Refugia are important areas for biodiversity conservation during climate change, but identification and application to management has been slow. Mapping hydrologic refugia requires quantification of the relationship between reduced water availability and biotic persistence. This is challenging due to limitations in modelling data from different scales and data-availability. The northern jarrah forests (NJF) of south-western Australia are one of the worlds remaining Mediterranean woodlands, and have undergone strong climatic and hydrological shifts since the 1970’s. Rainfall has decreased 15-20% and temperate increased 0.4%, resulting in decreases in surface run-off (40-90%), a falling water-table and shifts from perennial to ephemeral streams. Identifying refugia will help prioritise conservation action under change. Using comprehensive, fine-scale vegetation mapping data (31,000 plots, 432 km2, 120 m2 grid) and widely-available topographic variables, we automated community types to model their distribution under projected drying. Binary logistic regression models were fitted in R and accuracy assessed using k-folds cross-validation. To circumvent mismatches in the resolution and quality of available climate and topographic data, we use a non-parametric streamflow elasticity estimator modulated by rainfall to detect areas of refugia to 2090. We explore differences in the environmental profile and contraction dynamics of key indicator community types.
Results/Conclusions
Community types were automated with medium to high accuracy (mean accuracy = 0.74, SE = 0.02; mean kappa = 0.46, SE = 0.04). The weakest detectability (63-70%) was found in well-recorded, upland types, and the highest accuracy (80-88%) in rarer, range-restricted types. Changes in extent varied from 90% contraction in a riparian type (92-350 m elevation) underlain by deep clays and an impenetrable horizon, to >100% expansion in a granite-related type found in predominantly in exposed locations across ridge-tops and plateaus (92-567 m elevation). Patterns of expansion and contraction were overlain to indicate areas of potential refugia and mapped to a fine resolution for the study area (30 m grid). Perennial stream-zones provide the basis of refugia in the forest, with vulnerable floristic communities contracting tightly into remaining waterways. Our work quantifies the importance of lowland areas as important habitat for at-risk plant communities in the NJF. With streamflow-elasticity estimations available for over 500 catchments worldwide, fine-scale refugia modelling and mapping can be carried out using this method in landscapes where climate data may only be available at coarse resolutions. Landscape scale refugia maps provide local area managers with a basis for monitoring and conservation decision-making under climate change.