Groundwater change forecasts widespread forest dieback across an extensive floodplain system

1. Regulation of rivers for human demands has led to extensive forest dieback on many floodplains. If these important ecosystems are to be maintained under future drier climates, we need accurate tools for predicting forest dieback. In the absence of spatially explicit flooding histories for many floodplains, changes in groundwater conditions may be a good indicator of water availability and, therefore, an important environmental indicator. 2. Eucalyptus camaldulensis forests of Australia are an acute example of forest dieback, with 70% of the Victorian Murray River floodplain dying back. We quantified the relationship between forest dieback and ground water across this extensive floodplain (c. 100 000 ha of forest over 1500 km of river length). 3. A combination of extensive ground surveys, remotely sensed data and modelling methods was used to predict forest dieback at the time of the survey and in the past. This approach provides a valuable tool for accurately monitoring forest condition over large spatial scales. Forest dieback was estimated to have increased from 45 to 70% of the floodplain between 1990 and 2006. 4. Accurate groundwater data (depth and salinity) over a 20‐year period were obtained for 289 bores and summarised using nonlinear regression. Groundwater depth and salinity were strong predictors of stand condition. This suggests that changes in groundwater conditions could be used to signal areas vulnerable to forest dieback and prioritise the limited water available for managed flooding. 5. In the upper Murray, where ground water is predominantly fresh (<15 mS cm^?2), dieback increased with increasing groundwater depth. In contrast, the condition of stands in the lower Murray improved with increases in groundwater depth due its high salinity (>30 mS cm^?2). These regional differences in response of the same tree species to groundwater conditions show that our understanding of the drivers of forest dieback is best achieved at spatial and temporal scales representative of the problem.