Since the landscape is divided among the 32 processing nodes, a deer's search area may encompass more than one processor. Without knowledge of data across processor boundaries, it becomes impossible for the animal to select a new grid location similar to the one it would have chosen in the sequential model.
Similar parallel ecological models developed in the past have determined whether an animal should move to a new processor by using a decision method referred to as PMI (Preferential Moving Index) averaging. With this method, a PMI value is computed at the beginning of each simulation day, using data from all or portions of a processor's grid cells, and then broadcast to every other processor. If the animal cannot locate enough forage on its own processor, the nearest neighbor processor with the greatest PMI average above a threshold is chosen and the animal is sent to that processor to continue its foraging sequence [BU95].
Although PMI averaging may result in decreased parallel execution time for some ecological applications, this method was not used in the parallel SIMPDEL model for several reasons:
Therefore it was necessary to develop a different parallel method for determining deer movement locations based on the most recent forage levels. The parallel method for the deer forage search relies on all processors containing grid locations within a deer's search area to examine the necessary grid cells in parallel.