3  Results

3.1 Regeneration composition

3.1.1 Basal area

Composition of regeneration in terms of basal area per acre represented by each species in a 4-meter radius vegetation plot was modeled as a gamma distribution with a log link with fixed effects for treatment and species, and random intercepts for site x species interaction. Dispersion was modeled separately as a function of species, using a log link and the rate of zeros was modeled using the logit link, for each species as well (Listing 3.1).

Listing 3.1 
Family: Gamma (log) 
Conditional: ba_ha ~ treat * spp + (1 | site:spp)  
Dispersion: ~spp (log) 
Hurdle: ~spp (logit) 
Adding missing grouping variables: `spp`
Adding missing grouping variables: `spp`
Adding missing grouping variables: `spp`

According to predictions made from this model, there was not enough evidence to confirm a statistically significant difference between treatments. On average, we expect about 0.11 m2 ha-1 of basal area across treatments. The greatest basal area of other species was in the HD treatment which was 0.12 m2 ha-1 greater than in the HA treatment (p = 0.26). The GS and LD treatments were intermediate.

On average, for Douglas-fir, we expect about 0.17 m2 ha-1 of basal area across treatments. The greatest basal area of Douglas-fir was in the GS treatment which was 0.12 m2 ha-1 greater than in the HA treatment (p = 0.76). The LD, HA, and HD treatments were all comparatively similar.

Redwood basal area regeneration showed the greatest treatment response. Where the GS treatment had the greatest basal area of redwood regeneration at m2 ha-1, which was 9.28 m2 ha-1 greater than in the HD treatment (p = 0.19). The LD and HD treatments were intermediate.

Tanoak basal area regeneration was intermediate between that of redwood and Douglas-fir and other species. The GS and LD treatments had similar responses, as did the HA and HD treatments. The GS treatment resulted in 2.24 m2 ha-1 of tanoak basal area, which was 1.33 m2 ha-1 greater than in the HA treatment (p = 0.18).

Table 3.1: Grand means (m2 ha-1) for basal area of regeneration of each species across treatments 10 years after the initiation of a multi-age redwood forest.
spp 1 estimate SE df asymp.LCL asymp.UCL
other overall 0.11 0.04 Inf 0.03 0.19
df overall 0.17 0.06 Inf 0.05 0.28
rw overall 4.03 1.52 Inf 1.04 7.01
to overall 1.59 0.47 Inf 0.67 2.50 
Adding missing grouping variables: `spp`
Figure 3.1: Basal area (m2 ha-1) modeled at the vegetation plot level for four harvest treatments and four species classes (n = 16). Gray bars represent the 95% confidence interval (α = 0.05), black dots indicate the mean, and blue arrows provide a means of assessing the statistical significance of pairwise differences among treatments. Arrows are drawn so that when two arrows just meet, the p-value for that difference is 0.05 and overlapping arrows indicate a p-values greater than 0.05. 
Adding missing grouping variables: `spp`
Table 3.2: Basal area (m2 ha-1) modeled at the vegetation plot level for four harvest treatments and four species classes (n = 16).
spp treatment estimate SE df asymp.LCL asymp.UCL
other gs 0.13 0.08 Inf -0.03 0.29
other ld 0.11 0.06 Inf -0 0.23
other ha 0.04 0.02 Inf -0 0.07
other hd 0.16 0.07 Inf 0.02 0.3
df gs 0.28 0.12 Inf 0.04 0.52
df ld 0.11 0.05 Inf 0.02 0.21
df ha 0.16 0.08 Inf 0.01 0.31
df hd 0.11 0.05 Inf 0.01 0.21
rw gs 10.12 4.74 Inf 0.84 19.41
rw ld 3.63 1.91 Inf -0.12 7.38
rw ha 1.51 0.78 Inf -0.01 3.04
rw hd 0.85 0.52 Inf -0.17 1.86
to gs 2.24 0.79 Inf 0.7 3.79
to ld 1.94 0.69 Inf 0.58 3.3
to ha 0.92 0.33 Inf 0.28 1.56
to hd 1.25 0.44 Inf 0.39 2.11 
Adding missing grouping variables: `spp`
Table 3.3: Pairwise comparisons of treatments within species. P-values are adjusted using the Tukey method for comparing families of four estimates.
spp contrast estimate SE df p.value
other gs - ld 0.02 0.09 Inf 1
other gs - ha 0.09 0.08 Inf 0.6
other gs - hd -0.03 0.09 Inf 0.99
other ld - ha 0.08 0.06 Inf 0.53
other ld - hd -0.04 0.07 Inf 0.91
other ha - hd -0.12 0.07 Inf 0.26
df gs - ld 0.17 0.11 Inf 0.46
df gs - ha 0.12 0.12 Inf 0.76
df gs - hd 0.17 0.11 Inf 0.42
df ld - ha -0.05 0.08 Inf 0.93
df ld - hd 0 0.05 Inf 1
df ha - hd 0.05 0.07 Inf 0.91
rw gs - ld 6.49 4.6 Inf 0.49
rw gs - ha 8.61 4.56 Inf 0.23
rw gs - hd 9.28 4.64 Inf 0.19
rw ld - ha 2.12 1.8 Inf 0.64
rw ld - hd 2.79 1.83 Inf 0.42
rw ha - hd 0.67 0.79 Inf 0.83
to gs - ld 0.31 0.69 Inf 0.97
to gs - ha 1.33 0.65 Inf 0.18
to gs - hd 0.99 0.64 Inf 0.4
to ld - ha 1.02 0.58 Inf 0.29
to ld - hd 0.69 0.57 Inf 0.63
to ha - hd -0.33 0.37 Inf 0.8 
Adding missing grouping variables: `treat`
Adding missing grouping variables: `treat`

