TreeAgitation
Order of Operation and Downstream Considerations
SIMULATION.MODEL.TreeAgitation
…OR…
SIMULATION
MODEL
TreeAgitation
Properties
|
Property |
Default Arguments |
Description |
|
CycleStart |
0 |
Cycle to begin agitating trees. Tree agitation occurs at the beginning of each cycle before competition metrics are calculated. |
|
CycleInterval |
5 |
Number of cycles until tree agitation is repeated. If CycleStart = 1 and CycleInterval = 5, agitation will occur in cycle 1, 6, 11, 16… etc. |
|
MaxTrees |
25 |
Maximum number of trees permitted in a tree record. If this threshold is exceeded, then the tree record is partitioned into X separate records, where X = round up to nearest integer (record trees per hectare / MaxTrees). This maximum should correspond to the median tree factor (trees/ha) of individuals in calibration plots. DBH in each partition is adjusted as a function of Delta. |
|
DeltaDBH |
0.001 |
Maximum DBH change (+/-). If number of partitions is an odd number, tree DBH in the first partition remains unchanged. Half of the remaining evenly divisible partitions have DBH altered by delta / partition subset number, and the other half by negative delta / partition subset number. For five partitions with DeltaDBH = 0.01, the following DBH values would result for a 30 tree record: 1) 30.000 cm 2) 30.010 cm [30 + 0.01 / 1] 3) 30.005 cm [30 + 0.01 / 2] 4) 29.990 cm [30 - 0.01 / 1] 5) 29.995 cm [30 - 0.01 / 2] For two partitions: 1) 30.010 cm [30 + 0.01 / 1] 2) 29.990 cm [30 - 0.01 / 1] If the adjusted DBH is less than the minimum Variant supported DBH or if adjusted is DBH <= 0, then tree agitation is completely canceled for the subject tree record. If Simulation.Model.IsRandom = TRUE, then evenly divisible partitions are randomly assigned a DBH adjustment within +/- delta bounds. If DeltaDBH is set to zero, DBH will remain unchanged; however, this is not advised because when two or more tree records have different species with no DBH variation, one species may always be sorted first (given undue preference and social position) when calculating tree social position competition indices. This can lead to unrealistic species competition dynamics during simulation. By setting DeltaDBH > 0, species with the same initial DBH class will be able to more fairly compete with one another. |
Example Commands
The two examples below demonstrate commands required to turn off tree record agitation (active by default; see default properties above). If MaxTrees is set equal to or higher than 9999, agitation is turned off, regardless of other settings.
SIMULATION.MODEL.TreeAgitation.MaxTrees 9999
Alternatively, to turn of tree agitation, you could also set CycleStart higher than your simulation total cycles.
SIMULATION.MODEL.TreeAgitation.CycleStart 9999
The following commands will cause tree record agitation to occur before every cycle on records with > 50 trees/ha. DBH in each partitioned record will be varied by a maximum of ± 0.5 cm.
SIMULATION
MODEL
TreeAgitation
CycleStart 0
CycleInterval 1
MaxTrees 50
DeltaDBH 0.5
Purpose and Outcomes
Tree Agitation provides properties that control when tree records should be partitioned by OSM during simulation and how much to agitate (subtly change) each partition's DBH by. OSM is not technically an individual tree model, as each tree record simulated by the system represents a number of trees/ha sharing the same tree properties; e.g., DBH, species, height, and competitive social position. The number of trees/ha in a record can be less than 1, equal 1, or be greater than 1. In an extreme case, one tree record could represent all trees in the stand. There is no competition separation of individuals within a tree record. Therefore tree record partitioning is often required to maintain acceptable competition distribution among tree records during simulation. Poor competition separation among individuals often arises for cruise data using 1-2 cm or greater DBH classes because number of trees represented by some records can be very high. These ‘dense’ tree records can also arise during simulation from ingrowth predictions. In other cases, sapling subplot tallies, in otherwise large fixed area individual-tree plots, may also lead to high numbers of trees per record(s). When only a few tree records contain the majority of trees in a plot or stand simulation, competition between trees in the model decline. If these ‘dense’ tree records are not broken up during simulation they can lead to underestimated competition throughout the simulation, which in turn, can lead to overestimated growth and survival.
During calibration of new tree-level equations, the importance of record partitioning is not obvious, as all records used during calibration are individual trees; however, in practice, foresters have recognized that timber cruise tallies can often contain a very large number of trees/ha in very few tree records, and these dense records may manifest into tree-list model projection bias if ignored; depending on underlying Variant calibration and associated input assumptions.
Tree-level models are often calibrated with 400 m2 individual-tree inventory plots, so to match tree model inputs to underlying calibration context, we would expect no more than 25 trees/ha to be represented by any one record. Therefore, number of partitions is controlled by the maximum number of individuals you wish to represent per tree record in the model (MaxTrees property).
If you wish to make OSM work as an individual tree model, you could set MaxTrees = 1 to force maximum separation of individuals (<= 1 per record); however, this will increase memory usage, slow simulations, and may actually cause under prediction of growth and survival if DBH resolution of individuals is more than what was used to calibrate the models. Keep in mind also that not all trees in a hectare will compete with one another due to proximity, so modeling individual trees has limited applicability here unless some spatial-dependence was to be introduced.
