Mugumaarhahama Yannick

Yes! Spatial auto-correlation matters a lot! I replied to a similar question couple of year before, please find the link to it below

https://www.researchgate.net/post/How_do_we_use_spatial_autocorrelation_in_reality

Regards

you can deal this spatial auto-correlation effect:

1) by avoiding it using efficient sampling (combining systematic and random sampling at least 10th order lag distance)

2) by considering it effect in modelling

The first method is widely used and easy to perform.

Hope this help!

The question and the first answer suggest that an empirical confirmation of a negative impact of a biased set of point observations on the spatial modelling result is hard to find in the literature. I agree.

From first (statistical) principle, avoiding of geographic bias in observation seems advisable. We used a range of approaches to avoid such bias. In our mountain gorilla (Rwanda) and Phytophthora (Spain/Australia) articles, we used data-sets obtained by systematic sampling methods; in our endemic plant SDM (Italy), a database containing incidental observations from a wide range of observers over a long time period; in our bear research, incidental data from contrasting categories of observers recorded over years and applied a thinning/filtering of observation points that were found to be close in space or time; finally, in our African Swine Fever (ASF, China) paper, we applied a numerical method of spatial filtering/thinning of incidental observation points. Thinning/filtering assumes a minimum number of points, that may vary between SDM algorithms.

looking for forward to further responses to the question

Hello Mugumaarhahama, I am working with Spatial Point Processes but coming from a geoscientific background. Therefore my understanding of the whole picture is not as deep as the one that a statistician would have.

However, to my best understanding, the theoretical background upon which a Poisson Process is based assumes independence in your number of points. In turns, this is what defines/allows you to work with a Poisson distribution. Therefore, my answer would be yes, if you model your autocorrelated counts then you are neglecting part of the assumption of the model you use. As for how to deal with it, I do not have a suggestion because I have never had to deal with it. I can ask to a couple of statisticians that I work with if you are still interested. Just let me know. And sorry for not being of much more help than this.

Luigi

This book explains everything.

Spatial point patterns : Methodology and applications with R (Interdisciplinary statistics).

Baddeley, A., Rubak, E., & Turner, R. ( 2015). Spatial Point Patterns: Methodology and Applications with R.