Data from: Beyond floodplain specialization: Ecological flexibility in the white bald-headed uakari (Cacajao calvus)
Data files
Nov 10, 2026 version files 10.77 MB
-
abc_plot_corners.csv
3 KB
-
analysis_outputs.zip
3.55 MB
-
behavioural_scans.csv
1.51 MB
-
climate_chirps_precipitation_daily.csv
9.98 KB
-
climate_era5_temperature_daily.csv
11.26 KB
-
climate_terraclimate_monthly.csv
322 B
-
corrections_log.csv
37.66 KB
-
DATA_DICTIONARY.md
16.23 KB
-
feeding_morphotypes_daily.csv
3.57 KB
-
feeding_visits.csv
163.91 KB
-
fruit_morphometry.csv
446.21 KB
-
habitat_classes_12_19m_30m.tif
24.23 KB
-
hand_earth_engine_point_extraction.csv
238.68 KB
-
HAND_local_30m.tif
196.77 KB
-
HAND_regional_30m.tif
3.62 MB
-
phenology_census_records.csv
298.71 KB
-
README.md
66.24 KB
-
relocations.csv
416.20 KB
-
river_level_jurua_altimetry_hydroweb.csv
10.28 KB
-
terra_firme_plots_inventory.csv
121.81 KB
-
trail_tracks_gps.kml
24.97 KB
Nov 18, 2026 version files 13.78 MB
-
abc_plot_corners.csv
3 KB
-
analysis_outputs.zip
6.52 MB
-
behavioural_scans.csv
1.50 MB
-
climate_chirps_precipitation_daily.csv
9.98 KB
-
climate_era5_temperature_daily.csv
11.26 KB
-
climate_terraclimate_monthly.csv
322 B
-
corrections_log.csv
93.43 KB
-
DATA_DICTIONARY.md
16.63 KB
-
feeding_morphotypes_daily.csv
3.57 KB
-
feeding_visits.csv
163.91 KB
-
fruit_morphometry.csv
446.21 KB
-
habitat_classes_12_19m_30m.tif
24.23 KB
-
hand_earth_engine_point_extraction.csv
238.68 KB
-
HAND_local_30m.tif
196.77 KB
-
HAND_regional_30m.tif
3.62 MB
-
phenology_census_records.csv
298.71 KB
-
README.md
70.01 KB
-
relocations.csv
399.30 KB
-
river_level_jurua_altimetry_hydroweb.csv
10.28 KB
-
terra_firme_plots_inventory.csv
121.81 KB
-
trail_tracks_gps.kml
24.97 KB
Abstract
Two groups of white bald-headed uakaris were followed from September 2022 to August 2023 at the Lake Surara Biological Reserve – Instituto Juruá, lower–middle Juruá River, Amazonas, Brazil: one with access to a continuous mosaic of seasonally flooded forest (várzea) and upland forest (terra firme), the other confined to an island of várzea. The dataset contains 2,359 georeferenced relocations recorded every 10 min, with subgroup size, activity, vertical stratum and the habitat recorded in the field; 17,910 behavioural scan records of individuals; 2,509 records of fruiting plants from monthly phenological censuses on four transects in both habitats, with measurements of 2,295 fruits from 960 plant records (size, mass and puncture resistance); 2,754 feeding visits and daily counts of food morphotypes; the stem inventory of three 1-ha terra firme plots; 30-m rasters of height above the nearest drainage (HAND), used to classify habitat along the flood gradient; daily climate series and the level of the Juruá River from satellite altimetry; and a log of every correction made to the field data. R scripts deposited with the data reproduce the habitat-selection, home-range, phenology and fruit-trait analyses of the article from these files. The data can be reused to study habitat use by floodplain primates, the phenology of várzea and terra firme forests, and the physical properties of the fruits available to seed predators.
Dataset DOI: 10.5061/dryad.tht76hff0
Description of the data and file structure
These data and code support the article “Beyond floodplain specialization: ecological flexibility in the white bald-headed uakari (Cacajao calvus)”.
Dias da Silva, R. H. P., Castro-Sá, M. J., Barnett, A. A., Peres, C. A., & Gordo, M. Ecology and Evolution, https://doi.org/10.1002/ece3.74684 (manuscript ECE-2026-04-01111).
Two groups of Cacajao calvus (Surara and Tucumã) were followed in the lower–middle Juruá River, Amazonas, Brazil, from September 2022 to August 2023. The deposit contains the georeferenced behavioural relocations, the behavioural scan records, height above the nearest drainage (HAND) rasters, phenological censuses and fruit measurements in terra firme and várzea, the terra firme plot inventory, feeding records, climate and river-level series, a log of every correction made to the field data, and the R code that reproduces the analyses of the article and its appendices (version 16 of the package, which accompanies the article as corrected at proof, November 2026).
Files in this deposit
The scripts expect the package folder layout given in the last column. To run them, create a folder, put each data file in the sub-folder shown, and unzip analysis_code.zip (and, if wanted, analysis_outputs.zip) into the same folder.
