空间粒度变化对景观格局分析的影响

Effects of changing grain size on landscape pattern analysis

认识空间异质性的多尺度依赖性和景观格局特征对尺度效应关系的影响是进行空间尺度推绎的基础。以 2种真实景观 (中国广东粤北植被景观与美国凤凰城城市景观 )和 SIMMAP景观中性模型产生的 2 7种模拟景观为对象 ,利用景观格局分析软件 FRAGSTATS对 1 8种常用景观指数的尺度效应进行了系统的分析。根据这些指数对空间粒度变化的响应曲线和尺度效应关系 ,1 8种景观指数可分为 3类。第 1类指数随空间粒度的增大单调减小 ,具有比较明确的尺度效应关系 (幂函数下降 ) ,尺度效应关系受景观空间格局特征的影响较小 ;这类指数包括缀块数、缀块密度、边界总长、边界密度、景观形状指数、缀块面积变异系数、面积加权平均缀块形状指数、平均缀块分维数和面积加权平均缀块分维数。第 2类指数随空间粒度的增大将最终下降 ,但不是单调下降的 ;尺度效应关系比较多样 ,可表现为幂函数下降、直线下降或阶梯形下降 ,主要受缀块空间分布方式和缀块类优势度的交互影响 ;这类指数有 5种 :平均缀块形状指数、双对数回归分维数、缀块丰度、缀块丰度密度和 Shannon多样性指数。第 3类指数随空间粒度的变粗而增加 ,随缀块类优势度均等性的增加 ,尺度效应关系由阶梯形增加、对数函数增加、直线增加向幂函数增加过渡 ,尺度效应关系?

Spatial heterogeneity is ubiquitous across all scales of natural systems. Spatial pattern/heterogeneity is also scale dependent, i.e., spatial heterogeneity exhibits various patterns at different scales, therefore the observed pattern/heterogeneity is dependent on the scales of observation or analysis.Scale effects mean how ecological properties change with scales. Effects of changing scale on spatial analysis have been studied for decades in geography and ecology. The main goal of this study was to validate the scaling relations darived in our previous studies by analyzing additional real and simulated landscapes.To systematically investigate the effects of changing grain size on landscape pattern analysis, we chose two real landscapes (Northern Guangdong vegetation landscape representing relatively natural and undisturbed landscapes and Phoenix urban landscape representing highly managed and human- dominated landscapes) and 27 simulated landscapes generated by using the SIMMAP neutral landscape model. Three factors and three levels of each factor were considered while creating the simulated landscape maps. The first factor was patch richness (or number of classes), including three levels: 2, 5, 10. The second factor was class dominance (i.e., the proportion of the whole landscape area occupied by a particular class or patch type), including one-dominated, systematically decreasing, and equally dominated. The third factor was spatial distribution of patches, including clumped, moderately clumped and randomly distributed. These 29 landscapes represented a variety of landscapes with different spatial pattern characteristics. For changing grain size, we kept the extent the same as the original data sets (750 by 750 pixels for simulated landscapes and 1200 by 1200 pixels for the two real landscapes). Grain size was systematically changed from 1 by 1 to 100 by 100 pixels following the majority rule. We examined 18 landscape indices (see the next paragraph). The landscape pattern analysis package, FRAGSTATS 3.0, was used to compute the 18 selected landscape metrics. In total, these metrics were examined at 696 single scales for the 29 landscape data sets.The results in this study confirmed the scaling relutions found in our previous stueies.Based on the shape of the scale effect curves and scaling relations, the 18 landscape indices in this study were divided into three groups/types. Type I indices decreased monotonically with increasing grain size, showing a power-law decay scaling relation, with the characteristics of spatial pattern having little impact on scaling relations. This group included 9 landscape metrics: number of patches, patch density, total edge, edge density, landscape shape index, patch size coefficient of variation, area-weighted mean patch shape index, mean patch fractal dimension and area-weighted mean patch fractal dimension. Type II indices also decreased with increasing grain size, but not monotonically. There was no single scaling relation for each index, and scaling relations were related to spatial patterns, mainly influenced by the interactions of class dominance and spatial arrangement of patches. This group included 5 metrics: mean patch shape index, double-log fractal dimension, patch richness, patch richness density and Shannon's diversity index. Type III indices increased with increasing grain size. The shapes of the scale effect curves were various. There were three to five scaling relations for each index, and the scaling relations were mainly influence by class dominance. With increasing equality of class dominances, the scaling relations changed from staircase increase to logarithmic increase to linear increase to power law increase. There were 4 indices in this group: mean patch size, patch size standard deviation, largest patch index and contagion.Type I and II indices were very sensitive to grain change and decreased dramatically with increasing grain size below a critical value, whereas Type III indices increased dramatically with increasing grain size below a critical value. Spatia