Autocorrelation can be tested using an index called Moran's I (objectively). Subjectively you could also check by doing a variagram (variance between points vs seperation distance) and then try to notice whether there is any "structure" to the data -> lower variance in shorter sep. distance, and increasingly higher correlation in longer sep. distance, until at a point, the correlation levels off. I am less unsure about how to handle it in Maxent. I just remember some terms from long time ago , which may be relevant for you: Ripley's K function (to understand points pattern). But less sure of this. But please feel free to check this link where there is more discussion: https://www.researchgate.net/post/Accounting_controlling_spatial_autocorrelation_in_Maxent

Dear Angeliki,

In terms of spatial autocorrelation calculation, it doesn't matter if you use Maxent or other SDM method. And spatial autocorrelation of the input data is not a problem (indeed, it is normal). You should calculate spatial autocorrelation rather on the residuals of the SDM model. For this, follow these steps:

- train the SDM,

- predict the SDM to the observation points, and calculate the prediction-observation difference (i.e residual), if it cannot be taken out from the trained model object directly,

- calculate spatial autocorrelation of the residuals.

If the residuals are in RasterLayer fomat, use raster::Moran(), if they are points (which is your case), use spdep::moran().

HTH,

Ákos

Spatial autocorrelation is usually when areas close together have similar values, so that's not an issue for a single large area. Temporal autocorrelation refers to the relationship between successive values of the same variable. When temporal autocorrelation is statistically significant , it could adversely bias the parameter estimates.