Figure 3.2 shows the same model as Figure 3.1, but with an emphasis on treatment comparisons between redwood and tanoak. This shows that we expect on average, 7.88 m2 ha-1 greater redwood basal area than tanoak basal area in the GS treatment (p = 0.1), about 1.69 m2 ha-1 in the LD treatment (p = 0.4), and about 0.6 m2 ha-1 in the HA treatment (p = 0.48). In the HD treatment, we expect to see slightly higher tanoak basal area (0.55).

Uncertainty in average Redwood basal area across sites, indicated by the size of 95% confidence intervals, is much greater than that of tanoak in the GS treatment, but this difference diminishes such that GS > LD > HA > HD. In the HD treatment redwood and tanoak average basal area uncertainty across sites is very similar. (Figure 3.1).

Adding missing grouping variables: `treat`
Figure 3.2: Basal area (m2 ha-1) modeled at the vegetation plot level for four harvest treatments and two species classes (n = 16). Gray bars represent the 95% confidence interval, black dots—the mean, and non-overlapping blue arrows signify statistical significance (α = 0.05).

3.1.2 Other species

Other species included grand fir, madrone, and California wax-myrtle, of which there was a total of 23, 28, and 16 observations across our 16 macro plots (comprising 64 tree density plots). Generally, each plot had between 0 and 9 observations of other species, except for one macro plot with the LD treatment, which had 16 observations (data not shown).

3.1.3 Douglas-fir counts

Counts of regenerating Douglas-fir seedlings per vegetation plot (n = 16) were analyzed for differences between harvest treatments using a negative binomial response with a log link, fixed effects for treatment, random effects for site and site x treatment interaction (Listing 3.2).

Listing 3.2 
Family: nbinom1 (log) 
Conditional: n ~ treat + (1 | site) + (1 | site:treat)

This model for Douglas-fir counts does not indicate any statistically significant differences between treatments. Generally, we expect about 2 seedlings per 4-meter-radius plot, or about 413 seedlings per hectare (Figure 3.3).

Figure 3.3: Vegetation plot level counts of regenerating Douglas-fir seedlings in four harvest treatments 10 years after harvest (n = 16). Results have been scaled to stems per hectare (4-meter radium plots).
Table 3.4: Vegetation plot level counts of regenerating Douglas-fir seedlings in four harvest treatments 10 years after harvest (n = 16). Results have been scaled to stems per hectare from 4-meter radium plots.
Treatment estimate asymp.LCL asymp.UCL
gs 479 88 869
ld 394 60 728
ha 435 65 805
hd 632 149 1115

3.2 Sprout heights

3.2.1 Height increment

The selected height increment model used a normal response distribution on the identity link. It included treatment, growth period, species, and the interaction of species and growth period as fixed effects. A random intercept was included for tree (multiple observations) and macro-plot, and an another random effect allowed the response to vary by species differently for each macro plot. The dispersion parameter for the response was modeled (with a log link) as a function of treatment, growth period, species and all three-way interactions (Listing 3.3).

