library(tidyverse)
library(readr)
library(gt)
library(gtExtras)
library(leaflet)
library(sf)
library(terra)
source(here_rel("R", "funs_data.R"))
source(here_rel("R", "funs_graphics.R"))
source(here_rel("R", "funs_plot_tables.R"))
theme_set(theme_public())hab_map <- here_rel("notebook", "figs", "Habitat_map.png")
if (file.exists(hab_map)) knitr::include_graphics(hab_map)
The study was conducted across French Guiana (2-6° N, 51-54° W). The average temperature is around 26° C and precipitation ranging from 2000 mm in the south and extreme northwest to more than 4000 mm in the northeast is distributed in wet and dry seasons episodically perturbed by El Niño Southern oscillation events (Paget 1999). The geological substrate and associated landforms derive from the granitic Guiana shield (Hammond 2005; Kroonenberg, Jébrak, and Heuret 2020). Relief is generally subdued (mean elevation: 140 m), with a maximum of 831 m at Mt Bellevue. The geomorphology of the granitic substrate comprises a diverse set of landscapes forming a mosaic of plateaus, multiconvex hills, and valley bottoms associated with strong soil contrasts, analogous to terra firme and seasonally flooded topo-edaphic conditions (Guitet et al. 2013, 2015).
Evergreen ombrophilous forest covers more than 90% of the 83,000 km² territory (de Granville 1985; Gond et al. 2011), harbouring 1783 recorded tree species in 415 genera and 87 families (Molino et al. 2022). The French Guiana floristic background is deeply imprinted by the general Amazonian floristic background (11 % to 13 % of the 15,300 species recorded in the basin) and that of the Guiana Shield (20 % of species are endemic to the Guiana Shield), and has its own endemism rate of 5 % (Guitet, Brunaux, and Molino 2021; ter Steege et al. 2020). Within communities, species composition is marked by a group of 50 hyperdominant species (30 % of 50 % of stems inventoried at the regional level), but whose abundance per species does not exceed 15 % to the stand level. Taxonomic composition varies along regional gradients: coastal forests are dominated by Chrysobalanaceae, Lecythidaceae and slow-growing, hard-wooded Fabaceae, whereas southern forests are richer in Burseraceae, Urticaceae, and fast-growing Fabaceae species of the mimosoid clade (Guitet et al. 2018). In addition, white-sand forests occur in the northwestern sector (Blancaneaux et al. 1972), with a distinctive floristic assemblage (Baraloto et al. 2021).
An inventory unit is the statistical sampling unit used in analyses (i.e. the polygon or subplot for which tree and palm abundances and environmental predictors are summarised). Where original plots are larger than the modelling unit, plots are subdivided into subplots and these subplots are treated as inventory units.
A sampling site is a named spatial group (“Site” in the metadata) that aggregates neighbouring inventory units (potentially from different networks) within the same forest area. Sampling sites are used to stratify model training and external validation so that validation units remain geographically independent from the corresponding training units.
The inventory units come from three forest inventory networks: GuyaDiv (Molino, Sabatier, and Engel 2024), Guyafor (Mirabel et al. 2025), and GuyENTRY (Baraloto et al. 2021), covering the territory of French Guiana and resulting from a collective effort by field ecologists and botanists since 1980 (Figure 2). These three forest inventory networks gather respectively 1,739, 537 and 78 inventory units covering a cumulative area of 183 ha, 291 ha and 38.5 ha.
Shp_plots |>
st_drop_geometry() |>
group_by(DB) |>
summarise(
`# inventory units` = n(),
`Minimum area (ha)` = min(Area, na.rm = TRUE),
`Median area (ha)` = median(Area, na.rm = TRUE),
`Maximum area (ha)` = max(Area, na.rm = TRUE)
) |>
rename(`Inventory network` = DB)|>
gt() |>
fmt_number(columns = c(`Minimum area (ha)`, `Median area (ha)`, `Maximum area (ha)`), decimals = 3) |>
opts_theme()| Inventory network | # inventory units | Minimum area (ha) | Median area (ha) | Maximum area (ha) |
|---|---|---|---|---|
| GuyENTRY | 77 | 0.100 | 0.100 | 0.100 |
| GuyaDiv | 1682 | 0.010 | 0.030 | 2.000 |
| GuyaFor | 222 | 0.062 | 1.000 | 6.000 |
57,043, 158,344 and 11,056 individuals are inventoried, with a taxonomic identification rate at the species level of 98%, 68% and 82.5% in the GuyaDiv, Guyafor and gentry network plots respectively, and 100% at the genus level for all individuals. Species-name homogenisation relied on the R package WorldFlora version 2024-12-20 (Kindt 2020), and on the species list of French Guiana (Molino et al. 2022).
