空间幅度变化对景观格局分析的影响

Effects of changing spatial extent on landscape pattern analysis

景观格局指数是量化描述景观格局特征的主要方法之一 ,各种格局指数的尺度依赖性使比较分析景观格局特征和尺度推绎复杂化 ,分析不同指数随空间幅度变化的一般行为有助于景观格局分析结果的解释和降低空间尺度推绎的复杂性。研究以 2种真实景观和 2 7种模拟景观为分析对象 ,考查了 1 6种常用的景观水平格局指数随空间幅度变化行为。根据这些指数因幅度变化行为可预测性把它们分为两类 :第 1类随幅度变化可预测性强 ,指数与幅度之间的关系可用简单的函数关系来表达 ;这类指数包括缀块数、边界总长、景观形状指数和缀块丰度密度 ;前两者随幅度增加呈幂函数形式增加 ,而缀块丰度密度随幅度增加呈幂函数下降 ,景观形状指数随幅度增加呈直线增加。第 2类指数随幅度变化的可预测性较差 ,指数随幅度的变化存在多种可能 (不同形式的增加、减小或保持不变 ) ,不可用一种或多种简单的函数关系来描述所有的情况。这类指数包括缀块丰度、缀块密度、边界密度、最大缀块指数、平均缀块面积、缀块面积标准差、缀块面积变异系数、平均缀块形状指数、面积加权平均缀块形状指数、双对数回归分维数、聚集度指数与Shannon多样性指数。第 2类指数随幅度的变化行为受景观格局特征和指数本身算法的影响。总体上来说 ,第

Scale and scaling issues represent one of the foremost frontiers of landscape ecology and ecology in general. There is an urgent demand for ecological information at finer/smaller scales to be transferable to coarser/larger scales. Understanding spatial heterogeneity is one of the essential challenges of spatial scaling. We refer scale to spatial extent in this study. Several ecologists had demonstrated that changing extent could influence the results of spatial pattern analysis. But only a few landscape indices with a relatively narrow range of extents were analyzed. In our previous studies,we analyzed four real landscapes of USA to illustrate the effects of changing spatial extent on landscape pattern analysis. The main goal of this study was to validate the general scaling relations derived in Wu et al .(2002) and Wu(2003) by anoryzing an additional complex Chinese landscape and a number of sumulated landscapes. Two real landscapes, the Northern Guangdong vegetation landscape (GDV) and the Phoenix urban landscape (PHX), USA, and 27 simulated landscapes generated by the SIMMAP neutral landscape model were chosen for analysis. The two real landscapes were both derived from LANDSAT TM remote sensing images with a grain size of 30 by 30 meters and an extent of 3651 4 km 2 for PHX and 8944 3 km 2 for GDV. Three factors were considered in generating the simulated landscapes maps. They were number of classes (2, 5, 10), class dominances (equal, one dominated, and systematically decreasing) and spatial distribution characteristics (clumped, moderately clumped, and random). All simulated landscapes had an extent of 750 by 750 pixels. Each of the 29 landscapes was clipped into smaller ones by decreasing the extent by 100 × 100 pixels each time. Sixteen commonly used landscape indices were computed for each of the 233 landscapes using a landscape pattern analysis software package, FRAGSTATS. A suite of scalograms and scale effect curves were drawn for analyzing the scale effects and deriving general scaling relations. Our results confirmed the general conclusions in our previous studies. Based on the shape of the scale effect curves and the predictability of the scaling relations, the sixteen landscape metrics could be divided into two groups. The first group included 5 indices: number of patches ( NP ), total edge ( TE ), landscape shape index ( LSI ), patch richness ( PR ) and patch richness density ( PRD ). The behavior of this group of indices with change in extent was very predictable for all 29 landscape types, and there were simple scaling relations for the 5 indices. NP and TE increased in a power law function ( y=ax b, a≥0, b>0 ) with increasing extent. NPD decreased in a power law function ( y=ax -b , a≥0, b>0 ) with increasing spatial extent. There was an increasing linear relationship ( y=ax+b, a≥0, b>0 ) between LSI and extent. PR increased with increasing extent in staircase fashion for two real landscapes and remained unchanged for 27 simulated landscapes. These general scaling relations could be valuable in choosing the extent while comparing the spatial patterns of different landscapes and extrapolating ecological information among different scales. The second group included 11 indices: patch density ( PD ), largest patch index ( LPI) , edge density ( ED ), mean patch size ( MPS ), patch size standard deviation ( PSSD ), patch size coefficient of variation ( PSCV ), mean patch shape index ( MSI ), area weighted mean patch shape index ( AWMSI ), double log fractal dimension ( DLFD ), contagion ( CONT ) and Shannon's diversity index ( SHDI ). The scaling behavior of this group of indices was generally unpredictable and related to the specific spatial pattern of a landscape. In general, the predictability of scaling relations of these indices increased with increasing number of classes, equality of class dominance and randomness of spatial distribution. We divided the scale effect curves of the second group of indices into four types. Type I and