| file | content | folder in the package |
|---|---|---|
relocations.csv |
2,359 georeferenced behavioural relocations (one row per 10-min scan) | data/ |
behavioural_scans.csv |
17,910 scan records of individual uakaris | data/ |
phenology_census_records.csv |
2,509 records of fruiting plants on the phenological transects | data/ |
fruit_morphometry.csv |
3,834 rows, one per fruit (2,493 plant records; 2,295 rows carry at least one measurement) | data/ |
feeding_visits.csv |
1,630 feeding records (2,754 visits) | data/ |
feeding_morphotypes_daily.csv |
daily counts of food morphotypes (56 days) | data/ |
terra_firme_plots_inventory.csv |
1,815 stems of the three 1-ha terra firme plots | data/ |
abc_plot_corners.csv |
centre-line waypoints and corners of the three plots | data/ |
hand_earth_engine_point_extraction.csv |
earlier HAND extraction in Earth Engine (provenance only) | data/ |
climate_chirps_precipitation_daily.csv |
CHIRPS daily precipitation | data/ |
climate_era5_temperature_daily.csv |
ERA5 daily mean air temperature | data/ |
climate_terraclimate_monthly.csv |
TerraClimate monthly minimum and maximum temperature | data/ |
river_level_jurua_altimetry_hydroweb.csv |
Juruá River level from satellite altimetry, 2017–2025 | data/ |
corrections_log.csv |
every value changed in the field data, with the reason | data/ |
HAND_local_30m.tif |
HAND (m), 30 m, study area: the source of every HAND value in the article | data/hand_raster/ |
HAND_regional_30m.tif |
HAND (m), 30 m, study area plus 10 km (Appendix S1.5) | data/hand_raster/ |
habitat_classes_12_19m_30m.tif |
three-class raster cut at 12 and 19 m (Figure 8) | data/hand_raster/ |
trail_tracks_gps.kml |
GPS tracks of the phenological transects and trails (Figure 1d) | field_gis/ |
DATA_DICTIONARY.md |
description of every column (also reproduced below) | package root |
analysis_code.zip |
code/ (R scripts 00–11, run_all.R, figures/), gee/export_HAND_rasters.js, the package README and the MIT licence; published at Zenodo |
package root |
analysis_outputs.zip |
output/: tables (CSV), logs of the last complete run, Figure 4 and 300-dpi previews of the figures drawn by figures_600dpi.R and fig6_fig8.R (the 600-dpi TIFFs are not included; the scripts redraw them) |
package root |
Version 16 (November 2026). This version accompanies the corrected proof of the accepted article. While checking the proofs, the authors found that 100 rows of relocations.csv were
exact copies of other rows. All were Surara relocations: 86 of 22 September 2022, and 14 on six other
days. They were created in May 2025, when the field census was joined to the GPS waypoints: the GPS
export held the same waypoint 3–5 times under different names, with the same coordinates and second,
and each copy produced one more row. They are not in the field datasheets. They were found together
with three other problems. Version 16 therefore:
- removes the 100 copies (2,459 → 2,359 relocations);
- adds 21 subgroup counts that were missing and that the authors recovered from the field datasheets;
- removes 46 rows of
behavioural_scans.csvthat were entered twice (17,956 → 17,910); - corrects the year of 35 rows of
feeding_visits.csvtyped as December 2023 instead of December 2022
(feeding days 111 → 107).
Each of these 202 changes is a row of data/corrections_log.csv. Every script was re-run on the
corrected files (all logs, tables and figures in output/). The same code on the version-15 files
reproduces the version-15 logs and figures exactly, except the ctmm fit of one rejected candidate
model (OUf anisotropic), which converges differently between runs (Section 5.3). Scripts that do not
read the corrected files (08, 08b, 10, 10b and 11) give identical output.
Figures 6 and 8 are now also drawn in R (code/figures/fig6_fig8.R, same design as the Python scripts,
which need packages not available here). At proof, Figure 6 of the article was also replaced by the version drawn from the corrected files. Compared with the figure submitted with the manuscript, only two Surara outlines move: those
of the terrestrial phase and of the dry season, the periods that contain 22 September 2022. They move by up
to about 120 m at one point. The areas in Table 5 are those of the corrected files.
hand_earth_engine_point_extraction.csv is kept as exported.
It is an earlier extraction that no script reads, and it still contains the duplicated relocations.
Supplementary material is cited with the numbering of the article's Supporting Information (Appendices S1–S4, Tables S1–S24, Figures S1–S3). Version 15 of the package had changed only these labels (in this README, the data dictionary, the comments and printed headings of the scripts, and the logs, which were edited in the same way and not re-run); no data file, value or calculation changed. The file names in analysis_outputs.zip keep the earlier appendix numbering and correspond to the Supporting Information as follows:
file in output/ |
Supporting Information |
|---|---|
appendixA1_1_bandwidth.csv |
Table S1 |
appendixA1_2_bandwidth_sensitivity.csv |
Table S2 |
appendixA1_3_ctmm_selection.csv |
Table S3 |
appendixA1_5_domain.csv |
Table S5 |
appendixA1_6_positional.csv |
Table S6 |
appendixA1_7_hand_bands.csv |
Appendix S1.7 (HAND bands, availability against Surara use) |
appendixA1_8_thresholds.csv |
Table S7 |
appendixA2_2_events_effort.csv |
Table S9 |
appendixA2_5_index_per_census_field_only.csv |
Appendix S2.5 (field-measurement check of Table S12) |
appendixA2_6_plot_hand_profiles.csv |
Table S13 |
appendixA3_2_resampling.csv |
Table S15 |
appendixA3_3_kernel_composition.csv |
Table S16 |
appendixA3_7_trait_correlations.csv |
Table S20 |
appendixA3_figure1_hourly_activity.csv |
Figure S2 (data) |
appendixA4_1_monthly_index.csv |
Table S21 |
appendixA4_2_strata_by_month.csv |
Table S22 |
appendixA4_3_flood_amplitude.csv |
Table S23 |
figures/FigureA3_1_300dpi.png (the 600-dpi figures/FigureA3_1.tif is redrawn by figures_600dpi.R) |
Figure S2 |
Files and variables
1 | data/relocations.csv — 2,359 rows (2,459 up to v15; 100 duplicated rows removed in v16, README §3.13)
One row per georeferenced behavioural record (relocation), taken every 10 min while one or more
group members were in view. Coordinates are the observer's position at the moment of the scan.
HAND is not stored here: it is extracted from data/hand_raster/HAND_local_30m.tif by
code/01_analytical_base.R.
