ABSTRACT
Global warming would favour a rise in the rate of forest disturbance owing to an increase in meteorological conditions likely to cause forest fires (drought, wind, and natural ignition sources), convective winds and thunderstorms, coastal flooding, and hurricanes. Simulations suggest that changes in forest composition associated with global warming would depend upon the disturbance regime (Overpeck et al. 1990). Two sets of simulations were run. In the first set (‘stepfunction’ experiments), simulated forest was grown from bare ground under present-day climate for 800 years. This enabled the natural variability of the simulated forest to be characterized. At year 800, a single climatic variable was changed in a single step to a new mean value, which perturbed the forest. The simulation was then continued for a further 400 years. In each perturbation experiment, the probability of a catastrophic disturbance was changed from 0.00 to 0.01 at year 800. This is a realistic frequency of about one plot-destroying fire every 115 years when a 20-year regeneration period (during which no further catastrophe takes place) of the trees in a plot is assumed. In each of the stepfunction simulations, three types of climatic change (perturbation) were modelled: a 1°C increase in temperature; a 2°C increase in temperature; and a 15 per cent decrease in precipitation. In the second set of simulation runs (‘transient’ experiments), forest growth was, as in the step-function experiments, started from bare ground and allowed to run for 800 years under present climatic conditions. Then, from the years 800 to 900, the mean climate, both temperature and precipitation, was changed linearly year by year to simulate a twofold increase in the level of atmospheric carbon dioxide, until the year 1600 when mean climate was again held constant. As in the step-function experiments, the probability of forest disturbance was changed from 0.00 to 0.01 at year 800. In all simulation runs, a relatively drought-resistant soil was assumed, and the results were averaged from 40 random plots into a single time series for each model run.
