Table 1.
Generalized linear models of tree seedling and propagule density (stems/10 m2), and plot-level tree species composition.
| Generalized Linear Model | Distribution | df | pR2 (CV) | RMSE (CV) | MAE (CV) | P |
|---|---|---|---|---|---|---|
| Total Seedling & Propagule Stems = 3.26 + (FFI† × 0.74) − (MOIST‡ × 0.45) + 0.38(FFI × MS§) + (MS × 0.03) | Negative Binomial | 44 |
0.59 (0.44) |
56.92 (58.19) |
32.02 (39.13) |
<0.001 |
| Conifer Seedlings = 2.45 + (FFI × 0.99) − (TSF** × 0.8) + (%CON†† × 0.73) − (MS × 0.68) + 0.58(TSF × RO‡‡) + 0.38(FFI × MS) − (RO × 0.2) | Negative Binomial | 41 |
0.61 (0.43) |
54.29 (61.69) |
29.21 (39.63) |
<0.001 |
| Broadleaf Seedlings & Propagules = 0.98−2.32(%MIN§§ × BA***) − (%MIN × 2.13) − (BA × 1.4) − (TSF × 0.78) − (RO × 0.69) − 0.6(RO × MS) − (MS × 0.48) | Poisson | 41 |
0.45 (0.48) |
6.52 (7.26) |
4.64 (5.65) |
<0.001 |
| Proportion Conifer Stems = 0.60 + (FFI × 1.18) + (%MIN × 1.02) + (RO × 0.98) − 0.96(%MIN × MS) − (MS × 0.4) | Binomial | 43 |
0.49 (0.52) |
0.28 (0.3) |
0.22 (0.25) |
0.004 |
Significant (p ≤ 0.05) predictor variables are bolded in equations. Model fits are described using averages of 10-fold cross-validated (CV) root-mean-square-error (RMSE), mean absolute error (MAE), and pseudo R2 (pR2), derived from 100 repeats. Model degrees of freedom (n = 49) are reported in the column df. Model p-values were derived from χ2 tests of model deviance explained relative to a null model.
†FFI: Fire-free interval (years).
‡MOIST: Site moisture (from subxeric to subhydric).
§MS: Moisture stress (mm), as represented by site-specific cumulative anomalies of summer climatic moisture deficit (CMD) over 0 – 4 years post-fire.
**TSF: Time since last fire (years) at time of sampling.
††%CON: Proportion of conifer trees in the pre-fire cohort (%) calculated from basal area (m2 ha−1).
‡‡RO: Residual organic matter depth (cm).
§§%MIN: Proportion of exposed mineral soil and rock (%) at the surface.
***BA: Pre-fire basal area of trees (m2 ha−1).