Hi everyone.
I have a problem to interpret the result of spatial auto-correlation (Moran's I versus Getis-Ord General G) for species distribution.
In my analysis, the result of Moran's I indicating that there are some clustering (I=0.226, Z=5.194, P=0.000), but Getis-Ord General G indicating that there are no any clustering using same data (G=0.002, Z=0.725, P=0.469).
So, I did mapping clustering methods, such as Cluster and Outlier Analysis (local Moran's I) and Hot Spot Analysis (Getis-Ord Gi*). In the Hot Spot Analysis, there are two hotspots with high value, and one hotspot with low value.
My question is...Why spatial clustering did not found in G index?
Is it because two different spatial hotspots?
Thank you.
It's hard to tell what is happening without seeing the data. Both Moran's I and Getis-Ord General G presume isotropy and stationarity on the data. If there is only two hotspots with high value, and one hotspot with low value on a large area, it's not weird that General Getis-Ord didn't capture anything, as it is a global index.
Actually, these indexes are searching for patterns on different scales. I don't think it is a case that one of them is right and the other is wrong. They are just looking for different things.
Marcos, thank you for your comments that very helpful to me understanding interpretation of different results between Moran's I and General G.
Thank you.
I second Marcos on this. But just to add a little thing. It's quite hard to interpret the global test of Moran's I alone. I highly recommend that you run a autocorrelogram to see at which distance class you find positive correlation. That would give you much more precise description of your data.
Thank you Diogo for your suggestion. As I understand it, your comment indicates that I need a further analysis on the autocorrelogram using an Incremental Spatial Autocorrelation in ArcMap 10.1. Is it right?
If my understanding is correct, I run the autocorrelogram and got the result like attached figure (Incremental.jpg). In this case, all Moran's Indices along a distance class indicate that there were spatial clustering with significant p value. Am I did it correctly?
Still, I'm not sure how can I interpret the result of autocorrelogram in my case with an result of hotspot map (black point indicates High-High value derived by local Moran's I). What the results mean exactly?
Yes, the correlogram is capturing exactly the spatial structure and showing that you have a low, positive spatial autocorrelation all distance classes. I have never used ArcGIS to run a autocorrelogram, so I don't know if the black points in the graph are representing significant values.
Take a look at these two websites that show how the autocorrelogram behaves regarding the spatial structure of your data. Ch 7 of Legendre & Legendre 2012 has much more in depth explanation too.
http://www.petrkeil.com/?p=1050
http://rstudio-pubs-static.s3.amazonaws.com/9688_a49c681fab974bbca889e3eae9fbb837.html
Diogo, the websites will be useful to me for explanation on autocorrelogram.
Thank you for your help.
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology