terras

Modeling the social drivers of environmental sustainability among Amazonian indigenous lands using Bayesian networks

Paper title

Scripts are for the manuscript titled 'Modeling the social drivers of environmental sustainability among Amazonian indigenous lands using Bayesian networks' by Walker RS and J Paige.

Data

Data are from Indigenous Lands Territorial Consolidation Indicators System available at https://terrasmais.eco.br/v1/ and added to the project (terrasmais-2017.csv).

Phase 1 R package bnlearn

bnlearn (bnlearn.com) is an R package for learning the graphical structure of Bayesian networks, estimating their parameters and performing some useful inference. We use the boostrap feature for understanding causal structures in the WNAI data.

Phase 1 structure learning example: bootstrap multiple network structures

library(bnlearn)
boot <- boot.strength(df[,c(2:8)], R = 10001, cpdag = FALSE,
                      algorithm = "tabu")
boot[boot$strength > 0.5 & boot$direction >= 0.5, ]
avg.boot <- averaged.network(boot, threshold = .8)
plot(boot)
avg.boot

strength.plot(avg.boot, boot, threshold=.8,
              layout = "dot",
              shape = "rectangle")

Phase 2 R package brms

brms (paul-buerkner.github.io/brms/) is an R package that provides an interface to fit Bayesian generalized (non-)linear multivariate multilevel models using Stan. We use it to estimate the regression coefficients of the edges in our network while adjusting for spatial autocorrelation using a Gaussian process.

Phase 2 parameter learning example: bayesian path model with brms

library(brms)
mod1 <- bf(Environmental.integrity  ~ 1 + Legal.stability + Environmental.integrity.in.the.buffer.zone + gp(lat, long, scale=F))
mod2 <- bf(Governance ~ 1 + Legal.stability + gp(lat, long, scale=F))
mod3 <- bf(Absence.of.threats.due.to.infrastructure.projects ~ 1 + Territorial.integrity + gp(lat, long, scale=F))
mod4 <- bf(Environmental.integrity.in.the.buffer.zone ~ 1 + Absence.of.threats.due.to.infrastructure.projects + gp(lat, long, scale=F))
mod5 <- bf(Absence.of.pressure.from.infrastructure.projects ~ 1 + Absence.of.threats.due.to.infrastructure.projects + gp(lat, long, scale=F))
mv <- mvbf( mod1 + mod2 + mod3 + mod4 + mod5,
            rescor=FALSE )
m1 <- brm(data = df, 
         mv, prior = set_prior("normal(0,.1)", class = "b"), # group = ""),
         iter =  1e4, chains = 4, cores = 4, #save_all_pars = TRUE,
         control = list(adapt_delta = .999, max_treedepth = 20),
         seed = 1, backend = "cmdstanr")
prior_summary(m1)
m1
pl <- plot(m1, N = 4, ask = FALSE)
posterior_summary(m1)
bayes_R2(m1) # 
conditional_effects(m1, points=T)
plot(hypothesis(m1, "Governance_Legal.stability > 0"))