Deliver geospatial models direct to decision makers with SyncroSim › Forums › ST-Sim & State-and-Transition Simulation Models › Removing Stochasticity › Reply To: Removing Stochasticity

Hi Ayn,

There are several sources of stochasticity in the SyncroSim ST-Sim package including:

1. **Transition Pathways**: both the occurrence of a transition and, if there are multiple outcomes possible, which outcome is selected are uncertain. For a non-spatial model you could reduce stochasticity by setting probabilities to 1 and using area targets instead. However, which simulation cells are selected to satisfy the area targets could still be uncertain. For a spatial model you could instead provide transition spatial multipliers which define exactly which cells will transition.

2. **Initial Conditions**: For a non spatial model the number of cells initialized for a specific state class and age range is stochastic. You can remove stochasticity by using the *Calculate from distribution checkbox* whereby in the *Relative amount* column you specify the exact number of cells to assign to each record. Note that if the age ranges span multiple values (i.e., Max Age > Min Age) there will still be stochasticity in the age assigned to each cell. To remove uncertainty in this you need to have Max Age = Min Age for each record. For a spatial model be sure to include an initial age raster to eliminate uncertainty. Note that you can also specify time since transition for initial conditions both spatially and non spatially. This is important if your transitions include time since transition constraints.

3. **Transition locations**: in a spatial model, the cells selected to transition may vary spatially. As noted above you can specify transition locations spatially using transition spatial multipliers that can be set to specifically transition selected cells for each timestep of your model run.

Daniel et al. 2017 shows an example where ST-Sim was first used to replicate projections from a deterministic forest estate model by removing as much stochasticity as possible, then adding stochasticity back after to “stress test” the deterministic projections in the face of uncertainty around factors such as wildfire.