This archive folder includes tabular datasets, statistical models, and model outputs from a study of tree-level radial growth in five ponderosa pine-dominated experimental sites in Arizona, USA. At each site, restoration treatments (i.e., thinning and burning) were applied after an initial forest inventory. Subsequent inventories occurred at each site after treatment implementation, and increment cores were collected from a subset of trees in treated and untreated portions of each site in 2019 to supplement other available data. Descriptions of each dataset are provided below, but further information is available in Rodman et al. (2024). This folder was compressed in .zip format using Windows 11 v. 23H2x64
This data package is split into four subfolders, as described below. To replicate our analyses, the R project file “Rodman_MultisiteCoreData.Rproj” should be opened in a new session of R Studio software. This allows the use of relative file paths in the “here” package to re-run scripts described below.
This folder contains all R and Google Earth Engine (javascript) code used to process, analyze, and visualize data in this project. The three subfolders relate to each of these steps.
This folder includes the following scripts used to format data and create spatial datasets used in subsequent analyses. All files with the extension “.txt” were written in Java for use in cloud-based computing via Google Earth Engine. All files with the extension “.r” were written in R, for use in local processing.
Obtains data on average annual climate at each field site from the Daymet dataset (Thornton et al. 2021), and exports them as a .csv file and raster format (geotiff; .tif), for use in making Fig. 1b of Rodman et al. (2024).
Reads in list of all trees on the sites in 2019, to develop a stratified random sample of increment cores to be collected and dated.
Formats field and spatial data as inputs for a daily water balance model (Schlaepfer et al. 2012, Bradford et al. 2014), used to estimate available water capacity - a predictor of tree growth.
Reads in annual tree ring width information, cleans it, converts it to annual basal area increment, and organizes it in an analysis-ready format. These data are used as a response variable in subsequent analyses.
Reads in daily weather data for vapor pressure and available water capacity and converts them to seasonal means.
Takes tree lists, calculates live plot-level tree basal area in each field inventory year, and formats these data for use with tree ring data
Takes response data (annual basal area increment) and covariates (basal area, weather, tree size) and merges them into a named list object in .rds format for use as input data in JAGS software.
Fits a simplified version of our model, as a frequentist generalized linear mixed model, to create initial values for the Bayesian model in JAGS.
This folder includes the following files/scripts used to analyze data and evaluate fitted models.
A named list, produced in the “Step1g-PrepDataForJAGS.R” script, and used as an input into the initial JAGS model (4,000 iterations in each of three chains).
A named list, produced in the “Step1h-SettingInitials.R” script, which gives starting parameter values in the initial JAGS model (4,000 iterations in each of three chains).
This is a batch script used to initiate a processing task in the Monsoon high performance computing cluster (with SLURM task manager) at Northern Arizona University to run the initial Bayesian model of tree growth (4,000 iterations across each of three chains).
Used to define the structure of the initial Bayesian model of tree growth (4,000 iterations across each of three chains). It is called within the Monsoon cluster using the “main_job.sh” and “script.r” files
Runs the initial Bayesian model of tree growth (4,000 iterations across each of three chains) using JAGS and jagsUI. It is called in the Monsoon cluster using the “main_job.sh” file
This object is a named list, produced in the “Step1f-PrepDataForJAGS.R” script, and used as an input into the final JAGS model (20,000 iterations in each of three chains).
This object is a named list, produced in the “Step2b-GrowthModel_FinalInitsAndSettings.R” script, which gives starting parameter values in the final JAGS model (20,000 iterations in each of three chains).
This is a batch script used to initiate a processing task in the Monsoon high performance computing cluster (with SLURM task manager) at Northern Arizona University, to run the final Bayesian model of tree growth (20,000 iterations across each of three chains).
Used to define the structure of the final Bayesian model of tree growth (20,000 iterations across each of three chains). It is called within the Monsoon cluster using the “main_job.sh” and “script.r” files
Runs the final Bayesian model of tree growth (20,000 iterations across each of three chains) using JAGS and jagsUI. It is called in the Monsoon cluster using the “main_job.sh” file
Reads in the initial Bayesian model of tree growth (4,000 iterations across each of three chains), and uses it to set initial values and determine the number of iterations needed for convergence in the final model.
Evaluates the final bayesian model of radial growth (“growthMod_full.rds”; full model object is not included in this archive because of large file sizes - 1.4gb) to assess convergence and summarize/interpret fitted parameters.
Compares time since thinning, time since burning, and treatment intensity with model residuals from hierarchical bayesian regression of tree growth, using only the data from trees that experienced treatment. This evaluates whether treatment effects might have waned throughout the coarse of the study, and creates Figs. S2 and S3 in Rodman et al.