| column | type / values | description |
|---|---|---|
date |
date | date of the record |
time |
hh:mm:ss | local time (America/Manaus); missing in 1 row |
group |
Surara / Tucuma |
group being followed (2,165 / 194) |
group_label_corrected |
TRUE / FALSE |
TRUE for the three records relabelled from Tucumã to Surara (README §3.1) |
census_month |
month name | calendar month of date |
rainfall_season |
dry / rainy |
derived from the calendar month of date, as defined above |
hydrological_phase |
terrestrial / aquatic |
derived from the calendar month of date, as defined above (one May record had been labelled terrestrial; corrections log) |
species |
text | always Cacajao calvus |
group_size |
integer | individuals counted in the subgroup; recorded for every row since v16 (21 counts recovered from the field datasheets; README §3.7) |
activity |
foraging · locomotion · resting · social · agonistic |
main activity at the scan (Section 2.3 of the article) |
perpendicular_distance_m |
m | observer–animal offset to the nearest visible group member; recorded for 90.7% of rows |
height_above_ground_m |
m | height of the animal above ground (81.0% of rows) |
stratum |
dominated · dominant · emergent |
vertical stratum occupied (Brazilian National Forest Inventory classes) |
position_in_tree_code |
cp · rt · li · pi and combinations |
position in the plant as coded in the field: cp crown (copa), rt branch, li liana; pi is not defined in the surviving field key |
habitat_field |
varzea / terra_firme |
habitat recorded by the observer at the relocation |
weather |
sunny · cloudy · rain |
weather at the scan |
temperature_c |
°C | air temperature as recorded; two values below 15 °C (minimum 1.3 °C) are probably entry errors and are not used in any analysis |
waypoint |
text | GPS waypoint label (numbering restarts; not unique) |
latitude, longitude |
decimal degrees | observer position |
2 | data/phenology_census_records.csv — 2,509 rows
One row per record of a fruiting plant on the four phenological transects (two in várzea, two in
terra firme; 0.4 ha per habitat). Used for Table 1A, Figure 4, Table S22 and the ground-count
censoring rates (Table S10).
| column | type / values | description |
|---|---|---|
date |
date | date of the record (two 2023 year typos corrected, README §3.3; three records undated) |
census_month |
month name | monthly census (expedition) in which the record was made — the month used in the article |
rainfall_season, hydrological_phase |
as above | attached to the census month |
morphospecies |
text | field name of the plant (species where known; ni not identified) |
family |
text | family, where known |
life_form |
tree · liana · shrub · palm · others |
growth form as recorded; a few rows carry field shorthand (plt, vz_bx, vz_bx_ch, ni, herbacea) |
fruit_maturity |
immature · mature · both · flower |
maturation stage of the fruits |
fournier_score |
0–4 | Fournier (1974) crown intensity score |
habitat |
varzea / terra_firme |
transect habitat |
dbh_cm |
cm | diameter at breast height (for lianas, the host tree) |
height_m |
m | total height |
stratum |
dominated · dominant · emergent · undefined |
sociological position (National Forest Inventory) |
fruits_on_ground |
count | fruits counted on the ground within the 0.5-m transect strip; left-censored during the aquatic phase in várzea (Section 2.4.1) |
3 | data/fruit_morphometry.csv — 3,834 rows
The same censuses at the level of the measured fruit, with a plant identifier. This is the base
for Table 2 (A and B), Table 3 and Tables S9, S17, S19, S20, S21 and S24.
Row structure. One row per fruit measured (1–8 fruits per plant record). plant_id identifies a
plant within a monthly census and is never reused across months (2,493 identifiers), so counting
distinct plant_id per month and habitat gives the number of fruiting events. Plant-level columns
are filled on the first row of each plant and are empty on the other rows, except date,
census_month, plant_id and habitat, which are filled on every row; the habitat of every row is
that of the plant's first row.
Block of 11 August 2023. In the primary spreadsheet (Phenology_VF_2022_2023.xlsx), the
continuation rows of nine plants recorded that day carried the family, DBH, Fournier score,
habitat, stratum and ground count of other plants recorded the same day, and the header rows of six
of them repeat another plant's values (a paste offset). The continuation rows were cleared of these
values (the fruit measurements on them belong to the first plant and are kept), and the six plants
whose header rows are duplicates are flagged in plant_fields_flag and excluded from every analysis
that uses the Fournier score or the habitat of the plant. Their true values can only be recovered
from the field notebook.
| column | scope | description |
|---|---|---|
date |
plant | date of the record (carried to all fruits of the plant; three plants undated) |
census_month |
plant | monthly census in which the plant was recorded |
plant_id |
plant | plant identifier within the census |
morphospecies, family |
plant | as in §2 |
life_form |
plant | as in §2 |
fruit_maturity |
plant | as in §2 |
fournier_score |
plant | as in §2 |
habitat |
plant | varzea / terra_firme (habitat of the plant's first row) |
dbh_cm, height_m, stratum, fruits_on_ground |
plant | as in §2 |
fruit_colour, pericarp_consistency, fruit_type |
plant | field descriptors, partly translated (green, red, dry, fleshy, berry, capsule, pod, …); rare values are kept as written in Portuguese. Not used in any analysis |
length_mm, width_mm, thickness_mm |
fruit | dimensions, vernier calipers ±0.1 mm. Thickness was recorded only where it was a distinctive dimension of the fruit (Section 2.4.2.2); 11 of 410 readings outside 0.5–40 mm are recording errors and are removed by 08_fruit_traits.R |
mass_g |
fruit | fresh mass, portable scale ±0.01 g (zeros are readings below the instrument's resolution) |
puncture_resistance_lb |
fruit | pericarp puncture resistance (lb), prosthetic uakari canine on a Dial Geotester pocket penetrometer (Barnett et al., 2016); ceiling 50 lb |
pericarp_wall_mm |
fruit | pericarp wall thickness (16 plants) |
water_depth_m |
plant | water depth at the plant during canoe censuses in várzea (April–June 2023) |
plant_row |
fruit | 1 for the first row of each plant, 2, 3, … for the other fruits of the same plant |
plant_fields_flag |
plant | duplicated_from_other_plant for the six plants of 11 August 2023 described above; empty otherwise |
Aggregation rule of the article (Section 2.7): the plant is the unit; each variable is the mean of
the fruits on which it was recorded, independently of the others; the size index is the product of
the three mean dimensions (cm³); the immature block comprises plants bearing immature fruit, with all
their measurements.