Listing 3.3 
Family: gaussian (identity) 
Conditional: ht_inc ~ treat + year * spp + (1 | tree) + (0 + spp | plot)  
Dispersion: ~spp * year * treat (log) 
Adding missing grouping variables: `spp`
Adding missing grouping variables: `spp`

The model selected based on AIC lacks a treatment x species interaction, suggesting that there is not evidence that treatments affected species differentially. It also lacks a treatment x year interaction. This means that there was not enough evidence to support that treatment was related to changes in growth rate.

The inclusion of treatment factors in the model (0.001 ≤ p < 0.03) suggests that the levels of treatment were associated with different growth rates across species and years. And the species x year interaction (p < 0.001) suggests changes in growth rates are different for redwood and tanoak (Figure 3.4).

For tanoak, height increment was greatest in the GS treatment at 0.48 m yr-1. This was about 0.17 m yr-1 more than in the HA and HD treatments, which were very similar at about 0.31, 0.03, , 0.24, 0.37 m yr-1.

Redwood followed a similar pattern but with more pronounced differences between treatments. Height increment for redwood in the GS treatment was 0.96 m yr-1, which was about 0.4 m yr-1 greater than in the HD treatment (p = 0). Additionally, there was evidence that the GS treatment led to greater height increment than the LD treatment by about 0.17 m yr-1 (p = 0). And the LD treatment was higher than the HA treatment by about 0.15 m yr-1 (p = 0).

Adding missing grouping variables: `spp`
Figure 3.4: Estimated marginal means for the effect of harvest treatment on redwood and tanoak sprout height increment, averaged over two growth periods, ten years after harvest. Gray bars represent confidence intervals and statistical significance (α = 0.05) is indicated by non-overlapping blue arrows. 
Adding missing grouping variables: `spp`
Table 3.5: Estimated marginal means for the effect of harvest treatment on redwood and tanoak sprout height increment, averaged over two growth periods, ten years after harvest.
spp treatment estimate SE df asymp.LCL asymp.UCL
LIDE GS 0.48 0.034 Inf 0.41 0.54
LIDE LD 0.39 0.033 Inf 0.32 0.46
LIDE HA 0.31 0.032 Inf 0.24 0.37
LIDE HD 0.3 0.034 Inf 0.23 0.37
SESE GS 0.96 0.052 Inf 0.86 1.06
SESE LD 0.79 0.051 Inf 0.69 0.89
SESE HA 0.63 0.05 Inf 0.54 0.73
SESE HD 0.56 0.052 Inf 0.46 0.66 
Adding missing grouping variables: `spp`
Adding missing grouping variables: `spp`

Redwood growth slowed from 0.80 to 0.67 m yr-1 in the second period and tanoak slowed from 0.39 to 0.34 m yr-1.

Redwood grew faster than tanoak, but slowed down more relative to it in the second period. Height increment for redwood was 0.42 m yr-1 greater than tanoak in the first period and 0.33 m yr-1 greater than tanoak in the second period (Figure 3.5).

Adding missing grouping variables: `spp`
Figure 3.5: Estimated marginal means for the effect of growth period on redwood and tanoak sprout height increment, averaged over four harvest treatments, from years 1 to 5, and years 5 to 10 after harvest, plotted alongside actual data. Gray bars represent confidence intervals and statistical significance (α = 0.05) is indicated by non-overlapping blue arrows. 
Adding missing grouping variables: `spp`
Table 3.6: Estimated marginal means for the effect of growth period on redwood and tanoak sprout height increment, averaged over four harvest treatments, from years 1 to 5, and years 5 to 10 after harvest.
spp year estimate SE df asymp.LCL asymp.UCL
LIDE 5 0.39 0.017 Inf 0.35 0.42
LIDE 10 0.35 0.017 Inf 0.31 0.38
SESE 5 0.8 0.043 Inf 0.72 0.89
SESE 10 0.67 0.043 Inf 0.59 0.75

3.2.2 Height at year 10

Sprout heights at year 10 were modeled with a normal response and a log link. The best model included species and treatment, but no interactions in the fixed effects. This suggests that treatments do not affect species differentially in terms of the mean response (height at year 10). It also included a model for dispersion (log link) with predictors species, treatment, and their interaction (Listing 3.4).