This folder includes the following scripts used to visualize data and the results of the final Bayesian model of tree growth.
Takes climate data outputs from “GEE_GetClimate.txt” and other locations in this data package, compares them to the overall distribution of Rocky Mountain ponderosa pine, and plots time series of precipitation and temperature at each of the five experimental sites. This script makes Fig. 1b and 1c in Rodman et al., as well as Fig. S4 (a summary of modeled soil moisture over time)
Plots the mean +- 1.96*SE (95% interval) of annual tree basal area increment values at each of the five sites, as well as the timing of thinning and burning. This script makes Fig. 2 in Rodman et al.
Extracts posterior summary statistics from the final Bayesian model of tree growth, and creates some visual summaries of the model results - covariate effects, marginal effect plots, and summaries of antecedent climate terms (i.e., lags and seasons that most influence growth). This script makes Figs. 3-6 in Rodman et al.
This subfolder includes raw and summarized data used in Rodman et al., including tree ring data, forest inventories, water balance model outputs, and spatial data extractions (e.g., annual canopy cover, vapor pressure deficit). Individual files are split into additional subfolders, which are described below.
This folder includes files that have been cleaned and organized for use in the Bayesian hierarchical model of tree growth, and associated diagnostics.
This file is the final version of the data, used in statistical analyses for Rodman et al. (2024). Rows represent a single year of growth for an individual tree. Columns are as follows:
This file is a subset of the original data, used to evaluate the effect of time since thinning and time since burning on residuals of the hierarchical Bayesian model of tree growth. This analysis, used to create Fig. S2 in Rodman et al. (2024), was used to determine if the effects of treatment declined throughout the study period. Descriptions of rows and columns and rows are identical to those in “growthData_merged.csv” described above, but with the exception that rows are restricted to only trees and years that were considered “treated”, so pre-treatment years and untreated units are removed. One newly added column is as follows:
This folder includes information about seasonal vapor pressure deficit or available soil water at the five experimental sites. Raw daily outputs are not included here due to large file sizes (> 10 gb for SOILWAT2 outputs and daily VPD data), but are available upon request. Please contact Kyle Rodman (kyle.rodman@nau.edu) with requests for these data
This file includes summarized seasonal climate data, used as as covariates in the hierarchical Bayesian model of tree growth. Rows represent each combination of PlotID and year columns (identical to those described in “growthData_merged.csv” above). Columns are as follows:
This subfolder includes raw and summarized information on tree growth, forest structure, and the timing of thinning and burning at each of five experimental sites in Arizona, USA, included in Rodman et al.
This folder includes files used to create a stratified sample of increment cores for processing and dating in the laboratory. Files are created or used in “Step1a-CoreSubsetting.R” (see notes in this script for additional information) but are not described in detail here.
This description applies to five subfolders, one for each [Site]. Each subfolder has raw tree ring data, outputs of COFECHA software (Holmes 1983), and various notes from processing for each corresponding site. [Site] refers to AS = Apache-Sitgreaves, CF = Centennial Forest, FV = Fort Valley, GV = Grandview, MT = Mt. Trumbull. Files in each subfolder are as follows:
This object is a generalized linear mixed model, fit using the lme4 package in R, which can be read into R by calling “library(lme4)” and using the “readRDS” function. It was developed to predict bark thickness as a function of tree diameter and other covariates in Rodman et al. 2021. We used it to help develop predictions of basal area increment and reconstructed diameter at breast height. This model is used in the “Step1c-FormattingCoreData.r” script described above
This file includes forest inventory data from cored trees at each of five experimental sites in Arizona, USA. These data were used to help select a stratified random sample of cores to be dated and processed, and to link increment cores to tree-level data (e.g., diameter at breast height). Rows represent individual trees included in 2019 surveys that were cored (not a comprehensive list of trees on all sites). Columns are as follows:
This file includes information about all trees taller than breast height (1.37 m) on each of five experimental sites in Arizona, USA. Unlike the “treeData_allSites.csv” file, this list is comprehensive, and includes information about all trees on a site, including uncored trees and those of species other than ponderosa pine. Each row represents a tree in a specific field inventory. Thus, most trees are represented multiple times, as they may have been surveyed in each field inventory. Dead trees were typically not surveyed again after the first inventory in which they were identified as “dead”. New ingrowth (i.e., trees that grew above 1.37 m during the course of the study) was included in only later surveys in which the individuals were taller than 1.37 m. Columns are as follows:
This table gives the timing of thinning and burning at each site. Values give the first year in which the treated unit at a given site and block might be considered “Treated” by thinning and/or burning. As described in datasets above, this does not necessarily correspond to the calendar year of treatment - e.g., a treatment year of 2019 could indicate that a site was treated in the summer/fall of the previous year, or the spring of the current year. Numbers do not apply to untreated units at each site, where thinning and burning were never implemented. Rows represent each site and experimental block. Columns are as follows:
This dataset has cleaned and formatted data on tree basal area increment (in sq. mm/year) and DBH (diameter in centimeters at 1.4 m above ground level) for all trees and years included in hierarchical Bayesian models of tree growth. Rows identify individual growth years for a given tree from 1950 to 2018. Data were restricted to 1985 to 2018 for subsequent analyses based on the availability of covariates. This file was produced using the “Step1c-FormattingCoreData.r” script described above. Columns are as follows:
This dataset has cleaned and formatted data on annual tree ring width for all trees and years included in hierarchical Bayesian models of tree growth. Rows identify individual growth years for a given tree from 1950 to 2018. Data were restricted to 1985 to 2018 for subsequent analyses based on the availability of covariates. This file was produced using the “Step1c-FormattingCoreData.r” script described above. Columns are as follows:
This folder includes several files that provide the locations of each field plot/site, or extractions of spatial data (e.g., annual Landsat-derived canopy cover) at these locations, to be used as covariates in the hierarchical Bayesian model of tree growth. Individual files are further described below.
