1. Rapeepan Kantavichai, PhD student
Effect of Density Management and Fertilization in SMC Type I Installations on BH Branch Diameter
Rapeepan Kantavichai, PhD student
David Briggs
Eric Turnblom
April 25, 2007
SMC Spring Meeting
Vancouver, WA
File: Type_I_Branch_T_F_April_2007.ppt

2. Background

3. Knot diameter & product grade yield Grades of MSR Lumber Recovered
a. Largest knot on the log
b. LLAD “largest limb average diameter” = branch index (bix)
Grades of MSR Lumber Recovered
Knot diameters observed on log surface
With LLAD of log higher than 1.5 inches, very little yield of highest value grades
Is there a relationship between the branch index of a log and a simple tree measure?
Fahey, T.D., J.M. Cahill, GT.A. Snellgrove, L.S. Heath Lumber and Veneer Recovery from Intensively managed Young-Growth Douglas-fir. PNW-RP-437. USDA Forest Service PNW Research Station, Portland, OR.

4. Is there a relationship between the LLAD of a log and a simple tree measure?
Focus on BH region
Simple to measure
Hypothesis:
There is a relationship between the LLAD of a log and the diameter of the largest branch in the BH region (DLLBH) of the parent tree
LLAD or BIX of 16 ft log
Divide into N, S, E, W faces
Count and measure branches > 8mm diameter within 1 foot of either side of breast height i.e. from 3.5 to 5.5 feet
Sampled 12 trees per plot x 6 plots/spacing X 6 spacings = 432 trees
Diameter of largest branch at bh region

5. Divide into N, S, E, W faces
Count and measure branches > 8mm diameter within 1 foot of either side of breast height i.e. from 3.5 to 5.5 feet
Sampled 12 trees per plot x 6 plots/spacing X 6 spacings = 432 trees
DLLBH
Can translate a simple BH measurement into a log quality index or vice-versa

6. Branch Diameter before crown recession
Before crown recession above the branch it is alive and growing
Diameter at stem surface is increasing
Thin effect on mean branch diameter of trees in a stand
Chainsaw effect  increase, decrease, or not change depending on type of thinning
Response of residual stand  increased diameter growth for a longer time (slower crown recession)
Fertilizer effect on mean branch diameter of trees in a stand
Increased growth
Growth models that predict branch diameter at crown recession do not reflect what is observed on log surface later  next slide

7. Branch diameter after crown recession
After crown recession above the branch it is dead
Diameter at stem surface is decreasing (branch taper, loss of bark, shrinkage)
Thin effect on mean branch diameter of trees
Chainsaw effect  increase, decrease, or not change depending on type of thinning
Response of residual stand  may accelerate stem growth over dead tapered branch
Fertilizer effect on mean branch diameter of trees
May accelerate stem growth over the dead tapered branch

8. Experimental Sites

9. Sample: 9 Type I Installations with T/F Trials

10. Management regimes in the 9 Type I T/F Trials

11. Sample 6 plots x 9 installations = 54 plots
DLLBH from 2257 trees (about 40 trees/plot)
planted ( tpa; ave 500)
established (age 6-13; ave 9)
fertilized with 200 lb N as urea at establishment & every 4 years since
7 of 9 are King’s site class II; we used Flewelling’s SI in the analysis
22-32 years old at DLLBH measurement ( )
Virtually all BH branches were dead!

12. Analysis

13. Variables
Defined on next slide

14. Crown Recession Variables
Time of crown recession above BH (death of the largest BH branch)
Number of years since establishment until the first 4-year measurement when the crown base was above BH; no interpolation.
Also know years since crown recession until the DLLBH measurement.
DLLBH decreases as the bole grows over dead branch (taper, bark loss, & shrinkage)
Y_Total = Y_Since_CR + Y_Until_CR
Measurement points
Offset due to bark
Y_Since_CR (dead branch)
Y_Until_CR (live branch)
4 yrs
4 yrs
4 yrs
4 yrs
Our measure of crown recession above BH; no interpolation
Plot establishment

15. Plot (Stand) Level Statistics Variables

16. Confounding Problem with Stand Age
The 5 Installations planted 1980 & earlier were denser and on lower site quality than the 4 planted 1981 & later
Change in management planting philosophy
Bad luck in what was offered and when
AGE picks these historical patterns up  branch diameter decreases with age  not biologically meaningful so we dropped it.

