Abstract: A fractal analysis was performed for grassland patterns on the Songnen plain in Northeast China, a region with extremely heterogeneous vegetation because of alkalization. Four belt regions were selected according to the different grazing intensity and the flood condition in 1994. Field measurements on patch shape and size of plant communities were carded out along each region in 1997. The perimeter-area exponent and Korcak exponent were evaluated by regression method to determine the fractal dimension of the patch boundary and patchiness of the grassland. 1. The patchiness (Korcak exponent) of Leymus chinensis (Trin.) Tzrel, the dominant species of zonal vegetation, increased with the grazing gradient. The perimeter-area exponent (complexity of boundaries) was the highest (0.64) in the lightly grazed grassland, and lower in the forage grassland (0.56) and the heavily grazed grassland (0.59). 2. Both perimeter-area exponent and Korcak exponent of L. chinensis and two subdominant species (Puccinellia spp. and Aeluropus littoralis (Gouan) Parl. var. sinensis Debeaux) of zonal vegetation were lower in the flooded than in the non-flooded grassland. The possible reason is that flood reduces the heterogeneity of soil alkali. 3. In forage grassland, both Korcak exponent and perimeter-area exponent of the dominant species were lower than that of the sub-dominant species, but showed converse results in the heavily grazed flooded grassland. 4. Four of the six Durbin-Watson tests of regression evaluating Korcak exponent displayed significant autocorrelation in residuals. It can be concluded that the dynamics of patchiness occurred on at least two scales in the present data range. No significant autocorrelation was observed in residuals of regression evaluating perimeter-area exponent. It revealed that patchiness and complexity of boundary, had different scale regions.

Key words: Alkaline grassland, Spatial pattern, Disturbance, Fractal analysis, Scaling roles