4 | data/terra_firme_plots_inventory.csv — 1,815 rows
All stems with DBH ≥ 10 cm in three 1-ha permanent plots (500 m × 20 m; TRR A, TRR B, TRR C) censused
by ABC Expeditions.
| column | description |
|---|---|
plot |
TRR_A, TRR_B, TRR_C |
stem_id |
tag number |
taxon_field |
name recorded in the field |
family_code |
abbreviated family name as recorded (e.g. myris Myristicaceae, lecyt Lecythidaceae) |
dbh_cm |
diameter at breast height, cm; three stems below 10 cm or without DBH are excluded from densities |
life_form |
tree / liana |
accepted_name_reflora |
accepted name in the Flora e Funga do Brasil (REFLORA), where resolved |
5 | data/feeding_visits.csv — 1,630 rows
Feeding records made during focal follows: one row per plant (or item) fed on per day.
| column | description |
|---|---|
date |
date of the follow (35 rows typed as 2–5 December 2023 were corrected to 2022 in v16; README §3.15) |
feeding_visits |
number of feeding visits to that plant that day |
morphospecies, family |
plant fed on, where identified |
life_form |
tree, liana, palm, shrub, or field shorthand (epifita epiphyte, casca bark, tronco trunk) |
item_eaten |
part eaten, field code: seed, im immature, md mature, s seed, polpa pulp, flor flower, folha leaf, insetos insects, casca bark, and combinations |
maturity_code |
field code for fruit maturity: im immature, md mature, and combinations (im_md both); other values are item codes entered in this column |
fruit_type_code |
baga berry, capsula capsule, drupa drupe, vagem pod, pixidio pyxidium, sinconio syconium, apocarpico apocarpous, sincarpico syncarpous, … |
habitat_code |
habitat of the plant, field code: tf terra firme (plt plateau, vrt slope); vz/vr várzea (alta high, bx/baixa low, rst restinga, ch/chv chavascal), baixio low-lying ground, lago lake, beira margin; codes combining tf and vz are left unassigned in 09_feeding.R |
latex, colour, pericarp_surface, consistency |
field descriptors in Portuguese, as recorded; not used in any analysis |
6 | data/feeding_morphotypes_daily.csv — 56 rows
Photographic sorting of food items into morphotypes at the end of each observation day
(Section 2.4.1). Morphotypes are defined within days and are not tracked across days.
| column | description |
|---|---|
date, month |
observation day |
immature, immature_and_mature, mature |
fruit morphotypes by maturity |
flower, bark, leaf, invertebrate, play_item, vine |
other food morphotypes (play_item: object handled in play) |
morphotypes_total |
total morphotypes that day (row sum) |
habitat |
Terra Firme, Várzea alta (high várzea), Várzea baixa (low várzea), or Terra Firme e Várzea Alta (both) |
6b | data/behavioural_scans.csv — 17,910 rows (new in v7; 46 rows entered twice removed in v16, README §3.14)
The behavioural scan records of Section 2.3, deposited verbatim as exported from the field
spreadsheet (semicolon-separated, UTF-8, Portuguese codes). One row per individual per sampling
unit. Scans were made in repeated 3-min cycles following Barnett (2010): each cycle began with an instantaneous scan of
the nearest or visible individuals, in sequence, followed by 1 min of ad libitum observation and a 5-s interval
before the next cycle (Section 2.3 of the article as corrected at proof; versions 11–15 of this package described
two 60-s scans per cycle). hours is the time noted for the record, to the minute (to the second on five
days); successive times are often less than 3 min apart, so distinct date + hours values are not a
count of cycles (05b_scan_sampling.R estimates the number of cycles). Only uakari records are deposited
(the 884 records of other animals in the original export were removed at the authors' request).
| column | values | description |
|---|---|---|
date |
dd/mm/yyyy | day of observation |
hours |
hh:mm:ss | time noted for the record |
specie |
ukr_1, ukr_2, … |
uakari scanned within the unit (numbered within the unit) |
age |
ad adult, jv juvenile, vlh old, kid/flht infant, - not recorded |
age class |
sex |
m, f, - |
sex |
arboreal_stratum |
dmd dominated, dmt dominant, emg emergent, subosque understorey, aquatico in water, - |
vertical stratum |
place_in_the_tree |
cp, rt, li, pi, combinations, agua water, solo ground, - |
position in the plant, as in position_in_tree_code of §1 |
general_behavior |
forrageando foraging, locomovendo locomotion, repousando resting, sociais_e_reprodutivos social, agonistico agonistic, vocalizando vocalizing, outros other |
macro-category used in Figure S2 |
behavioral_category_follow_Bezerra_etal_2014, behavioral_subcategory |
Portuguese field codes | finer ethogram of the companion study; not used in this article |
habitat |
vz_bx low várzea, vz_alta high várzea, ecotono_vz_tf várzea–terra firme ecotone, tf terra firme |
habitat recorded in the field |
7 | Climate and river level
| file | columns | description |
|---|---|---|
climate_chirps_precipitation_daily.csv |
date, precip_mm |
CHIRPS daily precipitation (mm) at the study site, 1 Sep 2022 – 30 Aug 2023 |
climate_era5_temperature_daily.csv |
date, temp_C |
ERA5 daily mean 2 m air temperature (°C), same period |
climate_terraclimate_monthly.csv |
month, tmin_c, tmax_c |
TerraClimate monthly mean minimum and maximum temperature (°C) for the study area, September 2022 – August 2023, exported from Google Earth Engine |
river_level_jurua_altimetry_hydroweb.csv |
provider columns, semicolon-separated | satellite altimetry of the Juruá water surface at the virtual station adjacent to the study site, January 2017 – March 2025 (112 passes); HEIGHT (M) OF WATER SURFACE is the variable used |
8 | Rasters and spatial files
| file | description |
|---|---|
data/hand_raster/HAND_local_30m.tif |
height above the nearest drainage (m), 30 m, EPSG:4326, 22,203 ha around the study area; export of users/gena/GlobalHAND/30m/hand-1000 (Donchyts et al., 2016) |
data/hand_raster/HAND_regional_30m.tif |
the same product over the study area plus 10 km (Appendix S1.5) |
data/hand_raster/habitat_classes_12_19m_30m.tif |
the local surface cut at 12 and 19 m: 1 low várzea (0–12 m), 2 transition (>12–19 m), 3 terra firme (>19 m); used for mapping (Figure 8). Analyses cut the continuous raster instead |
data/hand_earth_engine_point_extraction.csv |
earlier extraction of HAND inside Earth Engine at 2,458 relocation coordinates (columns: date, longitude, latitude, hand_m_earth_engine, hand_class_earth_engine, activity, group_size, habitat_field, rainfall_season); provenance for Section 2.6, not used in analyses; made before the duplicated relocations were removed, and still contains them (README §3.13) |
gee/export_HAND_rasters.js |
Google Earth Engine script that exported the rasters |
field_gis/trail_tracks_gps.kml |
GPS tracks of the phenological transects and trails |
data/abc_plot_corners.csv |
waypoints of the three 20 m × 500 m terra firme plots of ABC Expeditions (TRR A, B, C): plot, point_code (000C–500C centre line every 100 m; 000E/000D and 500E/500D left and right corners at the start and end), position, distance_m along the axis, longitude, latitude (WGS 84, decimal degrees; corners recorded at lower precision), source. Used by 10b_plot_positions.R (Appendix S2.8) |
9 | data/corrections_log.csv
One row per value changed in the deposited field data: file, record (waypoint and time, or
plant identifier and row), column, old_value, new_value and reason.
