Improving the behaviour of forest gap models along drought gradients

Recent applications of forest gap models have shown weak performancealong drought gradients. Using the gap model FORCLIM, we examine twopossible explanations for this: (1) the parameters denoting the drought tolerance of the tree species need to be estimated more accurately, and (2) the soil water balance estimation schemes used in most gapmodels are not capable of tracking soil moisture content under dry conditions sufficiently well to reflect its influence on tree growth. The behaviour of the model along two drought gradients in Europe and eastern North America is used as a test case. Parameter adjustments alone turned out to be ineffective for improving the performance of FORCLIM. A theoretical analysis of the soil water balance model by Thornthwaite and Mather [Thornthwaite, C.W.. Mather. J.R., 1957, Instructions and tables for computing potential evapotranspiration and the water balance. Publ. Climatol. 10, pp. 183-311], which is used in FORCLIM and in many other gap models, showed that it yields inconsistent results when applied to simulate the year-to-year variability of drought occurrence. We therefore have developed a new water balance model with the goal of adequate performance in connection with a forest gapmodel. similar simplicity and low computational demand as in previous models. The new scheme uses a bucket approach and monthly time steps of temperature and precipitation alone. The performance of FORCLIM did not improve when the new water balance model was used together with the original estimates of the species' drought tolerance. However,when both a small number of drought tolerance parameters were adjusted and the new water balance model was incorporated, the performance of FORCLIM increased strongly along the drought gradients on both continents. In Europe, the transition from beech to oak forests could besimulated correctly, whereas in eastern North America the simulated pattern of aboveground biomass became much more realistic. We conclude that [ABSTRACT FROM AUTHOR]