Black spruce and balsam fir tree diameter growth increment (DBHI) models were developed from NL permanent sample plot (PSP) data using a Random Forests machine leaning approach as a function of Species, Age, Management (PCT or Planted – yes/no), DBH, basal area, basal area of larger trees (BAL), and a NL field assessed site productivity variable ‘IsGoodSite’. See Variant Input Tables for a description of the IsGoodSite variable. This model was represented in the OSM.NL variant using a large lookup table of predictor variables and responses, where diameter growth increment responses (cm/y) are linear interpolated between predictor value ranges in the lookup table. This approach was found to be superior in terms of model performance over non-linear approaches.
For black spruce and fir, some over prediction of DBHI was
evident for certain size trees (over predicted in small trees < 10cm DBH)
and stand basal area conditions (some over and some under depending on an
interaction between tree size and basal area). Therefore, additional bias
corrections were undertaken by first calculating the prediction modifier (mean
observed / mean predicted) by DBH and BA class and then predicting the modifier
as a function of basal area and DBH for each species. This approach was used to
put more equal emphasis on correcting bias across the full range of DBH and
stand basal area, rather than overly weight the corrections based on where most
of the data fell. Some weighting on DBH class was still used to promote
slightly better corrections for medium-large trees over small trees. See source
code for linear models developed.
For all other tree species, which all had <= 15,000 DBHI observations in NL to date, Nova Scotia diameter growth models were borrowed from the Acadian Variant (version 2.0.0.0, June 2024). Attempts were made to develop NL-specific other species models from these data, but model fits were often non-sensical or poorly extrapolated across all tree sizes or conditions. The Acadian NS diameter models require Biomass Growth Index (BGI) as a predictor, which is not an available site predictor in NL. Where NS models were leveraged, the BGI default value used by the NL variant was lowered to 3,000 kg/ha/year from a default value of 3,500 kg/ha/year in the Acadian model. Despite this change, most NS models still overpredicted NL DBHI, so species bias was further reduced by comparing NL observed to NS predicted DBHI to develop a bias correction multiplier for each of these ‘other’ species having >= 500 growth measurements. These modifiers were introduced as part of the NL DBHI final models.
After prediction modifiers were applied, predictions
improved slightly (see Table 1 below for accuracy statistics). In general, fir
and black spruce are predicted well, while ‘other’ species are considered ‘in
the ballpark’. These other species DBHI predictions should be re-evaluated with
more tree data in the future and alternative prediction methods explored.
Fit by C.R. Hennigar, FORUS Research 2024.
See also: Acadian Variant calibration.
Table 1. Diameter at breast
height growth increment model performance, comparing Acadian to NL, with and
without bias correction.
|
Species
and prediction model* |
r2 |
Mean Bias (cm/y) |
Mean Absolute Error (cm/y) |
DBHI Observations |
|
Black Spruce |
|
|
|
|
|
Acadian NS BGI=3000 |
41% |
-0.014 |
0.08 |
152,120 |
|
Acadian w Self. Calib. |
44% |
0.005 |
0.07 |
152,120 |
|
NL Variant |
57% |
-0.008 |
0.06 |
152,120 |
|
NL Variant w Bias Corr. Equ. |
58% |
0.001 |
0.06 |
152,120 |
|
Balsam Fir |
|
|||
|
Acadian NS BGI=3000 |
0% |
-0.071 |
0.11 |
161,417 |
|
Acadian w Self. Calib. |
38% |
0.003 |
0.08 |
161,417 |
|
NL Variant |
49% |
-0.007 |
0.07 |
161,417 |
|
NL Variant w Bias Corr. Equ. |
51% |
-0.001 |
0.07 |
161,417 |
|
White Birch |
|
|||
|
Acadian NS BGI=3000 |
0% |
-0.024 |
0.07 |
11,718 |
|
Acadian w Self. Calib. |
3% |
0.006 |
0.06 |
11,718 |
|
White Spruce |
|
|||
|
Acadian NS BGI=3000 |
37% |
-0.031 |
0.12 |
6,386 |
|
Acadian w Self. Calib. |
45% |
0.014 |
0.11 |
6,386 |
|
Tamarack |
|
|||
|
Acadian NS BGI=3000 |
0% |
-0.075 |
0.14 |
1,520 |
|
Acadian w Self. Calib. |
24% |
0.013 |
0.11 |
1,520 |
|
Red Pine |
|
|||
|
Acadian NS BGI=3000 |
60% |
-0.075 |
0.17 |
650 |
|
Acadian w Self. Calib. |
66% |
0.004 |
0.16 |
650 |
|
Red Maple |
|
|||
|
Acadian NS BGI=3000 |
3% |
-0.019 |
0.12 |
505 |
|
Acadian w Self. Calib. |
7% |
0.009 |
0.11 |
505 |
|
Trembling Aspen |
|
|||
|
Acadian NS BGI=3000 |
0% |
-0.140 |
0.17 |
517 |
|
Acadian w Self. Calib. |
2% |
-0.001 |
0.09 |
517 |
* Acadian NS BGI=3000 – Acadian Variant version 2.0.0.0, NS
zone, BGI = 3,000 kg/ha/y.
Acadian with Self. Calib. – same as Acadian default, but with OSM self-calibration
(simple scaling multiplier to correct mean bias).
NL Variant – models
fit for NL with only NL data, June 2024.
NL Variant with Bias Corr. Equ. – same as NL Variant default, but with linear bias corrections
as a function of DBH and stand basal area.