17. Models
Model 1: Fixed treatment effects plus installation-wide pre-treatment stand variables
Model 2: Substitute actual post-treatment plot variables
Model 3: Use mean tree variables on each treatment plot
Model 4: Combination of tree, treatment effect, & stand

18. Statistical Analysis (all in SAS)
Backward elimination
Main effects
2-way interactions (centered)
Fit & Assumption Checks
AIC, RMSE, R2adj
Residuals
Test normality & homogeneous variance
Hold-out procedure: use model based on 8 installations to predict 9th
Independent test:
Predict of DLLBH on other Type I’s

19. Results

20. Model 1 Fixed treatment effects, installation-wide pre-treatment stand conditions, timing of crown recession above BH
DLLBH = ISPA1 – ISPA ISPA FERT – ISTEMS SI IRD Y_Since_CR
(RMSE = 2.94, radj2 = 0.82)
No interactions were significant
DLLBH decreases with more original trees per acre and more time since the crown receded above BH (growth over dead tapered branches)
DLLBH increases if trees were on higher site quality and larger for the number of original trees and increases with PCT to wider spacing and use of fertilizer.

21. Model 2 Individual plot stand conditions just after treatment, timing of crown recession above BH
DLLBH = FERT – (PSTEMS - 685) (SI30 – 26.5) (Y_Since_CR – 5) – (PSTEMS - 685) (SI30 – 26.5) (PSTEMS - 685) (Y_Since_CR – 5)
(RMSE = 2.80, radj2 = 0.84)
DLLBH decreases with more trees per acre remaining after PCT and more time since the crown receded above BH (growth over dead tapered branches)
DLLBH increases if trees were on higher site quality with use of fertilizer.
The effect of site index is moderated by the density of trees  higher density on a site results in lower DLLBH (earlier crown recession above BH hence longer time to grow over dead tapered branch)
The effect of Y_Since_CR is moderated by the density of trees  higher densities increase DLLBH (slower rate of growth over dead tapered branch)

22. Model 3 Use Descriptors of Mean Tree on Treatment Plots
DLLBH = (DBH – 26.9) – (HT – 18.8) – (HT – 18.8)(DBH – 26.9)
(RMSE = 2.52, radj2 = 0.87)
DLLBH decreases with greater dbh
DLLBH decreases with greater height; taller trees are likely to have a crown that has receded further above BH  more time since branch death for bole to grow over dead tapered branches
The effect of dbh is moderated by height; taller trees reduce DLLBH more than shorter trees of the same DBH  tall vs short trees differ in when crown recession occurs hence time to grow over a dead branch.

23. Model 4 Combining Average Tree & Treatment/Stand Variables
DLLBH = SI DBH – HT – Y_Since_CR
(RMSE = 2.18, radj2 = 0.90)
DLLBH = SI DBH – HT CR
(RMSE = 2.19, radj2 = 0.90)
DLLBH = (SI30 – 26.5) – (Y_Since_CR – 5.0) (DBH ) – (HT – 18.8) – (DBH ) (HT – 18.8)
(RMSE = 1.99, radj2 = 0.92)

24. Model 5 Can we use LIDAR? Using just height, crown and tree count data
DLLBH = CR
(RMSE = 4.74, radj2 = 0.53)
If site index and treatment information from stand records are included the model improves
DLLBH = FERT SI PSTEMS CL CR
(RMSE = 2.95, radj2 = 0.82)
The potential for linking quality measures with LIDAR needs further exploration

25. Using the Models

26. Models 1, 2, 4a & 4c use Y_Since_CR Need to estimate when crown recession occurs
At each 4-year measurement calculate the % of trees in each treatment plot with crown above BH (n = 241)
Logistic model:
Enter values for PSTEMS and SI and calculate % of trees with crown above BH as stand age increases (Table next slide)
Example: Suppore wish to estimate DLLBH at age 25 of a 1200 tpha stand growing on site 20m land

27. % of Trees with Crown above BH Manager can decide on threshold, say 85% (yellow)
Density, trees/ha
Site index, m
Site index, m
Site index, m
A stand with 1200 trees per hectare on a 20 m site will reach 85% crown recession above BH by age 22. Y_SINCE CR is measured after recession
To estimate DLLBH, at age 25, Y_SINCE CR = 4

28. Models 3 and 4 Use mean total height and mean DBH
User may have HT40 and QMD from a model
Conversion equations for the 54 plots:
DBH = QMD (RMSE = 0.242, radj2 = 0.99)
HT = HT40 (RMSE = 0.054, radj2 = 0.93)

29. Conclusion
At the treatment plot level, DLLBH can be predicted by several alternative formulations  flexibility for adoption
Stand age is not in the DLLBH models; the older installations were planted more densely and on lower sites than the newer installations so density and site variables prevailed.
Connectivity with LIDAR is promising and should be explored further.
WARNING: Remember all BH branches were dead! These models would not be appropriate to use for trees with the live crown below BH.

30. Future
Counterpart models for predicting DLLBH for individual trees is underway and looks promising will present at IUFRO Forest Growth and Wood Quality Conference, Portland, OR. August 7-10, 2007.
Extending the crown recession model to predict % of trees in a stand with the crown recession to any height (multinomial logit).
Type III’s would be interesting so we can capture dynamics before/after crown recession above BH.