Performance of the forest productivity model 3-PG applied to a wide range of forest types.

The structure of the 3-PG model is outlined indicating the input data and parameter values required to run it and the changes introduced since the model description was originally published. Calibration procedures are described and the capacity of 3-PG to fit a wide range of forest growth data sets, from experimental and commercial plantings, is evaluated. The data came from the environmental range from sub-tropical Africa and Australia (Kwa-Zulu Natal 28°S and Queensland) to northern Europe (Flakaliden, 64°N, in Sweden). In all cases it was possible to simulate the observed data with useful accuracy. Values of the key parameters, and the way they vary are provided. To test whether the model can predict stand growth when it has not been fitted to the data we used independent sets of measurements, providing tests of model performance at three levels of severity: a test against data from the same site, where different treatments were applied; a test on data from a different location, where a similar experiment had been carried out; and a test against data from a spacing and thinning experiment. The results of the tests are assessed in terms of simple statistics (r² and bias). In the first and second tests, the r² values were all >0.9 (with one exception, where <F>r²=0.85</F>), although the bias in some of the results for the first test was relatively high (>20% in above-ground and >50% in below-ground variables). The second test produced unbiased results; r² values in the third test were lower, but the results were also essentially unbiased. The results presented show that the 3-PG model is robust and reliable and can be used with confidence to predict growth in areas where trees have not been grown, i.e. the model can be used to estimate site productivity. The model can also be used to explore the effects of environmental conditions on tree growth and productivity. Parameters in 3-PG that allow the model to account for species differences, and the problem of determining the appropriate values for the fertility rating used in the model are discussed. [Copyright &y& Elsevier]