PARENT SESSION
Oral Session #29: Spatial Ecology and Disturbance Ecology.
Presiding: J. Ludwig
Tuesday, August 6. 8:00 AM to 11:30 AM. Coconino Meeting Room, TCC.

Computational methods for ecological forecasting: Spatial models and algorithms.

Dietze, Michael^*,1, Govindarajan, Sathish¹, Agarwal, Pankaj¹, Clark, James¹, ¹ Duke University, Durham, NC

ABSTRACT- The ability to make sound ecological forecasts is an increasingly important priority in ecology. However, the models required to make such forecasts are often computationally intensive. This is particularly true for spatially explicit models, where interactions are often based upon the pair-wise distances between points on the landscape. Traditionally, spatial model runtime increases rapidly with total landscape size, generally on the order of the square of landscape size or slower. We present new computational methods that dramatically reduce the runtime of spatial models. These models are built around a novel data structure, the Quad Tree, which provides adaptive spatial resolution and an efficient hierarchical structure. Examples are drawn from the development of a new spatial forest model, ADOHI. Particular emphasis is placed on methods for calculating long distance dispersal and understory lighting, both of which we are able to reduce in runtime by over an order of magnitude compared to traditional computational methods. These algorithms also scale better than traditional methods, thus allowing larger landscapes to be simulated. This work emphasizes the need for efficient computational methods to allow for larger and more realistic models of global change.

KEY WORDS: algorithms, forest model, computation, spatial model