I see the importance of calculating spatial autocorrelation for species occurrence data, in order to guarantee spatial independence. However, I do not understand how this is done. Could any of the colleagues working on the topic help me?
From now on I thank you.
You should consider several aspects of spatial autocorrelation. That is, a first element is that of spatial autocorrelation of the environmental data using approaches like a semivariogram and spatial lag calculations. For exampleias , if you look at climate variation across a small area, there may be strong autocorrelation, such that environmental is very smooth across the area. If you look instead at satellite imagery of land cover over the same region, there may be much more detail and variation, such that more information is available. These analyses can help you to set a minimum separation distance between points, as points closer than that distance will not present independent samplings of environments.
Second, you may wish to seek clumping (autocorrelation) of the occurrence points. This will involve statistics such as Ripley's K. This may reflect differential reporting from a state or country, and these concentrations should probably be reduced further, to avoid overrepresenting particular regions and the associated environments.
Remember, autocorrelation can cause two sorts of problems: (1) bias in niche estimates owing to artificial over- and under-representation of particular environments, and (2) inflation of sample sizes and correspondingly artificial inflation of statistical power.
For example: I calculate the autocorrelation for the occurrence points, right? Disregarding environmental data.
Dear Joaquin,
Thank you for sharing your work!
But I can not understand one thing: when you do the semivariogram, to find the range, do you extract xi from the other xi, meter by meter, right? And for each variable? But, do you considered just the cells where do you have the species? I can't understand this! Because there is no semivariogram metric that considers dependence only on points of presence in space, or there is? I'm in trouble in this part. In the case, when we say "the points must be independents", actually what we are looking for is the independence of the conditions of the environmental variables.
And sorry about my english, I'm tring to improve this.
Hugs.
Marília
You should consider several aspects of spatial autocorrelation. That is, a first element is that of spatial autocorrelation of the environmental data using approaches like a semivariogram and spatial lag calculations. For exampleias , if you look at climate variation across a small area, there may be strong autocorrelation, such that environmental is very smooth across the area. If you look instead at satellite imagery of land cover over the same region, there may be much more detail and variation, such that more information is available. These analyses can help you to set a minimum separation distance between points, as points closer than that distance will not present independent samplings of environments.
Second, you may wish to seek clumping (autocorrelation) of the occurrence points. This will involve statistics such as Ripley's K. This may reflect differential reporting from a state or country, and these concentrations should probably be reduced further, to avoid overrepresenting particular regions and the associated environments.
Remember, autocorrelation can cause two sorts of problems: (1) bias in niche estimates owing to artificial over- and under-representation of particular environments, and (2) inflation of sample sizes and correspondingly artificial inflation of statistical power.
Dear Dr. Andrew Townsend Peterson,
Thanks for your ansewer!
I will explain the purpose of my work to you, to see if my logic is correct, all right?
The goal of my work is to model the niche of a moth species throughout Brazil, a moth that causes injuries to human health.
I have species occurrence points correctly geo-referenced from southern Brazil to the northeast of the country, however there are few points throughout this region and they are a little scattered in space. And I only have presence data ranging from 1995 to 2015. In total there are 36 correctly geo-referenced presences.
To use some data in the training and others in the test, during the niche modeling, I need to show that these two data are spatially independent.
Well, I also leave my apologies for such basic questions, but I am looking for the information according to the suggestions in the topic!
Thank you again!
Hugs.
Dr. Janet Lane,
Thanks for your ansewer!
I'll look the papers by Dr. Kumar, Thanks!
I already applied Ripley's K to my data, but it does not cost me to read more about it!
Hugs.,
Marília
Dear Marilia,
In our study (see attached paper), we checked for spatial autocorrelation of the presences. As in your case, we also had a few presences.
I hope it helps.
xavier
Article Declining relict plants: Climate effect or seed dispersal di...
Dear Xavier Rotllan-Puig,
Thank you for sharing your work! I promise that I will read it carefully and soon I return with my doubts here in the topic.
Hugs,
Marília
Dear Jane, sorry about that. I went to write "dear" and I forgot what I was doing. But, I think you'll be a great doctor, I swear!
So, about the doubt... I maked the Ripley's K in ArcGis 10.4. The software have a page suporte for that: http://pro.arcgis.com/en/pro-app/tool-reference/spatial-statistics/multi-distance-spatial-cluster-analysis.htm.
Again sorry.
Hugs,
Marília
- Badges
- Science topic
- Similar topics
- Applied Mathematics
- Calculations