Climate change and anthropogenic land use are increasingly affecting the resilience of natural ecosystems. In Mediterranean ecoregions, forests and woodlands have shown progressive declines in health. This study focuses on the decline of an endemic woodland tree species,Eucalyptus wandoo(wandoo), occurring in the biodiversity hotspot of southwest Western Australia. We determined the change in health of wandoo stands between 2002 and 2008 across its geographic and climatic range, and associated this change in health with non‐biotic variables focusing on: (1) fragment metrics; (2) topography; (3) soil characteristics; and (4) climate. Only fragment metrics and climate variables were found to be significantly related to the observed change in health. Stands that were small with high perimeter/area ratios were found to be most sensitive to health declines. Recent increases in autumn temperatures and decreases in annual rainfall were negatively affecting health of wandoo most prominently in the low rainfall zone of its climatic range. Together, these results suggest the onset of range contraction for this ecologically important species, which is likely to be exacerbated by projected future changes in climate. Our results emphasize the importance of establishing monitoring programs to identify changes in health and decline trends early to inform management strategies, particularly in the sensitive Mediterranean ecoregions.

Changes in climate trends and extreme climatic events are increasingly impacting on forests around the world. In order to better understand how and where major ecological and climatic changes will affect our forested ecosystems, tools based on landscape sensitivity analysis need to be developed to help inform sustainable forest management. This study was undertaken in the Northern Jarrah Forest (NJF) region in the Mediterranean climate of southwest Western Australia. Extreme drought and multiple heatwaves in 2010/2011 resulted in large-scale tree canopy dieback in the NJF. In this study, we used Landsat satellite imagery to (1) accurately map the damaged areas, (2) produce a damage probability model and (3) compare the model with a probability model derived from data collected through an airborne flight survey. We found that the Landsat-derived Normalized Difference Vegetation Index was a good indicator of drought/heat induced damage in the NJF region. Both probability models identified the same set of topography and climate-related factors for determining the probability of drought/heat damage within the landscape. Extrapolation of the Landsat satellite method-based model, however, produced a more deterministic and useful drought/heat damage sensitivity map for the NJF region. The techniques and tools developed, and applied, in this study can readily be transferred to other regions around the world and can assist in the sustainable management and timely climate adaptation efforts to accommodate our future forests.

Spatial modelling approaches are increasingly being used to direct forest management and conservation planning at the landscape scale. A popular approach is the use of buffer-radius methods, which create buffers around distinct forest habitat patches to assess habitat connectivity within anthropogenic landscapes. However, the effectiveness and sensitivity of such methods have rarely been evaluated. In this study, Euclidean and least-cost buffer-radius approaches were used to predict functional ecological networks within the wooded landscape of the Isle of Wight (UK). To parameterize the models, a combination of empirical evidence and expert knowledge was used relating to the dispersal ability of a model species, the wood cricket (Nemobius sylvestris Bosc.). Three scenarios were developed to assess the influence of increasing the amount of spatial and species-specific input data on the model outcomes. This revealed that the level of habitat fragmentation for the model species is likely to be underestimated when few empirical data are available. Furthermore, the least-cost buffer approach outperformed simple Euclidean buffer in predicting presence and absence for the model species. Sensitivity analyses on model performance revealed high sensitivity of the models to variation in buffer distance (i.e. maximum dispersal distance) and permeability of common landscape features such as roads, watercourses, grassland and semi-natural habitat. This indicates that when data are lacking with which to parameterize buffer-radius models, the model outcomes need to be interpreted with caution. This study also showed that if sufficient empirical data are available, least-cost buffer approaches have the potential to be a valuable tool to assist forest managers in making informed decisions. However, least-cost approaches should always be used as an indicative rather than prescriptive management tool to support forest landscape conservation and planning.