This file provides the locations of all field plots at these five experimental sites. This may be useful for future users to extract spatial datasets at the locations of field inventories. Many plots that were not included in our analyses (because they contained no cored trees or were on blocks that experienced fire or another disturbance during the course of the study) are included here. Rows represent each field plot. Columns are as follows:
This subfolder is intentionally empty, but included in this data package to permit reproducibility of R scripts in “Code/Step3-Visualization”.
This subfolder contains JAGS (Just Another Gibbs Sampler; v. 4.3.0) model outputs and associated model diagnostics created during this project. All .rds files can be opened in R using the “readRDS” function.
Within the ModelOutputs folder, this subfolder contains a hierarchical Bayesian model used to predict tree growth as a function of treatment (i.e., thinned/burned or not treated), plot-level basal area, tree size, and interannual climate variability. The folder also includes associated diagnostics. This model and set of diagnostics are from the initial model (4,000 iterations in each of three chains), used to develop initial values and the number of iterations used in the final model (contained in “growthV3_bayesianFinal” described below). Code and input data used to develop this initial model are in “Code/Step2-Analysis/ForMonsoon/Growth_v2_4000its”
This subfolder contains traceplots (created with the mcmcplots package in R) for tracked parameters in the the initial model, used to help assess convergence. Individual traceplots are in .png data formats. To view all of these plots in one place, open the “mcmcPlots_Growth4000its.html” file
This model object, fit using JAGS v. 4.3.0 and the jagsUI package in R, was created on the Monsoon high performance computing cluster at Northern Arizona University. The model object can be read into R using the ‘readRDS’ function in R to summarize information about posterior distributions of each parameter.
This object gives estimated time to convergence (following Raftery and Lewis 1992) for the initial model (4,000 iterations across each of three chains). The object can be read into R using the ‘readRDS’ function in R.
Within the ModelOutputs folder, this subfolder contains a hierarchical Bayesian model used to predict tree growth as a function of treatment (i.e., thinned/burned or not treated), local canopy cover, tree size, and interannual climate. The folder also includes associated diagnostics. These models and diagnostics are from the final model (20,000 iterations in each of three chains), presented in Rodman et al. Code and input data used to develop this final model are in “Code/Step2-Analysis/ForMonsoon/Growth_v3_Final”
This subfolder contains traceplots (created with the mcmcplots package in R) for tracked parameters in the the final model, used to help assess convergence. Individual traceplots are in .png data formats. To view all of these plots in one place, open the “mcmcPlots_GrowthAllits.html” file
This object gives summarized information on posterior distributions (median; 95% CI) of each model covariate in the final 20,000 iteration model of tree growth. These values are plotted in Fig. 3 of Rodman et al. (2024). The object can be read into R using the ‘readRDS’ function in R.
This object gives an additional 1,000 iterations in each of three chains (using final values from the final model with 20,000 iterations), used to calculate goodness of fit statistics. The object can be read into R using the ‘readRDS’ function in R.
This object gives values of all 20,000 iterations in each of three chains for tracked parameters in the final Bayesian model of tree growth, used to create traceplots and evaluate model fit. The object can be read into R using the ‘readRDS’ function in R.
This object gives convergence statistics (r hat values following Gelman and Rubin 1992) for the final model (20,000 iterations across each of three chains). R hat values close to 1 indicate stability of a tracked parameter across the three chains. The object can be read into R using the ‘readRDS’ function in R.