simulating landscape - how much realism?

[bowlerbear]

I played with a couple of functions to simulate a species occupancy landscape. From random fields (used in the AHMbook package), we can get something like this:

Here I changed extent by adding in an effect of latitude. But this effect is independent on the underlying spatial field. I should be able to include a trend in the spatial field using RMtrend but I cant figure it out

[bowlerbear]

alternatively in the gstat package there is another function, in which the trend is directly modelled. This gives me this:

[bowlerbear]

In both cases, there are 3 main parameters, mean occupancy (linked to area of occupancy) trend or latitude effect (can be linked to extent of occupancy) and clumpiness

But maybe this is more complex than we need?? Need to think more about it.

Anyhow these functions are in the simulate_landscape.R

[coreytcallaghan]

Wow - these are awesome looking.

I'm not entirely sure either. But I guess more realism is totally fine, and even believable.

One thing I came upon when I was writing the quick and dirty function the other day was where to add the random draws: before OR after the starting landscape. So, do we want to randomize the starting landscape AND the change in the scenarios (random, marginal, and core)? Or, do we just want to focus on the change from point 1 to point 2.

I feel like this is relevant and it could go either way. But if we simulate/randomize the whole thing (including the starting landscape) then complex or not probably doesn't matter too much... but if we just want to simulate the change through time then perhaps it would be necessary to start with different realistic landscapes?

I'll keep thinking, but I would lean towards simulating the whole thing and picking some cutoffs in one of these functions to ensure a starting point falls in high EOO high AOO or high EOO low AOO etc. etc.