arranged (near 0), tested against a p-value.
Global or local Moran's I?
Use global Moran's I to test for any overall pattern; follow with local Moran's I (LISA) to locate the individual clusters and outliers.