Spatial autocorrelation describes the extent to which a variable is either dispersed or clustered through space. In the case of a raster image, spatial autocorrelation refers to the similarity in the values of nearby grid cells. This tool measures the spatial autocorrelation of a raster image using the global Moran's I statistic. Moran's I varies from -1 to 1, where I = -1 indicates a dispersed, checkerboard type pattern and I = 1 indicates a clustered (smooth) surface. I = 0 occurs for a random distribution of values. image_autocorrelation computes Moran's I for the first lag only, meaning that it only takes into account the variability among the immediate neighbors of each grid cell.

The user must specify the names of one or more input raster images. In addition, the user must specify the contiguity type (contiguity; Rook's, King's, or Bishop's), which describes which neighboring grid cells are examined for the analysis. The following figure describes the available cases:

Rook's contiguity

...
010
1X1
010

Kings's contiguity

...
111
1X1
111

Bishops's contiguity

...
101
0X0
101

The tool outputs an HTML report (output) which, for each input image (input), reports the Moran's I value and the variance, z-score, and p-value (significance) under normal and randomization sampling assumptions.

Use the image_correlation tool instead when there is need to determine the correlation among multiple raster inputs.

NoData values in the input image are ignored during the analysis.

See Also

image_correlation, image_correlation_neighbourhood_analysis

Function Signature

def image_autocorrelation(self, rasters: List[Raster], output_html_file: str, contiguity_type: str = "bishop") -> None: ...

Project Links

WbW Homepage User Manual Support WbW