https://doi.org/10.5061/dryad.69p8cz9cs

Description of the data and file structure

Between 2018 and 2022, we conducted five camera-trap surveys in Nui Chua NP with different study designs and survey efforts. These surveys include four fine-scale surveys where cameras were spaced from 300 m to 600 m in selected areas of the park, and one large-scale survey with camera spacing of approximately 2.5 km across the protected area. Camera-trap photographs were managed and the intermediate outputs for analysis produced using the package camtrapR 2.1.1. 0 in R software 4.0.5

To characterize the landscape-scale habitat composition in Nui Chua NP, we ran a random classification forest model in Google Earth Engine using the smileRandomForest algorithm. For input variables in the classification, we used 14 bands derived from a cloud-free Sentinel-2 image composite in 2021, obtained via GEE-PICX. We implemented a principal component analysis (PCA) to reduce the of three major habitat probabilities into two non-correlated dimensions PC1 and PC2

We constructed the Royle-Nichols model and included PC1 and PC2 as covariates in the abundance sub-model to capture the habitat association of communities across Nui Chua NP. We ran the model using JAGS 4.3.0 via the rjags 4-13 in R software 4.0.5.

Files and variables

File: supporting_information_S1_S12.docx

Description: Appendices that support the study's findings

Items

• APPENDIX S1: Map of the Greater Annamites ecoregion and two of its landscapes, the Annamites mountain range and the dry forest in the southern coastal area of Vietnam, derived from Baltzer et al. (2001). The historical extent of semi-arid forest landscape was derived from Schmid (1974)
• APPENDIX S2: Summary of species recorded in each of camera-trap surveys. We excluded Chinese francolin, green-legged partridge, serow, and yellow-bellied weasel from the analysis due to the low number of detections or detected stations
• APPENDIX S3: Photographs of three major habitats and their canopies
• APPENDIX S4: Habitat classification using random forest model
  • Figure S4.1. The inputs (a, b) and outputs (c, d, e, f, g, h, i, j, k) of the random forest model. (a) The cloud-free satellite images in 2021; (b) the training polygons of eight habitat classes across the study site; (c) habitat classification across study site; probabilities of broadleaf wet evergreen forest (d), semi-dry forest (e), dry forest (f), grassland (g), water surface (h), bare ground (i), build-up (j), and agriculture areas (k).
  • Table S4.2. Descriptions of bands derived from cloud-free Sentinel 2 satellite images in 2021 and used in the random forest model. The sentinel 2 satellite images and bands were obtained via GEE-PICX
  • Table S4.3. Confusion matrix and statistics from random forest model. In total, for each habitat, 3850 data points were used for training random forest model, and 1650 data points were used for random forest model validation.
  • Table S4.5. Java code to train random forest model and predict habitat probabilities in Google Earth Engine.
• APPENDIX S5: Principal Component Analysis for random 2000 pixel at 10m resolution across the study site
  • Table S5.1. Eigenvalues
  • Table S5.2. Variables
  • Figure S5.3. Visualized graph of individual pixels from Principal Component Analysis. Habitat classes derived from random forest model (no threshold)
  • Figure S5.4: Spatial visualization of PC dimension 1 and PC dimension 2 across Nui Chua National Park
• APPENDIX S6: Correlation plot of habitat probabilities at camera-trap stations in Nui Chua National Park. E = broadleaf wet evergreen forest; S = semi-dry forest, D = dry forest, D1 = dimension 1 from PCA, D2 = dimension 2 from PCA, (S+E) = sum probability of evergreen and semi-dry forests, (S+D) = sum probability of evergreen and dry forests, and (S+D) = sum probability of semi-dry and dry forest. Only D1 and D2 were used as covariates of local abundance in the Royle-Nichols model.
• APPENDIX S7: Moran’s I statistic and Bayesian p-values of each species
• APPENDIX S8: Species and community responses to dimension 1 and dimension 2. Negative response to pca_D1 indicates that species has higher occurrence in dry forest. Positive response to pca_D1 indicates that species has higher occurrence in broadleaf wet evergreen forest or semi-dry forest. Positive response to pca_D2 indicates species has higher occurrence in semi-dry forest. The light blue bar indicates non-significant response (75% BCI overlaps zero); the dark blue bar indicates moderate significant response (75% BCI does not overlap zero and 95% BCI overlaps zero); the red bar indicates strong significant response (95% BCI does not overlap zero).The black vertical lines are 0-value.
• APPENDIX S9: Mean occupancy estimates of mammal and birds species across three major habitats (using threshold of class probability higher than 0.7)
• APPENDIX S10: Community Royle-Nichols model description
• APPENDIX S11: JAGS model code
• APPENDIX S12: Prior distribution (red area) and posterior distribution (gray area) of community intercepts and coefficients (See Appendix S10 and Appendix S11 for the detail descriptions of the these hyperparameters)

File: supporting_information_S13_recordTable.csv

Description: Output file from the R package camtrapR 2.1.1. 0. See Niedballa et al. (2016) for more details

Variables

• no.: Order number
• station: camera-trap station ID
• Camera: camera ID
• Species: Common name of species
• DateTimeOriginal: Date and time of the species detection
• Date: Date of the species detection
• Time: Time of the species detection
• delta.time.secs: Delta time in second
• delta.time.mins: Delta time in minute
• delta.time.hours: Delta time in hour
• delta.time.days: Delta time in day
• Directory: Camera-trap photograph directory
• FileName: File name
• n_images: Number of photographs
• survey: Survey ID

File: supporting_information_S14_covariateTable.csv

Description: Detailed information and covariate values of each camera-trap station in Nui Chua National Park

Variables

• no.: order number
• station: camera trap station ID
• camera: camera ID
• date_setup: setup date
• date_retrieval: retrieval data
• Problem1_from: Date that camera start to stopped working. NA means camera was functioning
• Problem1_to: Date that "stopped working period" end. NA means camera was functioning
• survey: covariate "survey"
• pca_D1: covariate "PC1"
• pca_D2: covariate "PC2"

Code/software

R software 4.0.5

Google Earth Engine

JAGS 4.3.0