I’ve been experimenting with spatial vs. non-spatial modeling in ST-SIM using the LANDFIRE library, and I have noticed that I get very different output (in terms of the mean distribution of state classes) depending on which I choose.
I first noticed this when running a larger spatial model containing a number of different vegetation types- the mean distributions I recorded (after 10 iterations of 1000 timesteps) did not match up with the reference distributions in the LANDFIRE database. I have since experimented by running models with only one vegetation type, while keeping the total area and pixel size constant and starting with equal proportions of each state class, and still I get different distributions between the spatial and non-spatial versions.
Can anyone shed any light on why this happens? I’m guessing it must be related to how disturbance is modeled spatially, but any further information would be greatly appreciated! Thank you.
Depending on how your spatial model is configured (size distributions, primary vs. secondary groups, etc…) it is possible that you will end up with different results between spatial and non spatial simulations. The most likely cause would be some kind of spatial constraint in the way some transitions are spreading, or some error in configuration (for example in the assignment of transitions to primary vs. secondary groups). Without seeing your specific configuration it would be hard to say for sure. A good approach to diagnosing potential problems is to compare the expected area transitioning for each transition type (sum product [area by state]x[probability by state] at timestep t-1) vs. actual area transitioning at timestep t. If you see large differences for some transition types this may help you find out which aspects of your model configuration may have problems or at least be leading to counter intuitive results.