Lat_Saul <- 3.6163888889
Lon_Saul <- -53.2011111111
# Fill color = inventory network
pal_net <- colorFactor(
palette = c("red", "darkgreen", "navy"),
domain = c("GuyaDiv", "GuyaFor", "GuyENTRY")
)
# Outline = train/validation
pal_split <- colorFactor(
palette = c("white", "darkred"),
domain = c("train", "validation")
)
leaflet(Shp_plots) |>
setView(lng = Lon_Saul, lat = Lat_Saul, zoom = 7.5) |>
addProviderTiles("Esri.WorldStreetMap") |>
addCircleMarkers(
lng = ~Lon, lat = ~Lat,
label = ~ as.character(PlotID),
fillColor = ~ pal_net(DB),
color = ~ pal_split(type),
fillOpacity = 0.8, opacity = 1,
weight = 2, radius = 4,
popup = ~ paste0(
"<strong>Inventory unit ID: </strong>", PlotID, "<br>",
"<strong>Network: </strong>", DB, "<br>",
"<strong>Sampling site: </strong>", Site, "<br>",
"<strong>Area (ha): </strong>", Area, "<br>",
"<strong>Protocole: </strong>", Protocole, "<br>",
"<strong>Reference census year: </strong>", RefCensus, "<br>",
"<strong>Train/validation: </strong>", type, "<br>"
)
) |>
clearControls() |>
addLegend(
position = "topright",
pal = pal_net, values = ~DB,
title = "Inventory network (inside)",
opacity = 0.8
) |>
addLegend(
position = "bottomright",
pal = pal_split, values = ~type,
title = "Train/validation (outline)",
opacity = 0.8
)if(!file.exists(here_rel("notebook", "figs", "phylo_GF.png"))){
g <- ggtree::ggtree(phylodata, layout = "fan", size = 0.5, aes(color = selected)) +
ggtree::geom_tiplab(size = 0.5, align = TRUE, linesize = .5) +
scale_colour_manual(values = c("0" = "grey75", "1" = "darkred", "NA" = "grey25"), limits = c("0","1"))+
labs(color = "Selected species") +
theme(legend.position = "bottom") +
theme_void()
ggsave(plot = g, filename = here_rel("notebook", "figs","phylo_GF.png"), bg = "white", width = 10,height = 10, dpi = 1500)
}
knitr::include_graphics(here_rel("notebook", "figs","phylo_GF.png"))
Estimating species’ environmental preferences requires a minimum number of occurrences across inventories and presence in distinct sites. We therefore excluded species with fewer than 30 total occurrences and presence in fewer than 10 distinct inventory units. Under these criteria, 569 species in 237 genera and 63 families were retained.
The final inventory dataset covers a cumulative area of 517 ha divided into 2,307 forest inventory units of various shapes and sizes. Each inventory unit consists of the measurement of the abundance of palm trees and trees ≥ 10 cm DBH (i.e. diameter at breast height, 1.30 m above the ground). To limit the taxonomic identification error, we selected the most recent available census for each inventory unit (reference census year).
A total of 1,981 inventory units had no missing data, representing 226,443 distinct individuals. The 175,675 individuals measured and identified to species span 88 families, 370 genera, and 1,732 species. We studied 569 species in 237 genera and 63 families, representing 83.9% of inventoried individuals and 31.9% of the described trees and palms species from the French Guiana checklist (Molino et al. 2022). The excluded individuals from retained inventory units are gathered in a unique class.
Plot_table |>
mutate(
Reference = glue::glue("<br><span style='font-size:50%'>({Reference})</span>"),
`Data reference` = glue::glue("<br><span style='font-size:50%'>({`Data reference`})</span>",
DOI = glue::glue("<br><span style='font-size:50%'>{DOI}</span>"))
) |>
gt() |>
cols_align(align = "left") |>
opts_theme() |>
opt_interactive(use_compact_mode = TRUE, use_highlight = TRUE, use_filters = TRUE) |>
opts_theme()