Code/software
2 | How to run
R (tested with R 4.5.1 on Windows 11): terra, sf, sp, adehabitatHR, ctmm, lme4,
MASS, ggplot2. From the package root:
Rscript code/run_all.R
Rscript code/figures/figures_600dpi.R
Rscript code/figures/fig6_fig8.R
Each script can also be run on its own after 01_analytical_base.R, which writes
output/analytical_base.rds and output/availability.rds. Printed results go to
output/logs/<script>.log; tables go to output/*.csv. A full run takes about 25 minutes, most of it
in the two continuous-time scripts (03 and 03b).
Bootstrap intervals (Section 2.7 of the article) resample sampling days with replacement, 2,000
replicates, with the seed reset to 20260916 before every interval; the same seed is set before every
ctmm model selection (Section 5.3). Spearman correlations use R's
default P value: exact for the 12-month series, t approximation for larger samples.
Figures. figures_600dpi.R (R packages ragg, terra, adehabitatHR) draws Figures 3, 5, 7
and 9 and Figure S2 as 600-dpi TIFF (LZW, 170 mm wide) in output/figures/. It follows
the Python scripts that drew them for the submission (same panels, colours, 2-m bins, reference
bandwidth and 500 × 500 grid), which could not be re-run because no Python installation was
available; Figure S2 uses uakari records only (the original export also held 884 records of other
animals, which are not deposited; with them the hourly indices differ by at most 0.036). Up to v15,
Figures 6 and 8 were drawn by their Python scripts (numpy, pandas, rasterio, pyproj, scipy,
matplotlib, Pillow, FreeSans font; e.g. python code/figures/fig6.py). Since v16 they are drawn
in R by fig6_fig8.R, with the same design: four panels of 95% kernel isopleths by hydrological phase
and by rainfall season over the HAND ramp, with h = mean SD × n^(−1/6), a 500 × 500 grid and extent 1;
and bars of relocations per HAND class against availability in the pooled 100% MCP, with the ridge
relocation of Section 3, item 4, in low várzea. The figure's contours use the figure-script
bandwidth, so their areas differ slightly from Table 5, which uses adehabitatHR's reference
bandwidth; the log prints both.
3 | Data processing
Every change is listed in data/corrections_log.csv.
- Hydrological phase of one relocation (new in v5). Relocation
ukr_mai_165(18 May 2023,
11:39) was labelled terrestrial, although May belongs to the aquatic phase by the calendar rule
of the article and another relocation of the same day was labelled aquatic. Phase and rainfall
season are now derived from the date for every relocation. Effect: Surara phase subsets
1,418 / 847 (were 1,417 / 848); areas unchanged; terra firme Wᵢ in the terrestrial phase
2.38 (1.64–2.99); phase ratio 1.39 (0.91–1.93); UDOI between phases (95%) 0.73; daily median HAND
by phase 15.5 m (70 days) against 18.8 m (42 days), P = 0.046 (was 17.0 against 17.5 m, P = 0.07,
with the 18 May day counted in both phases). - Plant-level fields of 11 August 2023 (new in v5). In the primary spreadsheet
(Phenology_VF_2022_2023.xlsx, rows 1382–1403), the continuation rows of plants 1114–1122
(identifiers 1027–1035 here) carry the family, DBH, Fournier score, habitat, stratum and ground
count of other plants recorded that day (plants 1125–1140), and the header rows of 1117–1122
repeat those plants' values. The fruit measurements on these rows belong to the first plant. The
continuation rows were cleared of plant-level values, the habitat of every row was set to that of
the plant's first row (this also corrects plant 1053), and the six plants whose header rows are
duplicates are flagged (plant_fields_flag) and excluded from the analyses that use the Fournier
score or the habitat of the plant. The notebook should be checked for their true values.
