城市地表温度与NDVI空间相关性的尺度效应

城市地表温度与NDVI的空间相关性已被广泛研究,但是其尺度效应常常被忽略,给研究结果带来不确定性。本文以郑州市为例,基于四幅Landsat8影像,经辐射传导法反演地表温度(LST),运用半变异函数识别地表温度的空间相关性分析尺度,并结合空间相关指数Moran's I,从多尺度、多季节、多邻接范围3个方面对地表温度与植被覆盖的空间相关性的尺度效应进行了探讨,结果表明：①LST和NDVI的单变量空间自相关和双变量空间相关尺度均在300 m左右;②300 m相关尺度内,单变量空间自相关性存在显著尺度效应,相比之下双变量空间相关性尺度效应较为微弱;③LST和NDVI的单变量空间自相关性和双变量空间相关性尺度效应均表现出明显的季节差异;④随着邻接范围增大,LST和NDVI的空间自相关性减弱,尺度效应更明显。因此度量LST和NDVI的空间相关性要考虑时空尺度效应,本文研究结果有助于进一步认识LST和NDVI间空间相关性的尺度效应。

Scale effect of the spatial correlation between urban land surface temperature and NDVI

The spatial correlation between urban land surface temperature (LST) and vegetation coverage (NDVI) has been widely studied, but its scale effect is often ignored, which brings uncertainty to the results. Taking Zhengzhou City as an example and based on four Landsat8 images, this study retrieved the land surface temperature by the radiation conduction method, and identified the spatial correlation analysis scale of the land surface temperature by using the semivariance function. It then combined the spatial correlation index Moran's I to discuss the spatial correlation between land surface temperature and vegetation coverage from three aspects: multi-scales, multi-seasons, and multi-adjacent ranges. The results show that: (1) Both the univariate spatial autocorrelation scale and bivariate spatial correlation scale of LST and NDVI are around 300 m; (2) Within the 300 m correlation scale, there is a significant scale effect in the univariate spatial autocorrelation, but the scale effect of bivariate spatial correlation is much weaker by comparison; (3) The univariate spatial autocorrelation and bivariate spatial correlation scale effects of LST and NDVI show significant seasonal differences; (4) With the increase of adjacent range, the spatial autocorrelation of LST and NDVI weakens, and the scale effect is more obvious. Therefore, to measure the spatial correlation between LST and NDVI, spatiotemporal scale effect should be taken into consideration. This study should be helpful for further understanding the scale effect of spatial correlation between LST and NDVI.