In general, striping or banding in the change prediction output image from LCM tends to happen when it is allocating cells to a category that needs to have more cells (based on the predicted # of cells in that category from the Markov prediction), and it has to choose from cells that are equally unlikely/likely. For example, if the transition probability map doesn't has sections of that map that are equally probable (usually equally unlikely), then it starts to just fill them in from that category.
There is some variability between runs, as both MLP and DecisionForest use randomness in their starting points, so each run will be slightly different. For this reason, if the striping is just for a small area, you should try running the model again.
You might want to look at the area that is being allocated and think about which category/categories should get more cells allocated to them, based on what their suitability is.
If you find that the same category is being consistently over-allocated, you can manually edit the Markov matrix on the top panel of the Change Prediction tab. That tells Change Prediction how much of each category to try to allocate.
Other tips:
Make sure your landcover images are accurate and your driver variable images make sense. The quality of these images will affect your result no matter what method you use to predict the future.
You might also try running it with only 1 iteration. That sometimes leads to less artifacts.
0 Comments