Scale is an important issue both in landscape ecology and remote sensing. Although scale in landscape ecology involves both grain and extent, changing grain size is much more concerned by scientists. Choosing data of proper resolution has always been a big problem for landscape pattern analysis using satellite images. Many studies have been conducted trying to investigate the effects of remote sensor spatial resolution on landscape pattern analysis. However, the effects of remote sensor spatial resolution on urban landscape pattern analysis have rarely been reported.
Land use transects across urban center of Shanghai were clipped from the land use maps of 2000 and 2002, which were created by manual interpretation from ETM+ and IRS-PAN images, to produce Land use transect I and II (LUT-I and II). To study the impacts of road corridors on urban landscape pattern, road corridors and urban patches in LUT-II were merged to create Land use transect III (LUT-III). The three transects were then converted to ArcGrid formats at the grain size of 5, 10, 15, 20, 25, 30, 35, 40, 50, 60, 80, 100, 120, 140, 160, 180, 200, 250m using ArcView Spatial Analyst. Seven landscape indices were examined from both landscape and class-level including percent of coverage (PLAND), patch density (PD), edge density (ED), mean patch fractal dimension (MPFD), contagion (CONTAG), largest patch index (LPI) and Shannon diversity index (SHDI).
The results showed that scale effects of selected landscape indices were different in urban landscape. PD in LUT-II dropped as grain size increased from 5m to 10m, and increased as grain size reached 30m, and then decreased again with increasing grain size. Linear corridors such as roads were the major reasons leading to scale effects in urban landscape, because they are quite sensitive to varied remote sensor resolution and grain size.
PD, ED, MPFD, CONTAG and LPI were more sensitive to changing grain size comparing with PLAND and SHDI. Although PD, ED, MPFD and LPI could be used to quantify urban fragmentation, their behavior was different with changing grain size, highlighting the necessity of measuring urban fragmentation from different aspects.
The results also suggested that satellite images from IKONOS, SPOT or IRS-PAN with fine resolution were necessary for examining urban fragmentation, while data from TM/ETM+ with coarser resolution might be used to monitor urban sprawl. For data derived from IRS-PAN image, grain size about 5-10m is necessary to avoid scale effects in landscape pattern analysis, while the optimal grain size for data derived from TM/ETM+ images is 30-80m.