When using ArcGIS for spatial statistics and analysis, many users are exposed to two key functions: "spatial grid partitioning" and "spatial autocorrelation". Whether conducting point density analysis, grid calculation, or region clustering, the reasonable setting of grid size directly affects the accuracy and effectiveness of the results. And "spatial autocorrelation" analysis (such as Global Moran's I), although the calculation results intuitively display the degree of spatial clustering, its statistical significance and interpretation of output parameters are not easy for beginners.
一、Reasons and solutions for invalid spatial grid size in ArcGIS
Many users may find that the set grid size does not work when using "Fishnet" or "Point Cluster Grid Analysis" in ArcGIS, or the tool outputs empty or even reports errors after running. This usually stems from several easily overlooked reasons.
1. The coordinate system unit is inconsistent with the grid size unit
If your layer uses a geographic coordinate system (such as WGS84), the unit is "degrees" instead of "meters" or "kilometers". If you directly set the grid size to 100, ArcGIS will assume it is 100 degrees, resulting in a grid that is too large to see data coverage.
Solution: Project the layer onto a plane coordinate system, such as UTM or CGCS2000, so that the unit becomes meters and the grid size setting is meaningful.
2. The grid size is far from suitable for the data distribution range
When the grid is too small, ArcGIS may fail due to dense data; When the grid is too large, it may result in all elements falling into only one grid, losing analytical value.
Suggestion: Based on the research area, it is reasonable to use a grid of 100-500 meters for urban areas, 1-5 kilometers for county areas, and more than 10 kilometers for national level areas.
3. There is no spatial overlap between the grid and the data
If the data points are too concentrated or the starting point for grid generation is set incorrectly, it may result in the generated grid not covering the data area at all.
Solution: Use the "Minimum External Rectangle" tool to obtain the data coverage range, and then manually set the grid starting point and number of rows and columns based on this.
4. Cache or layer not updated causing tool failure
ArcGIS sometimes has set parameters that do not actually take effect due to temporary caching or layer locking.
Suggestion: Clean up temporary folders, close layers and reload, or restart the software before running the tool.
二、How to view ArcGIS spatial autocorrelation results
Spatial Autocorrelation is a statistical method used to determine the clustering and randomness of a phenomenon in space. The most commonly used tool in ArcGIS is Global Moran's I, which can determine whether there is spatial clustering in the entire region.
1. How to interpret Moran's I value
The value of this index is between -1 and+1.
A value close to+1 indicates that similar values tend to cluster together in space (such as high values clustered together).
A value close to -1 indicates that the distribution of similar values is very scattered (spatially distributed in a discrete or checkerboard pattern).
A value close to 0 indicates a random distribution in space, without clustering or discrete trends.
2. What do you think of Z-score and P-value
The Z-score represents the degree of deviation between the current observed value and the theoretical expectation. The higher the Z-score, the more significant the deviation from the random distribution.
The P-value measures the probability of this outcome occurring. If the P-value is less than 0.05, it indicates that the result is statistically significant and not "accidental".
3. Output chart information in the report
After running, ArcGIS usually automatically opens an HTML report containing Moran's I value, Z-score, P-value, and spatial random distribution simulation graph.
Comparing the point cloud in the figure with the red line, we can intuitively see the degree of autocorrelation of the current data: the more the point cloud deviates from the centerline, the stronger the clustering.
4. Pay attention to data preprocessing before analysis
In order to make the results of spatial autocorrelation analysis more reliable, it is recommended to clean up outliers, unify projections, eliminate outliers, and select meaningful fields for analysis in advance.
For example, when analyzing the "housing price distribution", the "housing price" field should be selected instead of "number of floors"; When analyzing 'crime density', the number of cases per unit area should be selected instead of the total number.
三、How to determine if grid and autocorrelation analysis are suitable for your data
Not all spatial data is suitable for grid analysis and spatial autocorrelation. Some situations require special attention:
1. Too little or too sparse data volume
There are only a few dozen points, so doing autocorrelation may be meaningless, and Moran's I value tends to be random.
2. The spatial distribution itself is unstable or fragmented
If there are multiple isolated areas in a city that cannot form a continuous clustering pattern, it is recommended to use local Moran's I instead of global analysis.
3. Attribute fields have no spatial meaning
Such fields as ID number, customer number and order number are not suitable for spatial analysis. You need to select fields that have spatial dependencies, such as price, temperature, and accident rate.
4. Analyze areas that are too large or cross coordinate systems
If a layer contains multiple countries or provinces at the same time, the coordinate system may not be consistent, and it is necessary to unify the coordinate system before analysis.
