# Building an integral projection model
# https://tidyecology.com/posts/building-an-integral-projection-model/
#
# The R code of this Tidy Ecology tutorial, chunk by chunk, in the order of the post.
# Run it top to bottom in a fresh R session. Code the post shows but does not run
# is commented out. The prose, and the numbers it quotes, are in the post itself.
# Post last modified: 2026-10-15

## ---- setup ----
library(ggplot2)
library(dplyr)
library(tidyr)

te <- list(ink="#16241d", body="#2c3a31", forest="#275139", label="#46604a",
           sage="#93a87f", paper="#f5f4ee", line="#dad9ca", faint="#5d6b61",
           rust="#b5534e")
theme_te <- function(base_size = 11) {
  theme_minimal(base_size = base_size) +
    theme(text = element_text(colour = te$body),
          plot.title = element_text(colour = te$ink, face = "bold", size = rel(1.02)),
          plot.subtitle = element_text(colour = te$faint, size = rel(0.9)),
          axis.title = element_text(colour = te$label),
          axis.text = element_text(colour = te$faint),
          panel.grid.major = element_line(colour = te$line, linewidth = 0.3),
          panel.grid.minor = element_blank(),
          panel.background = element_rect(fill = te$paper, colour = NA),
          plot.background = element_rect(fill = te$paper, colour = NA),
          legend.key = element_rect(fill = te$paper, colour = NA),
          strip.text = element_text(colour = te$forest, face = "bold"))
}

## ---- simulate-data ----
set.seed(4021)
n <- 2600
pars <- list(a_s=-1.4, b_s=0.50,             # survival (logit scale)
             a_g= 1.9, b_g=0.65, sig_g=0.90,  # growth (Gaussian); equilibrium 1.9/0.35
             a_f=-6.0, b_f=0.80,              # flowering probability (logit)
             a_r= 0.42, b_r=0.24,             # recruits per flowering plant (log)
             mu_r=2.2, sig_r=0.70)            # recruit size (Gaussian)

z    <- rnorm(n, 5.2, 1.9)
z    <- z[z > 0.2 & z < 11.5]; n <- length(z)
surv <- rbinom(n, 1, plogis(pars$a_s + pars$b_s * z))
zt1  <- ifelse(surv == 1, rnorm(n, pars$a_g + pars$b_g * z, pars$sig_g), NA)
flow <- rbinom(n, 1, plogis(pars$a_f + pars$b_f * z))
nrec <- ifelse(flow == 1, rpois(n, exp(pars$a_r + pars$b_r * z)), 0)
rec_sizes <- rnorm(sum(nrec), pars$mu_r, pars$sig_r)

dat <- data.frame(z = z, surv = surv, zt1 = zt1, flow = flow, nrec = nrec)
c(individuals = n, survivors = sum(surv), flowered = sum(flow), recruits = sum(nrec))

## ---- fit-vital-rates ----
m_surv <- glm(surv ~ z, family = binomial, data = dat)
m_grow <- lm(zt1 ~ z, data = dat[dat$surv == 1, ])
sig_g_hat <- summary(m_grow)$sigma
m_flow <- glm(flow ~ z, family = binomial, data = dat)
m_rec  <- glm(nrec ~ z, family = poisson, data = dat[dat$flow == 1, ])
mu_r_hat  <- mean(rec_sizes); sig_r_hat <- sd(rec_sizes)

cs <- coef(m_surv); cg <- coef(m_grow); cf <- coef(m_flow); cr <- coef(m_rec)
round(rbind(survival = c(cs, NA), growth = c(cg, sigma = sig_g_hat),
            flowering = c(cf, NA), recruits = c(cr, NA)), 3)

