Why can't process-based models kill trees when modelling drought-induced mortality? How can we fix that?

Background/Question/Methods

Plant mortality events following droughts are occurring throughout the globe and rising mortality rates have been documented in major biomes. The American southwest and central plains regions have experienced extreme droughts both in the past few centuries that have resulted in dramatic changes in vegetation.  There is currently considerable debate on the direct causes of mortality, particularly whether carbon starvation, hydraulic failure, or an interaction drives mortality. Although interest in simulating and forecasting vegetation mortality has grown substantially, no model has been developed specifically to simulate mortality because no one knows what are the empirical thresholds beyond which mortality is unavoidable?  Our main goals were to investigate the advantages and disadvantages of different model assumptions for examining tree mortality during drought, and to determine the physiological mechanisms resulting in mortality, emphasizing hydraulic failure and carbon starvation as mechanisms. To parameterize the models, we used results from an ecosystem-scale precipitation reduction experiment at a piñon pine-juniper woodland, as well as field data from central Texas in which trees exhibited substantial mortality.

Results/Conclusions

Our modeling scenarios illustrated that some common levels of hydraulic failure and carbon starvation processes were predicted to co-occur in dying trees by all models and therefore interactions between water use and carbon acquisition and usage made assigning a single cause for mortality difficult. Despite the large variety of model complexity used, all simulations suggested that carbon starvation was more likely than hydraulic failure per se as the primary driver mortality. In the simpler model, equations accounting for hydraulic failure and carbon starvation were only represented by a single module within the soil-plant-atmosphere continuum.  Requiring the fewest parameters, this model over predicted tree mortality from carbon starvation, emphasizing the importance of using species-specific parameters to accurately predict tree mortality. The source-driven approach used in the more complex models simulated low NSC values, with the lowest predicted NSC values found when maximum death rate was recorded.  As an example, trees that died had, on average, only 40% of their maximum NSC reserves, whereas trees that survived averaged 60% over the same period.  Notably, those models allowed NSC to decline to zero rather than imposing a threshold.  Given that NSC never reaches zero in dead trees, it is likely that imposing a threshold NSC values in models is appropriate. A common challenge to all models was process understanding belowground, including rooting depth, root cavitation, embolism refilling and carbohydrate partitioning.