Problems with ML methods at the edge of maps

[bniebuhr]

The user Laura Nery Silva commented on an issue dealing with corridor simulations when the source-target (ST) points are near the edges of the resistance map (closer than the scale parameter), when using one of the ML methods (MLavg, MLmin, or MLmax).

She noticed that corridors for these ST points were not created, and this may be related to what happens at the edges of the map, and to the moving-window process to create the input resistance map for these methods.

This is just a reminder for us to take a look at that.

[pdodonov]

Greetings! I had previously replied privately to Bernardo because I didn't know how github works (shame on me, shame! :-) ). So now I'm responding here as well, for us to have the conversation registered here etc. As I see it, what we have is a boundaries effect issue. When the window overlaps the study area's boundary, part of it are occupied by NA (or NULL?) values - missing data - and something must be done about it. Bernardo said that the GRASS function used in LSCorridors exludes the NULL values and calculates the summary statistics for the other cells, but perhaps this is not happening here. In any case, I think there are three ways of dealing with boundaries effects in our case: 1) Do not create the corridor and provide a warning or error message: "Corridor not created due to proximity to the map's boundaries. Try selecting a smaller window size." 2) Calculate average, min and max values considering only the pixels for which there is information. This will likely increase variation at the map's edges, as uncertainty is greater when there are fewer pixels. 3) Mirror the map inside the window: for example, if we have PASTURE PASTURE FOREST FOREST FOREST NULL NULL NULL..., this becomes PASTURE PASTURE FOREST FOREST FOREST FOREST FOREST FOREST PASTURE PASTURE. Not really realistic, but may be a plausible alternative. Another option is to wrap the map - when the eastern corner is reached, the calculation continues from the eastern corner. Not really realistic, but might make sense if land use patterns are stationary. Spatial pattern analysis also sometimes uses zero-padding: artificially replacing NULL value with zeroes, but this would make no sense whatsoever here. It'd be nice also if the output map could show the parts of it that are subjected to boundary effects, as results in these areas must always be interpreted with caution. It will probably be a small part of the map, but perhaps not. Best, Pave

…

On 4/12/18, Bernardo Niebuhr ***@***.***> wrote: The user Laura Nery Silva commented on an issue dealing with corridor simulations when the source-target (ST) points are near the edges of the resistance map (closer than the scale parameter), when using one of the ML methods (MLavg, MLmin, or MLmax). She noticed that corridors for these ST points were not created, and this may be related to what happens at the edges of the map, and to the moving-window process to create the input resistance map for these methods. This is just a reminder for us to take a look at that. -- You are receiving this because you are subscribed to this thread. Reply to this email directly or view it on GitHub: #6

-- ------- Pavel Dodonov Post-doctorate fellow – Universidade Estadual de Santa Cruz – UESC Laboratório de Ecologia Aplicada à Conservação - LEAC Rodovia Ilhéus-Itabuna, km16 /Salobrinho, Ilhéus , Bahia, Brasil, CEP 45662-000 Skype: pdodonov1 Papers: https://www dot researchgate dot net/profile/Pavel_Dodonov Lattes CV: http://lattes dot cnpq dot br/9008153877455949 Linux user #624671 (via linuxcounter.net) *"The highest function of ecology is understanding consequences." - Frank Herbert, Dune, 1965*