It is probably unnecessary to agitate records in every cycle unless the ingrowth model is introducing very ‘dense’ tree records, and even then, you may see no major change in results by waiting a few cycles to partition these ingrowth records. As agitation increases (by increasing either MaxTrees or CycleInterval), influence on model results will increasingly decrease. It may be wise to identify threshold MaxTrees and CycleInterval values that provide good performance and accurate results; i.e., the minimum amount of partitioning possible without substantially altering long-term simulation outcomes.
This concept is not new. Later versions of STAMAN circa 2012 (New Brunswick’s empirical tree list model, predecessor to OSM) introduced a ‘ReDistribution’ command to essentially perform the same task as described here.
Extreme Case Examples
In an extreme example where only one tree record represents all trees in the plot or stand (e.g., 10 thousand maple trees with DBH = 4 cm class), all individuals will experience the same level of competition. In this extreme case, if an individual’s competition measure is based on social position (e.g., basal area of larger trees [BAL]), the single tree record and all trees represented by it will have zero competition from above, even though we would expect competition between individuals in this record.
For a 400 m2 plot, ten thousand maple trees/ha tallied at 4 cm DBH (1 cm class size) would equate to 400 (10,000/25) individual trees measured between 3.6 and 4.5 cm DBH. This tree tally would results in a single tree record, which may cause the maple to grow substantially faster and survive longer because trees cannot compete with one another, compared to the same plot tally when individual trees are measured to the nearest millimetre (400 trees).
In another example, assume 1000 spruce and 1000 pine trees are all tallied in a DBH class of 30 cm (2 cm class size); if the pine is taller than the spruce, pine will always be sorted first in terms of social dominance in OSM (sorted by DBH then by HT), meaning pine will have an undue competition advantage over spruce. In reality, some spruce are probably slightly larger diameter and able to out compete some of the pine. Partitioning these tree records into 40 spruce and 40 pine tree records with slight random adjustments to DBH (+/- 1 cm) will promote more realistic competition dynamics within and between species during simulation.
In the extreme examples above, growth and survival would probably be substantially overestimated without agitation, depending on the Variant.
The figures below show the effect of simulating a single tree record (4000 trees/ha of balsam fir at 8 cm DBH) in the Acadian Variant (version 1.19.6.1) with different A) CycleInterval and MaxTrees settings and B) CycleStart settings, including a no tree agitation scenario. Effect of tree agitation will vary with model calibration, so these results below may be outdated. In this case, the Acadian Variant is shown to be extremely sensitive to extremely dense tree records. CycleInterval has little effect in this case because once trees are partitioned, they stay partitioned for the remaining of the cycle, and therefore repeated agitation has no effect. It appears that MaxTrees should definitely be set lower than 1000, but whether MaxTrees is set to 25 or 50 makes little difference on results. Tree agitation should always be started in cycle zero as the longer the delay (setting CycleStart higher) will cause increasingly over estimated predicted volumes (Figure B below).
While these examples are atypical of the likely distribution of tree records in most stand surveys, the importance of tree agitation should be clear.
One approach that may help to reduce the effect of dense tree records on model projections is to explicitly consider this issue during calibration. For example, basal area of larger trees, a common predictor variable that is susceptible to tree record density, could be altered to also include half of the basal area of the subject tree record. This theoretically ought to reduce over-estimation of growth because half of the trees’ basal area within the tree record will contribute toward increased competition from neighbors within the record. This, to our knowledge has not been tested, but should be!
Order of Operations and Downstream Considerations
Tree agitation is the first operation to occur at the beginning of each cycle; e.g., before competition metrics are calculated, before height predictions are made on trees with missing heights, and before the tree list is reported.
When tree agitation occurs, only the number of trees/ha (Stems), DBH, and Weight properties are modified in each partition. Both Stems and Weight properties are divided by the number of partitions and DBH is modified according to rules stated under DeltaDBH in the properties table.
If agitation occurs in Cycle 0, then agitation will likely influence missing tree height predictions, if height predictions depend on DBH or tree social position.
A user trying to compare trees in an input tree list to the output tree list for Cycle 0 will notice, if tree agitation has occurred, that there may be multiple records with the same tree ID. Partitions retain all properties of the original record including tree ID, except those properties stated above (Stems, DBH, and Weight).
If the OSM.Simulation.Model.Calibrator is calibrating, tree agitation is turned off in all cycles.
During OSM.Simulation.Model.HeightModel.Calibration in cycle 0 (local height model calibration), OSM takes into consideration the possibility that tree records may have been agitated when calculating number of user height observations in the tree list. If 5 height observations are required for local calibration, and partitioning increases observed height observations from say 3 to 9, local calibration will still not occur as the actual number of tree observations is less than 5. Also, since tree Weight property values are divided by partitions, weight integrity is preserved across all tree records. Therefore there should be no adverse effects of tree agitation on local height calibration.