Characterizing the spatial distribution of giant pandas (Ailuropoda melanoleuca) in fragmented forest landscapes

To examine the effects of forest fragmentation on the distribution of the entire wild giant panda (Ailuropoda melanoleuca) population, and to propose a modelling approach for monitoring the spatial distribution and habitat of pandas at the landscape scale using Moderate Resolution Imaging Spectro-radiometer (MODIS) enhanced vegetation index (EVI) time-series data. Five mountain ranges in south-western China (Qinling, Minshan, Qionglai, Xiangling and Liangshan). Giant panda pseudo-absence data were generated from data on panda occurrences obtained from the third national giant panda survey. To quantify the fragmentation of forests, 26 fragmentation metrics were derived from 16-day composite MODIS 250-m EVI multi-temporal data and eight of these metrics were selected following factor analysis. The differences between panda presence and panda absence were examined by applying significance testing. A forward stepwise logistic regression was then applied to explore the relationship between panda distribution and forest fragmentation. Forest patch size, edge density and patch aggregation were found to have significant roles in determining the distribution of pandas. Patches of dense forest occupied by giant pandas were significantly larger, closer together and more contiguous than patches where giant pandas were not recorded. Forest fragmentation is least in the Qinling Mountains, while the Xiangling and Liangshan regions have most fragmentation. Using the selected landscape metrics, the logistic regression model predicted the distribution of giant pandas with an overall accuracy of 72.5% (κ = 0.45). However, when a knowledge-based control for elevation and slope was applied to the regression, the overall accuracy of the model improved to 77.6% (κ = 0.55). Giant pandas appear sensitive to patch size and isolation effects associated with fragmentation of dense forest, implying that the design of effective conservation areas for wild giant pandas must include large and dense forest patches that are adjacent to other similar patches. The approach developed here is applicable for analysing the spatial distribution of the giant panda from multi-temporal MODIS 250-m EVI data and landscape metrics at the landscape scale.