A revised copy of the spreadsheet received from the authors on 26 September 2026 gives the twelve
continuation rows a date, the nameniand no identifier, which would make them twelve separate
plants. That change was not adopted: the fruits on these rows match the first plant's fruit in
size and shape (e.g. pods of 117 and 110 mm with the 130 mm pod of plant 1035; fruits of 127 and
132 g with the 112 g fruit of plant 1033), and seven of the twelve rows repeat the plant-level
fields of plants recorded later that day (1037, 1040, 1041, 1045, 1048, 1052, 1053). - Group labels. Three relocations (
ukr_mrc_135,ukr_mrc_138, 23 March 2023;ukr_abr_63,
21 April 2023) were labelled Tucumã in the field sheet while falling inside continuous Surara
follow sequences. They are assigned to Surara and flagged ingroup_label_corrected. - Ridge relocation. One Tucumã relocation (waypoint 83, 5 December 2022) has HAND 23 m in the
local raster, 7 m in the earlier Earth Engine point extraction and 7–16 m at the seven relocations
within 150 m, and was recorded as várzea in the field.01_analytical_base.Rassigns it to low
várzea for categorical analyses and keeps its raster value in continuous HAND analyses. - Dates. Two phenological census records and two fruit-morphometry plant records dated
11 November 2023 were corrected to 2022. In the fruit file the date is carried from the first row
of each plant to its other fruits (three plants have no date). - Census month. Phenological records carry the monthly census in which they were made. The eight
terra firme records dated 12 December 2022, labelled November in the field sheets, belong to the
December census (authors, 26 September 2026) and are assigned to it in both phenology files (new
in v7). - Group size. Up to v15 it was missing for 20 Surara relocations (one entered as "several
together", 19 blank) and 1 Tucumã relocation. The authors recovered all 21 counts from the field
datasheets, and v16 enters them, so every relocation now has a group size. - Plot inventory. Three stems below 10 cm DBH (or without DBH) remain in the file and are
excluded by10_plots_density.R(1,812 stems analysed). - Codes and names. Portuguese codes were translated; free-text field descriptors that are not
used in any analysis are kept as recorded. Field names given to unidentified plants
(morphospecies, e.g.bundinha,catore_falso,uxirana_amarela) were kept stable across
censuses by the field team and are used by08b_individuals_across_months.R. - Fruits per plant. Twenty plants have four to eight measured fruits. Each plant's trait value
is the mean of all its fruits, whatever their number (08_fruit_traits.R). - Water depth (new in v7). Four depths recorded with a question mark were confirmed by the
authors (plants 1588, 1613, 1659 and 1831: 1.9, 3.4, 2.4 and 0.5 m); three others
(1586, 1594, 1789) remain uncertain and are not used. Appendix S4.4: 130 plants measured in May and 68 in June. - Behavioural scans (new in v7; uakari records only since v8). Deposited line by line as
exported, keeping the 17,956 uakari records; the 884 records of other animals seen with or near the
groups were removed at the authors' request. Scans were made in repeated 3-min cycles, each with one instantaneous scan followed by
1 min of ad libitum observation and a 5-s interval (Barnett 2010; Section 2.3 of the article as corrected at
proof). Versions 9 and 10 described 90-s and 2-min units, and versions 11–15 two 60-s scans per cycle; the
authors corrected both. - Duplicated relocations (new in v16). One hundred rows of
relocations.csvrepeated another
row in every column. All are Surara: 86 on 22 September 2022, where 31 relocations had become
117 rows; 7 on 22 February 2023; 1 each on 23 and 27 February and 23 May; 2 each on 3 March and
14 August 2023.- Origin. They are absent from the field census sheet and appear in an intermediate file of
May 2025, when the census was joined to the GPS waypoints. On those days the GPS export holds
the same waypoint 3–5 times under different names, with the same coordinates and second (the
GPS clock is UTC; local time is UTC−4), and the join repeated the record once per copy. - Treatment. The first row of each set is kept. The other 57 date–time keys that still occur
twice (56 Surara, 1 Tucumã) differ in coordinates, subgroup, activity or secondary fields. They
are kept as recorded;03_akde_ctmm.Rkeeps the first record of each timestamp, as before.
- Origin. They are absent from the field census sheet and appear in an intermediate file of
- Scans entered twice (new in v16). Forty-six rows of
behavioural_scans.csvrepeated another
row in every column. In 43 of them, the same record already appears twice in the field
spreadsheet. They are removed (17,910 rows). - Feeding dates (new in v16). Thirty-five rows of
feeding_visits.csvdated 2–5 December 2023,
after the end of the study, were corrected to 2–5 December 2022. Those days already held relocations, scans and
34 other feeding rows. The re-dated rows are not copies of those rows (no row matches in every
other field). Feeding days: 107 (were 111).
4 | Two HAND sources
All HAND values in the article come from HAND_local_30m.tif, an export of the global 30 m HAND
product of Donchyts et al. (2016) (users/gena/GlobalHAND/30m/hand-1000 in Google Earth Engine),
sampled at the relocation coordinates in R. HAND_regional_30m.tif is the same product over the
study area plus 10 km, used only for the buffered availability domains of Appendix S1.5 (within the
pooled 100% MCP both give a terra firme availability of 0.24). hand_earth_engine_point_extraction.csv
is an earlier extraction inside Earth Engine, deposited because Section 2.6 refers to it. It was made
before the duplicated relocations were removed (Section 3, item 13) and still contains them; it has no
time column and no script reads it.
5 | Reproduction of the reported values (article as corrected at proof, November 2026)
Every script prints the values it computes next to the section of the article in which they appear.
The logs of the last complete run (v16, on the corrected files) are in output/logs/. The values of
the article, its tables, Figures 3 and 5–9 and the Supporting Information were updated at proof from
these logs.
5.1 | Reproduced
Except for the items of Section 5.2: all values of Sections 2.1, 2.3, 2.6, 3.2, 3.3.1, 3.3.2, 3.3.3, 3.4,
3.5 and 3.5.1–3.5.2; Tables 1–6; Tables S1, S2, S4–S7, S9, S13, S14, S15, S16 (since v16), S19, S20,
S21, S22 and S23; the AKDE and its model selection (Table S3; see Section 5.3 for the OUf anisotropic
row); the stem
densities; the rainfall-season means, TerraClimate temperatures and river-level statistics of
Section 3.1 (2,458 mm and 26.1 °C, from the deposited CHIRPS and ERA5 series, which the authors
confirmed as the extraction over the pooled 100% MCP); the size of the scan dataset in Section 2.3 and
Figure S2; the share of relocations within 15 m of another HAND class (26.7%, Section 2.3). The monthly Spearman
P values of Section 3.4 (0.011 and 0.035) are exact tests; an earlier version of this README wrongly
listed them as not reproduced, because the script then used the t approximation (0.008 and 0.031).
5.2 | Not reproduced from these files
| Article | Reported | This package | Note |
|---|---|---|---|
| Table S10 | zeros 3.4% and 2.4% (terrestrial phase) | 3.3%, 2.3% | |
| S2.4, S2.5, S3.4, S3.5 | mass per event (e.g. terrestrial phase P = 0.65) | P = 0.56 from field measurements only | the 2,218-event dataset with the volumetric mass prediction (Section 2.4.2.3) is not deposited. Table S12 is now per census: each value is the earlier value per census date multiplied by the number of census dates of that month and habitat (Table S9); appendixA2_5_index_per_census_field_only.csv gives the same index from field measurements only, as a check of its order of magnitude. The terra firme values for November and December were rescaled in v34 by the change in that field-only index after the 12 December 2022 records were assigned to December |
| S4.4, April | 130 trees measured | 128 | |
5.3 | Changes of method and sensitivity analyses (v5 to v7)
-
Fruiting index per census (v6; census described in v12). Each transect was censused once per
month, over at most five days per expedition (up to three in one habitat and two in the other);
plants that began fruiting after the census and were found on the same trails later in the
expedition were added to that month's census, so the records of a census span 2 to 9 dates. Within
a census almost no tree was recorded on two dates (5 of 2,002 by trail, morphospecies and DBH).