Listing 3.4 
Family: gaussian (log) 
Conditional: ht ~ treat + spp + (0 + spp | plot)  
Dispersion: ~spp * treat (log) 
Adding missing grouping variables: `spp`
Adding missing grouping variables: `spp`

Because the best model did not contain a species x treatment interaction for the mean response, treatment comparisons are parallel between species. The GS treatment resulted in greater heights in year 10 than the other treatments (0.001 < p < 0.05). Predicted mean height for redwood ranged from 10.64 m in the GS treatment to 6.3 m in the HD treatment. For tanoak, predicted mean height ranged from 5.2 in the GS treatment to 3.08 in the HD treatment. Predicted mean heights followed the pattern GS > LD > HA > HD (Figure 3.6).

Adding missing grouping variables: `spp`
Figure 3.6: Predicted mean height and 95% confidence intervals (gray bars) for redwood and tanoak stump sprouts 10 years after harvest using four different harvest treatments. Non-overlapping blue arrows indicate statistically significant differences between treatments within a species. 
Adding missing grouping variables: `spp`
Table 3.7: Height (m) of measured redwood and tanaok sprouts 10 years after harvest treatments with four different over-story densities.
spp year estimate SE df asymp.LCL asymp.UCL
LIDE GS 5.2 0.41 Inf 4.4 6
LIDE LD 3.9 0.31 Inf 3.3 4.5
LIDE HA 3.3 0.27 Inf 2.8 3.8
LIDE HD 3.1 0.25 Inf 2.6 3.6
SESE GS 10.6 0.9 Inf 8.9 12.4
SESE LD 8 0.69 Inf 6.6 9.3
SESE HA 6.7 0.59 Inf 5.5 7.8
SESE HD 6.3 0.55 Inf 5.2 7.4 
Adding missing grouping variables: `spp`
Table 3.8: Pairwise comparisons of treatments within species for height (m) of measured redwood and tanoak sprouts 10 years after harvest.
spp contrast estimate SE df p.value
LIDE GS - LD 1.3 0.5 Inf 0.05
LIDE GS - HA 1.93 0.48 Inf 0
LIDE GS - HD 2.12 0.47 Inf 0
LIDE LD - HA 0.63 0.4 Inf 0.4
LIDE LD - HD 0.82 0.39 Inf 0.16
LIDE HA - HD 0.19 0.36 Inf 0.95
SESE GS - LD 2.66 1.03 Inf 0.05
SESE GS - HA 3.94 0.98 Inf 0
SESE GS - HD 4.33 0.97 Inf 0
SESE LD - HA 1.29 0.82 Inf 0.4
SESE LD - HD 1.68 0.81 Inf 0.16
SESE HA - HD 0.39 0.75 Inf 0.95

3.3 Fuels

3.3.1 Pre-pct

Gamma distributed, linear multi-level models, with a log link were used for all six fuel class responses. Random intercepts were specified for three levels of nesting, representing sites, treatment blocks, and transect corners. All models except for the duff & litter model included a hurdle model to account for zero, which was modeled with a logit link. For the 10-hr fuel model, the hurdle portion was modeled as a function of treatment, and for the others, it was modeled as a single rate for all observations. The 10-hr fuel model also included a dispersion model, which was modeled with a log link, using treatment as a predictor (Table 3.9).

Table 3.9: Model specifications for six fuel classes before pct.
class Family Link Conditional Dispersion (log) Hurdle (logit)
Duff & Litter Gamma log load ~ treatment + (1 | site) + (1 | block) + (1 | corner) ~1 ~0
1-hr Gamma log load ~ treatment + (1 | site) + (1 | block) + (1 | corner) ~1 ~1
10-hr Gamma log load ~ treatment + (1 | site) + (1 | block) + (1 | corner) ~treatment ~treatment
100-hr Gamma log load ~ treatment + (1 | site) + (1 | block) + (1 | corner) ~1 ~1
1,000-hr Gamma log load ~ treatment + (1 | site) + (1 | block) + (1 | corner) ~1 ~1
Vegetation Gamma log load ~ treatment + (1 | site) + (1 | block) + (1 | corner) ~1 ~1

TODO: I’m here

For Duff & Litter, the largest difference was between the HD and HA treatments. The HD treatment had about 1.4 times more duff and litter (p = 0.07). Generally, all treatments were similar, with estimated loading of around 50 Mg ha-1. One-hour fuels were around 50% higher in the HA treatment compared to the LD and GS treatments (p = 0.07, and p = 0.01, respectively), with mean differences of around 0.5 Mg ha-1. Ten, hundred and thousand-hour fuels were statistically, very similar across treatments (p = 0.7 — p = 1). Point estimates varied by about 1, 3, and <20 Mg ha-1 for ten, hundred, and thousand-hour fuels, respectively. Vegetative fuel loading was greatest in the GS treatment, with an expected value of 28.5 Mg ha-1, which was about 2.7 times greater than in HA (p = 0.01) (Figure 3.7).

