The factors file describes the input dataset. The input data is always classified a priori, it is never continuous data. The characteristics used to describe the forest, as well as the number of classes used to define those characteristics, are written in this file. In the toy dataset, this file is called "SEfactors.txt", and describes the forest using the following characteristics:
- site classes
This is an example of the content you may find in a "factors.txt" file :
vol 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 age 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 siteclass SE11 SE12 SE21 SE22 SE23 SE31 SE32 SE33 species BT OB PA PS
Volume and age classes should be used to form the dynamic state space while the choice of the other factors is up to the modeller. Generally speaking, it makes sense to separate between different “forestry types” if they can be expected to either grow differently or to be managed differently. The number of “forestry types” could, from a technical view point, be large but it is important to keep in mind that the larger the matrix, the poorer the estimate of the parameters of the model.
The “age” factor should preferably be set up so that the width of the age classes coincides with the time between the two observations of the NFI plots used to estimate the "no-management" Transition Probabilities of the model. The number of age classes should be big enough to let the state-space span all existing forests today, as well as those that may be created during simulations.
The “volume” factor merits additional consideration. According to experience, the number of volume classes is appropriate within the order of 10 to 15 classes. We must have an adequately large volume span in the model to avoid that forest growth continues and hits the “roof” of the volume dimension. It is also advantageous if the probability of growing “out” of a state is of the same magnitude for young and older forests, which means that we should define our volume classes so that they reflect the growth pattern of a forest. When the volume classes are defined according to the growth pattern, the plots will be more concentrated in and close to the “diagonal” cells of the state-space. In principle, this means that the number of volume classes should be equal to the number of time periods (age classes) that it takes for mean age increment to reach its maximum. Of course, this case is theoretical because it would only occur if the growth function used for defining class limits was 100 % true. In reality, there will anyway be plots of the same age class but belonging to different volume classes. Even more so since we are using same number of volume classes for all site qualities and species. A “broad” volume class (i.e. wider that the actual one-period growth) implies that the plots will be distributed over a several age-volume classes instead of one age-volume class. “Stretching” volume classes over several age classes is not a good idea for several reasons. It can lead to some problems in the following model specification and reduce the information that the model would be otherwise able to carry in the analyses. The disadvantages of “broad” volume classes can be summarized as follows:
- The model might be less sensitive to modifications of final felling age. Changing the age class of final felling might have no effect on the harvested volume because it will still be the mean of the same volume class.
- The above applies also to thinning timings
- It might complicate the definition of the thinning. This means even one-class drop might be too large for a realistic thinning representation
Narrowing volume classes below a single growth time-step does not seem to have disadvantages other than a reduction of the number of plots per class available for the estimation of the transition probabilities which is, of course, a very serious limitation.