Because the census days are not flagged in the records,07_fruiting_intensity.Rapproximates
them by the three dates with most records in one habitat and the two in the other and repeats the
comparisons of Table 2 on those records (1,665 of 2,050 plant records, 81%): every comparison keeps
its direction and conclusion (Appendix S4.1). The number of census dates is therefore not a measure of
effort, and the index (sum of Fournier scores ÷ 4 ÷ 0.4 ha) is now expressed per census
(appendixA4_1_monthly_index.csv, columnindex). The version divided by census dates, used up to
v5, is kept for comparison (columnindex_per_day); the conclusions are the same with either
(Appendix S4.1). The chain of Appendix S2 follows: events per census are no longer divided by the
number of dates (appendixA2_2_events_effort.csv), and E/N is a nominal event-to-stem ratio, not
a proportion of stems in fruit (it exceeds 1 in terra firme in December: 1.02). -
Observation time (v11) (
05b_scan_sampling.R). Scans were made in repeated 3-min cycles (one instantaneous scan, then 1 min of ad libitum
observation and a 5-s interval; Section 2.3), and records were made only during the cycles.
Each row of the file is one individual, with the time noted to the minute. Successive recorded times
are exactly 3 min apart in 36.6% of cases, as expected for consecutive cycles, but less than 3 min
apart in 30.2%, so the 5,791 distinct recorded times are not a count of cycles. The smallest number
of 3-min cycles that contains every recorded time (a greedy cover within each day, which is optimal
for intervals of fixed length) is 4,587: at least 229.4 h of observation. The log of05b_scan_sampling.Ralso splits this time into 152.9 h
of scans and 76.5 h of ad libitum recording; that split follows the earlier description of two 60-s scans per
cycle, kept in the script's comments, and is not used in the article.
For reference, taking every recorded time as the start of a cycle and counting overlapping time once
gives 243.8 h; 5,791 × 3 min = 289.6 h counts overlapping cycles twice; and counting each row as a
2-min unit (598.5 h; 597.0 h after the duplicate rows were removed in v16, which change no recorded
time and no cycle) counts the same scan once per individual. (Versions 9 and 10 used 90-s and 2-min
units and gave 144.8 h and 175.5 h.) The 373.2 h of the submitted version
(11,196 units × 2 min, on 134 days) cannot be derived from the scan file: no count in it gives 11,196
(distinct times 5,791; cycles 4,587; rows 17,956, now 17,910), and 134 is the number of relocation
days (scan days: 114). -
Feeding values (v10) (
09_feeding.R). Up to manuscript version 36 the article reported values
computed from a cleaned copy of the feeding sheet that is not deposited (2,642 visits on 107 days;
920 / 1,266 visits in terra firme / várzea; 676 items on 55 days; IRR 0.89, 0.70–1.11, P = 0.30,
n = 42). It now reports the values computed from the deposited files: 2,754 visits on 111 days;
927 / 1,361; 682 items on 56 days, 51 of which have feeding-visit records; IRR 0.89 (0.70–1.12),
P = 0.32, n = 41. The conclusion is the same. Since v16, which corrects the year of 35 records
(Section 3, item 15), the article reports:- 2,754 visits on 107 days, the várzea visits on 83 days;
- richness against visits ρ = 0.78, P < 0.001;
- β = 0.66 (0.50–0.83);
- IRR 0.97 (0.75–1.27), P = 0.84, n = 41.
The conclusion is again the same.
-
Repeated individuals (v6) (
08b_individuals_across_months.R). Records with the same field name,
in the same habitat and in different months, with DBH within 0.5 cm or 3%, are grouped as one
candidate individual; groups that contain two records from the same month are ambiguous and left
unlinked. Of 100 candidate groups (305 plant records), 37 are ambiguous, leaving 63 individuals
recorded in two to four censuses (139 records);sensitivity_individual_links.csvgives the same
counts for tolerances from identical DBH to 1 cm or 5%. With the linked individual as the unit, the
Table 3 comparisons keep their direction and significance (sensitivity_table3_individuals.csv). -
Duplicated timestamps (
03b_akde_sensitivity.R). One Surara record has no recorded time, and 114 share one of 56 timestamps
(coordinates differ in 15); keeping the first record of each timestamp removes 58 of them. Before the 100 copies were removed
in v16, 256 records shared one of 98 timestamps and 158 were removed; the series fitted by ctmm (2,106 fixes) is the same in
both versions. Averaging their coordinates instead of keeping
the first record leaves the selected model, the area estimate and its effective sample size
unchanged (OU anisotropic; 8.70 km²; DOF 152.8 against 152.7). -
Tucumã and ctmm (
03b_akde_sensitivity.R). Earlier versions of the article stated that 194 relocations
do not support ctmm model selection. They do: an anisotropic OU process is selected by 495.8 AICc units
over the independent model, with an effective sample size for area of 37.9 and an AKDE of 654.9 ha
(463.2–879.3). The estimate is imprecise and is reported in Appendix S1.3 only. -
Random seed of the ctmm model selection (v6) (
03_akde_ctmm.R,03b_akde_sensitivity.R).
ctmm.selectdraws random numbers while fitting the candidate models, and up to v5 no seed was set.
The OUf anisotropic candidate is sensitive to this: in some runs its fit converged to a degenerate
solution with the effective sample size of the independent model (ΔAICc 11,646.19; DOF 2,105), as
listed in Section 5.2 of the v5 README. With the seed set to the package's value (20260916, the
same one used for every bootstrap) the fit is reproducible, and it gives the values of Table S3
(8,059.55; 1,016.76; 8,056.60 and 1,017.10 when duplicated timestamps are averaged). Two runs with
the same seed gave identical tables. The selected model and every estimate derived from it were the
same in every run, with or without a seed. In the November 2026 runs, the seed did not fix this
candidate.03_akde_ctmm.Rgave the degenerate solution (11,646.19; DOF 2,105) both on the v15 files
and on the corrected v16 files.03b_akde_sensitivity.Rgave it on the v15 files but not on the v16
files. Table S3 keeps the values above. The candidate is never selected; every other row, the
selected model and the AKDE (696.9 ha, 590.8–811.7; DOF 152.7) are identical in all runs.
