Dear all, I would like run spatial autocorrelation analysis with my data in R (or other software such as Minitab, Past or Python). My data comprise 100 1m2 plots with control paired plots 1m far away treatment. In all plots I measured plant cover and I want to measure species co-ocorrence in each plot. All plots are georeferenced with lat and long in degree, minutes and seconds. I want know if had autocorrelation in my sampling. Can someone help me?
Best wishes,
Jhonny
Hi Jhonny. What measure of spatial autocorrelation will you be using? Maybe Moran's I or Geary's C?
http://www.ats.ucla.edu/stat/r/faq/morans_i.htm
https://www.youtube.com/watch?v=NLyjdmyokio
Hi Eric! Thanks! I read that Moran's I is more suited for data as mine.
So you want to test for a difference between control and treatment, but you suspect that pairs of plots are not independent replicates?
As an exploratory exercise, imagine looking at the difference in plant cover between all possible pairs of plots. Then plot a metric of differences over the distance between the pairs of plots. If you have autocorrelation, then differences between near-by plots should be small. Differences between plots should increase with distance between plots until space has no effect and you just have background randomness between plots. This is essentially what a semivariogram does. An advantage of a semivariogram is that it is a model of the distance. It's not a single number, it is actually something you can use to "remove" the effect of autocorrelation if you want. There are many ways to do this in R. Just google search for R and semivariogram.
If you need to do something more like inference or hypothesis testing, then a Conditional Autoregressive (CAR) model may be appropriate. The idea here is that the value at any given location is partially explained by the values at nearby location. You would want to to "subtract" that effect of space and then perform the analysis on whatever is left over.
Good luck!
Hi Jhonnhy.
You can fit linear models with correlation structures for the error using package nlme (https://cran.r-project.org/web/packages/nlme/nlme.pdf). There is an argument `correlation` in the `lme` function to model spatial correlation. Also the function `Variogram` is used to compute the semi-variogram. Argument `form = ~ x + y` represents a two-dimensional position vector with coordinates x and y, which I think is your case.
Yes Raúl. This is my question! Thank you for your answer. Can you help me more information? How I must organize my matrix (or dataframe)? For exemplo, is it correct site in rows and coordinates in columns? Sorry, but I'm beginning in this analyses!!!
Matt, thank you for your answer. My problem is not about my pairs, but about the distance between plots. My study site have 20 ha and my plots are distributed for whole area.
You need a data frame with your data (response variable and factors) and two additional columns for coordinates (it can be any distance measure, e.g. meters). But there are several issues to consider when selecting a correlation structure, which are still not clear for me. Here you have an example using a Gaussian structure:
http://www.ats.ucla.edu/stat/r/faq/spatial_regression.htm
Thank you so much! I will learn about this before my analyses!
Good morning everybody! I want ask some help, how can I analysis correletion between spatial nc data and time series (txt or excel) data, I need spatial result? Can I run in the Minitab?
- Badges
- Science topic
- Similar topics
- Filing