To test the validity of the LUCAS model, ten pixel-based replicates of
the historical transition simulations for the Little Tennessee River
Basin were analyzed. Each replicate began with an initial land cover
map for the year 1975. Two 5-year time steps using the historical
transition probabilities from 1975--1986 were simulated followed by
one time step using the 1986--1991 historical transition
probabilities. The simulated 1980, 1985 and 1990 maps were then
compared against the historical 1980, 1986 and 1991 maps.
Turner [35] reported that the proportion of land cover of each cover type simulated by LUCAS corresponded closely to the actual historical land cover types. The simulation for forest, which was the dominant land cover type, had little room to redistribute because it comprises 90% of the landscape and hence was spatially distributed accurately. For the other cover types, however, LUCAS simulated many more patches than were present in the historical maps, i.e., the landscape became more fragmented. This is not surprising considering that the pixel-based model used in the simulation examines only independent grid cells. For example, after one time step 3417 patches of grassy land cover with an average patch size of 1.09 cells were simulated, whereas the actual map contained 1583 grassy patches with an average patch size of 2.4 cells. If a patch-based simulation were used, it would likely cause somewhat less fragmentation. However, due to the strict definition of a patch in the current implementation, i.e., contiguous grid cells with identical landscape condition labels, many patches tend to degenerate into one-cell patches or pixels. This is because the chance of at least one of the components of the LCL differing from its neighbor is great, thus creating small patches. A complete set of statistics from the model validation is available in [13].