Access information
Other publicly accessible locations of the source data:
- HAND: global 30 m height above the nearest drainage (Donchyts et al., 2016), Google Earth Engine asset
users/gena/GlobalHAND/30m/hand-1000. - Precipitation: CHIRPS daily (Funk et al., 2015); air temperature: ERA5 (Hersbach et al., 2020) and TerraClimate (Abatzoglou et al., 2018), via Google Earth Engine.
- River level: Theia Hydroweb, virtual station R_AMAZONAS_JURUA_KM2695 (ID 0000000001175, product version 2.0), https://doi.org/10.24400/329360/HYDROWEB_WATER_LEVEL (Crétaux & Calmant, 2015).
6 | Colour decision for the figures
code/figures/palette_check.py measures the worst-case OKLab distance of candidate colours against
all stops of the HAND ramp under normal vision, protanopia and tritanopia. No pair of colours
separates reliably from the full ramp, so the article uses colour for group identity (Surara
#2C7FB8, Tucumã #D95F0E) and a black casing on every marker and contour drawn over the raster.
7 | Licence and citation
Data: CC0 1.0 (Dryad default). Code on Zenodo: MIT licence (LICENSE).
Dias da Silva, R. H. P., Castro-Sá, M. J., Barnett, A. A., Peres, C. A., & Gordo, M. (2026). Data
from: Beyond floodplain specialization: ecological flexibility in the white bald-headed uakari
(Cacajao calvus). Dryad Digital Repository. https://doi.org/10.5061/dryad.tht76hff0
Donchyts, G., Winsemius, H., Schellekens, J., Erickson, T., Gao, H., Savenije, H., & van de Giesen, N.
(2016). Global 30m height above the nearest drainage. Geophysical Research Abstracts, 18,
EGU2016-17445-3.
Study site and period. Two groups of white bald-headed uakaris (Surara and Tucumã) were followed at the Lake Surara Biological Reserve – Instituto Juruá, lower–middle Juruá River, Amazonas, Brazil (4°30'35"S, 66°43'20"W), from September 2022 to August 2023. Monthly expeditions comprised 5 days of phenological census and 15 days of behavioural observation; during the habituation period (September–November 2022) expeditions were longer than 20 days.
Relocations (relocations.csv). While one or more group members were present, the observer's position was recorded every 10 min with a Garmin eTrex 20 GPS, together with time, subgroup size, activity (foraging, locomotion, resting, social interaction, agonistic), perpendicular distance to the nearest visible animal, height above ground, vertical stratum, habitat recorded in the field and weather: 2,359 relocations, 2,165 of the Surara group and 194 of the Tucumã group. Group identity was assigned in the field.
Behavioural scans (behavioural_scans.csv). Behavioural data were collected in repeated 3-min cycles following Barnett (2010): each cycle began with an instantaneous scan of the nearest or visible individuals, in sequence, followed by 1 min of ad libitum observation and a 5-s interval before the next cycle; each row is one individual recorded during a scan (17,910 records on 114 days).
Height above the nearest drainage (HAND_local_30m.tif, HAND_regional_30m.tif, habitat_classes_12_19m_30m.tif). The analysis code extracts HAND at each relocation from a 30-m raster of the global HAND product (Donchyts et al., 2016) exported from Google Earth Engine; habitat classes are low várzea (0–12 m), transition (>12–19 m) and terra firme (>19 m).
Phenology and fruit traits (phenology_census_records.csv, fruit_morphometry.csv). Four transects of 4 km × 0.5 m, two in várzea and two in terra firme, were censused once per month. For each fruiting tree or liana we recorded morphospecies, diameter at breast height, Fournier intensity (0–4), fruit maturity, vertical stratum and the number of fruits fallen within the transect strip, and measured usually up to three fallen fruits per plant (length, width, thickness where it was a distinctive dimension of the fruit, fresh mass, and puncture resistance with a penetrometer fitted with a prosthetic uakari canine). Fruit measurements began in November 2022. From February to July 2023 the várzea transects were censused from a canoe.
Feeding records (feeding_visits.csv, feeding_morphotypes_daily.csv). Feeding visits to plants recorded during follows, with the habitat recorded in the field, and daily counts of food morphotypes sorted from photographs of the fruits collected under the plants.
Terra firme plots (terra_firme_plots_inventory.csv, abc_plot_corners.csv). Inventory of all stems of at least 10 cm diameter at breast height, trees and lianas, in three 1-ha plots (500 × 20 m) established by the ABC Expedition, with the centre-line waypoints and corners of each plot.
Climate and river level. Daily precipitation (CHIRPS) and daily mean air temperature (ERA5), monthly minimum and maximum temperature (TerraClimate) for the study area, and the level of the Juruá River from Sentinel-3A altimetry (Theia Hydroweb virtual station R_AMAZONAS_JURUA_KM2695, 2017–2025).
Processing and code. Every change made to the field spreadsheets is listed in corrections_log.csv, and DATA_DICTIONARY.md describes every column. Sixteen R scripts (R 4.5.1; analysis_code.zip, published at Zenodo under the MIT license) reproduce the tables and appendix values of the article from these files; analysis_outputs.zip contains the tables and logs of the last complete run. The README lists the reported values that the deposited files do not reproduce. Version 16 (November 2026) removed 100 duplicated relocations and 46 duplicated scan records, added 21 subgroup counts recovered from the field datasheets and corrected the year of 35 feeding records; each change is a row of corrections_log.csv.
Changes after Nov 10, 2026: Version 16 (November 2026). While checking the proofs of the accepted article, the authors found and corrected four problems in the field data: 100 duplicated rows in relocations.csv (now 2,359 relocations, previously 2,459); 46 behavioural scan records entered twice (now 17,910 records, previously 17,956); 21 missing subgroup counts, now recovered from the field datasheets; and 35 feeding records dated December 2023 instead of December 2022. Each change is a row of corrections_log.csv. All analyses were re-run on the corrected files (analysis_outputs.zip), and the code adds fig6_fig8.R (analysis_code.zip). The README, the data dictionary, the abstract and the methods were updated, and the description of the behavioural sampling protocol now follows the article as corrected at proof (Barnett 2010). The conclusions of the article are unchanged.