Figure 3.7: Estimated marginal means (black dots) confidence intervals (gray bands) and comparisons (blue arrows) of fuel loading across four treatments for six different fuel-class models. Non-overlapping blue arrows indicates statistical significance at the α = 0.05 level.
Table 3.10: Estimated marginal means (Mg ha-1) for six fuel classes and four overstory treatments before pct
class treatment estimate SE df asymp.LCL asymp.UCL
dufflitter gs 47.93 5.69 Inf 36.76 59.09
dufflitter ld 45.09 5.35 Inf 34.6 55.59
dufflitter ha 40.01 4.75 Inf 30.71 49.32
dufflitter hd 54.4 6.46 Inf 41.74 67.06
onehr gs 0.55 0.11 Inf 0.33 0.77
onehr ld 0.66 0.13 Inf 0.4 0.92
onehr ha 1.2 0.24 Inf 0.73 1.68
onehr hd 1.01 0.2 Inf 0.61 1.41
tenhr gs 3.71 0.65 Inf 2.43 4.98
tenhr ld 3.31 0.52 Inf 2.3 4.32
tenhr ha 2.9 0.63 Inf 1.68 4.13
tenhr hd 2.91 0.45 Inf 2.04 3.79
hundhr gs 11.17 1.92 Inf 7.41 14.93
hundhr ld 10.03 1.76 Inf 6.58 13.47
hundhr ha 8.9 1.52 Inf 5.92 11.88
hundhr hd 8.76 1.49 Inf 5.84 11.69
thoushr gs 43.08 13.84 Inf 15.95 70.22
thoushr ld 26.25 8.58 Inf 9.43 43.06
thoushr ha 28.06 9.58 Inf 9.29 46.83
thoushr hd 39.75 12.94 Inf 14.38 65.11
veg gs 29.94 7.78 Inf 14.69 45.19
veg ld 20.88 5.32 Inf 10.45 31.3
veg ha 10.99 2.86 Inf 5.39 16.6
veg hd 16.86 4.32 Inf 8.4 25.32
Table 3.11: Grand means (Mg ha-1) for six fuel classes before pct.
class 1 estimate SE df asymp.LCL asymp.UCL
dufflitter overall 46.86 4.21 Inf 38.6 55.1
onehr overall 0.86 0.12 Inf 0.63 1.1
tenhr overall 3.21 0.28 Inf 2.65 3.8
hundhr overall 9.71 1.1 Inf 7.56 11.9
thoushr overall 34.29 6.31 Inf 21.93 46.6
veg overall 19.67 3.52 Inf 12.77 26.6

3.3.2 Post-pct

The response for all six, post-pct fuel classes were modeled with a gamma distribution and a log link, and included the same multi-level random effects as for the pre-pct models. Dispersion models with treatment as the only predictor were included for 1-hr and 100-hr fuel classes. All models included a hurdle portion to model zeros using a logit link. For 100-hr fuels, this model included treatment and site as predictors, and for the rest, a constant rate for all observations was used (Table 3.12).

Table 3.12: Model specifications for six fuel classes after pct.
class Family Link Conditional Dispersion (log) Hurdle (logit)
1-hr Gamma log load ~ treatment + (1 | site) + (1 | block) + (1 | corner) ~treatment + site ~1
10-hr Gamma log load ~ treatment + (1 | site) + (1 | block) + (1 | corner) ~1 ~1
100-hr Gamma log load ~ treatment + (1 | site) + (1 | block) + (1 | corner) ~treatment + site ~treatment + site
1,000-hr Gamma log load ~ treatment + (1 | site) + (1 | block) + (1 | corner) ~1 ~1
Vegetation Gamma log load ~ treatment + (1 | site) + (1 | block) + (1 | corner) ~1 ~1
Vegetation Difference Gamma log load ~ treatment + (1 | site) + (1 | block) + (1 | corner) ~1 ~1

