A new system is presented for predicting tree mortality, in order to improv
e Swedish long-term forest planning. A three-step approach was used, which
consists of (I) estimating the probability of mortality on a sample plot; (
II) quantifying the mortality in terms of proportion of basal area., and (I
II) distributing the mortality among individual trees. The system predicts
the mortality for 5 yr periods. Data from permanent sample plots of the Swe
dish National Forest Inventory were used. Independent variables used for st
eps I and II were specific to site, stand and plot characteristics. In the
step III models, which were tree-species specific. competition indices were
also included. Logistic regression was used for steps I and III models, wh
ile linear regression was used for the step II models. A fair performance o
f the functions was observed, although mortality is a highly stochastic pro
cess. In applications, random simulation in all steps can mimic this.