关键词: 空间异质性, 特征尺度, 变异函数, 小波方差, 局部方差

Abstract: The characteristic scales in real landscapes reflect the spatial patterns and scales of human interactions with the environment. Identifying spatial structure and its characteristic scale is very important as well as necessary for exploring the spatial variability of different landscapes within remotes sensing images. Landscapes in mountainous area of South China is characterized as with strong variability and controlled by topographic conditions to a certain degree. Further researches are needed to quantitatively identify the characteristic scale of typical landscapes in those areas and its relationship with natural and anthropogenic processes. The prime objective of this study was to explore the characteristic scale of main landscapes at mountainous area in South China with semivariogram, wavelet transform and local variance using SPOT 10 m image. The first principal component of SPOT image is used for analysis. Results of variograms, wavelet variance, local variance for forest, city, agricultural and water landscapes all demonstrate that forest landscape possesses great spatial variability as quantified by the variogram or wavelet, local variance, and it partially originated from topographic complexity. City landscape also exhibits strong spatial variability and it's principally influenced by anthropogenic processes. Agricultural landscape is more heterogeneous than forest and city landscapes, but water landscape is most heterogeneous and no further variogram modeling conducted. Two different spatial structures were detected from wavelet analysis and semivariogram modeling in forest, city and agricultural landscapes individually. Results from variogram modeling are more precise and show that the range of the first spatial structure of city, agricultural and forest landscape is 16 m, 79 m and 95 m, and the second one is 133 m, 1031 m and 483 m respectively. Only one smaller characteristic scale is spotted from local variance. The mean characteristic scale quantified by mean length scale varies from 111 m to 569 m over city, forest and agricultural landscapes. The characteristic scale of forest in mountainous areas is relatively smaller than usual with the introduction of complicated landform. The largest spatial structure is detected in agricultural landscape which is the mosaic of agricultural crop fields surrounded by rivers and roads. In conclusion, a combined process for identifying characteristic scale with wavelet and semivariogram analysis is proposed. Firstly, wavelet method could be applied to distinguish the different spatial structures and the rough value of characteristic scale; then semivariogram modeling might be utilized to gain the exact value, with the number of combined models and parameters derived from results of wavelet analysis.

Key words: spatial variability, characteristic scale, semivariogram, wavelet transform, local variance