Post-pct resulted in greater stratification of treatments (Figure 3.8). One-hour fuels for most treatments were around 2.4 Mg ha-1, but the HA treatment had around half of that amount (p = 0.01 to p = 0.02). The GS treatment had the greatest 10-hr fuel loading with 8.8 Mg ha-1, which was about 1.6, 2.3 and 2.9 times greater than the LD, HA, and HD treatments respectively (p = 0.03, p < 0.001, for the others, respectively). The LD treatment also had about 1.7 times more 10-hr fuels that the HD treatment (5.4 vs. 3 Mg ha-1, p = 0.001). Hundred-hour fuels were also greatest in the GS treatment, with an average of about 19 Mg ha-1, which was about 2.6 times greater than in the HD treatment (7 Mg ha-1, p < 0.001). Thousand-hour fuels were greatest in the HD treatment, with 80 Mg ha-1, which was about 2.7 times greater than the LD and HD treatments (p = 0.03 and p = 0.05, respectively). Fuel loading for live vegetation was similar across treatments at around 2.5 Mg ha-1. The pre-post vegetation difference was greatest in the GS treatment at about 31 Mg ha-1, which was 2.5 and 2.8 times the HD and HA treatments, respectively (p ≈ 0.01).

Figure 3.8: Estimated marginal means (black dots) confidence intervals (gray bars) and comparisons (blue arrows) of fuel loading across four treatments for six different fuel-class models. Non-overlapping blue arrows indicates statistical significance at the α = 0.05 level. Vegetation difference equals the transect level difference in vegetation load in the pre and post-pct conditions. This represents slash fuels recruited to the forest floor following the pre-commercial thinning.
Table 3.13: Estimated marginal means (Mg ha-1) for six fuel classes and four overstory treatments before pct
class treatment estimate SE df asymp.LCL asymp.UCL
onehr gs 2.6 0.5 Inf 1.63 3.6
onehr ld 2.8 0.55 Inf 1.71 3.9
onehr ha 1.4 0.25 Inf 0.89 1.9
onehr hd 2.2 0.38 Inf 1.47 2.9
tenhr gs 9 1.7 Inf 5.68 12.3
tenhr ld 5.5 1.05 Inf 3.44 7.6
tenhr ha 3.9 0.73 Inf 2.41 5.3
tenhr hd 3.1 0.59 Inf 1.92 4.2
hundhr gs 19.1 4.52 Inf 10.26 28
hundhr ld 13.2 2.97 Inf 7.36 19
hundhr ha 10.5 2.41 Inf 5.81 15.3
hundhr hd 7.4 1.74 Inf 4.03 10.8
thoushr gs 43.9 10.61 Inf 23.08 64.7
thoushr ld 22.8 5.33 Inf 12.31 33.2
thoushr ha 23.2 6.21 Inf 11.04 35.4
thoushr hd 60.5 16.55 Inf 28.05 92.9
veg gs 2.7 0.87 Inf 0.96 4.4
veg ld 1.9 0.61 Inf 0.7 3.1
veg ha 3.2 1.12 Inf 1.06 5.4
veg hd 2.2 0.71 Inf 0.78 3.6
veg_diff gs 29.6 7.99 Inf 13.95 45.3
veg_diff ld 17.2 4.22 Inf 8.89 25.4
veg_diff ha 10.6 2.98 Inf 4.74 16.4
veg_diff hd 11.8 3.02 Inf 5.91 17.7
Table 3.14: Grand means (Mg ha-1) for six fuel classes after pct.
class 1 estimate SE df asymp.LCL asymp.UCL
onehr overall 2.3 0.36 Inf 1.5 3
tenhr overall 5.4 0.84 Inf 3.7 7
hundhr overall 12.6 2.2 Inf 8.3 16.9
thoushr overall 37.6 5.61 Inf 26.6 48.6
veg overall 2.5 0.65 Inf 1.2 3.8
veg_diff overall 17.3 3.33 Inf 10.8 23.8

3.3.3 Pre-post commercial thinning comparison

Pre-commercial thinning led to a small increase in average 100-hr fuel loading, only for the GS treatment, increased 10-hr fuels in the GS and LD treatments, and increased 1-hr fuels for all but the HA treatment (Figure 3.9), although these results are not statistically comparable, due to slightly different model structures.

Figure 3.9: Estimated marginal means (black dots) and confidence intervals (colored bars) of fuel loading across four treatments and five different fuel classes, before and after PCT. Pre- and Post PCT models within a treatment are from similar, but not necessarily identical models.