## ---- fig-vital-rates ----
zg <- seq(0, 12, length.out = 200)
pnl <- function(d) factor(d, levels = c("Survival s(z)", "Growth z' | survived",
                                        "Flowering probability", "Recruits per flowering plant"))
pts <- bind_rows(
  transform(data.frame(z = dat$z, y = dat$surv), panel = "Survival s(z)"),
  transform(data.frame(z = dat$z[dat$surv == 1], y = dat$zt1[dat$surv == 1]), panel = "Growth z' | survived"),
  transform(data.frame(z = dat$z, y = dat$flow), panel = "Flowering probability"),
  transform(data.frame(z = dat$z[dat$flow == 1], y = dat$nrec[dat$flow == 1]), panel = "Recruits per flowering plant"))
pts$panel <- pnl(pts$panel)
lns <- bind_rows(
  transform(data.frame(z = zg, y = plogis(cs[1] + cs[2] * zg)), panel = "Survival s(z)"),
  transform(data.frame(z = zg, y = cg[1] + cg[2] * zg), panel = "Growth z' | survived"),
  transform(data.frame(z = zg, y = plogis(cf[1] + cf[2] * zg)), panel = "Flowering probability"),
  transform(data.frame(z = zg, y = exp(cr[1] + cr[2] * zg)), panel = "Recruits per flowering plant"))
lns$panel <- pnl(lns$panel)

ggplot() +
  geom_point(data = pts, aes(z, y), colour = te$sage, alpha = 0.28, size = 0.7,
             position = position_jitter(height = 0.02, width = 0)) +
  geom_abline(data = data.frame(panel = pnl("Growth z' | survived")),
              aes(slope = 1, intercept = 0), linetype = "dashed", colour = te$faint, linewidth = 0.35) +
  geom_line(data = lns, aes(z, y), colour = te$rust, linewidth = 0.9) +
  facet_wrap(~panel, scales = "free_y") +
  scale_x_continuous(breaks = c(0, 2.5, 5, 7.5, 10)) +
  labs(x = "Size z at time t (log scale)", y = NULL) +
  theme_te(11)

## ---- build-kernel ----
build_K <- function(cs, cg, sig_g, cf, cr, mu_r, sig_r, L, U, m) {
  h    <- (U - L) / m
  mesh <- L + (1:m - 0.5) * h                                   # bin midpoints
  S  <- plogis(cs[1] + cs[2] * mesh)                            # survival at each z
  G  <- outer(mesh, mesh, function(zp, z) dnorm(zp, cg[1] + cg[2] * z, sig_g))
  P  <- sweep(G, 2, S, "*") * h                                 # survival-growth
  Pf <- plogis(cf[1] + cf[2] * mesh)                            # flowering probability
  Rn <- exp(cr[1] + cr[2] * mesh)                               # expected recruits
  Cr <- dnorm(mesh, mu_r, sig_r)                                # recruit-size density
  Fk <- outer(Cr, Pf * Rn) * h                                  # fecundity
  list(K = P + Fk, mesh = mesh, h = h, S = S, colP = colSums(P))
}

L <- 0; U <- 12; m <- 100
fit    <- build_K(cs, cg, sig_g_hat, cf, cr, mu_r_hat, sig_r_hat, L, U, m)
ev     <- eigen(fit$K)
lambda <- Re(ev$values[1])
w      <- Re(ev$vectors[, 1]); w <- w / sum(w)                  # stable size distribution
lambda

## ---- fig-kernel ----
kdf <- data.frame(z = rep(fit$mesh, each = m), zp = rep(fit$mesh, times = m),
                  val = as.vector(fit$K))
ggplot(kdf, aes(z, zp, fill = val)) +
  geom_raster(interpolate = TRUE) +
  geom_abline(slope = 1, intercept = 0, linetype = "dashed", colour = "#ffffffaa", linewidth = 0.4) +
  scale_fill_gradient(low = te$paper, high = te$forest, trans = "sqrt", name = "K(z', z)") +
  coord_equal(expand = FALSE) +
  labs(x = "Size z at time t", y = "Size z' at time t+1") +
  theme_te(11) + theme(panel.grid = element_blank())

## ---- recover-truth ----
tru <- build_K(c(pars$a_s, pars$b_s), c(pars$a_g, pars$b_g), pars$sig_g,
               c(pars$a_f, pars$b_f), c(pars$a_r, pars$b_r), pars$mu_r, pars$sig_r, L, U, m)
lambda_true <- Re(eigen(tru$K)$values[1])
c(fitted = lambda, true = lambda_true, difference = lambda - lambda_true)
