Post on 27-Jan-2017
Using Bayesian hierarchical models to estimate forest demographic processes
Carrie R. Levine1, Natalie S. van Doorn1,2, and John J. Battles1
1 Environmental Science, Policy and Management, UC Berkeley2 US Forest Service Pacific Southwest Research Station
LTER ASM 2015Working Group: Sources of Uncertainty in Ecosystem Monitoring
1 September 2015
Advantages of Bayesian models for monitoring
Model accounts for variation at multiple scales.• This example: The likelihood for the entire community is
the product of the species level probability and the community level probability.
• Other possible sources: over space, over time, uneven sampling intervals, uneven sample size, etc.
Scant observations for rare species are informed by the overall community mean and the estimates for individual species and the overall community are more robust than with a non-hierarchical model.
RGR[i,j](log norm dist)
p[i](log norm dist)
μ(unif dist)
τ.with(gamma dist)
τ.btw(gamma dist)σ Species [i]
Community
Individual [i,j]
Hierarchical model structure:Relative growth rate (% yr-1)
LevelsModel
for j IN 1:ind
for i IN 1:spp
0 1 2 3 4 5 6 7 8 9 100
0.2
0.4
0.6
0.8
1
1.2
1.4 1957-...G
row
th ra
te o
f tre
es >
24 c
m D
BH
(% y
r-1)
Douglas-fir
Comm. meann = 546
Relative growth rate for an old-growth Sierra Nevada mixed-conifer forest
White fir
Incense
cedar
Ponderosa
pine
Sugar pine
n = 3390n = 3339n = 3792
Relative growth rate for an old-growth Sierra Nevada mixed-conifer forest
Constraining estimates of rare species using the community-level mean
Example: Mycorrhizal species abundance measured pre- and post-fire
Species Number
Cha
nge
in a
bund
ance
(OT
units
)