Simulating complex landscapes with a generic model: Sensitivity to qualitative and quantitative classifications

Abstract The sensitivity of a generic cellular automaton model to qualitative and quantitative landscape classifications is tested regarding the spatio-temporal self-organization of the historical landscape pattern of southern Wisconsin (USA). The qualitative classification is R.W. Finley's “Original Vegetation of Wisconsin” and relies on criteria determined by an expert individual. The classification accounts for descriptive information and expert knowledge, but is not easily reproducible due to locally subjective decisions of class delineations. The quantitative classifications rely on a numerical-objective algorithm that ensures classification reproducibility and tests for robustness, but do not account for local ecosystem knowledge or qualitative detail. For model development, a cell in the forest–landscape lattice is chosen according to three generic and stochastic rules. The “uncorrelated” rule chooses a cell randomly, the “correlated” rule picks a cell within two distances of random length, and the “raster” rule chooses randomly one of four immediate neighbors in the lattice. The so chosen cell is then replaced by a cell randomly identified within a circular neighborhood of radius r (1  r Comparisons between model simulations and the empirical forest landscapes include temporal dynamics (cluster probability) and spatial patterns (fractal dimension, landscape diversity). Results suggest that the simulated landscape (1) exhibits self-organization for intermediate neighborhoods ( r  = 3), independent of the classification approach, three model-development rules, and two initial and boundary conditions; (2) achieves temporal (cluster probability) dynamics and spatial patterns (fractal dimension, landscape diversity) consistent with the empirical landscape. It is, therefore, concluded that a generic model calibrated independently from specific ecological processes may suffice to replicate major statistical characteristics of a complex landscape with simulations that are robust to various landscape-classification approaches.