Using the analytical hierarchy process to create variable weights to build a current day goshawk nesting habitat suitability model for Utah national forests and use the Forest Vegetation Simulator to forecast potential changes to nesting habitat suitability out the year 2150

This project repository contains files associated with the Utah State - Climate Adaptation Science goshawk project and the PhD dissertation chapter for Marilyn E. Wright (Utah State University, 2022). It includes literature review, analytical hierarchy process, current condition suitability model, Forest Vegetation Simulator runs, and future habitat suitability models to the year 2150 under four climate scenarios (base, RCP45, RCP60, and RCP85).

Literature Review and Expert Knowledge Solicitation

We conducted a literature review of northern goshawk habitat suitability models that had been completed in western states, USA to determine habitat variables to include in the model and define suitable values for these variables. The journal articles used and extracted values can be found in 'data/lit_review/HSM_variable_parameters.xlsx' excel sheet. Geometric means of each lower and upper thresholds of numeric values were calculated in the excel sheet and can be found on the DATA tab.

Variables along with their threshold values were included in a survey that was sent to experts in the field of northern goshawk research (state and federal government employees, Intermountain Bird Observatory, Rocky Mountain Research Station, graduate student researcher). Participants were asked to rank each variable in pairwise comparisons, using an interactive Excel spreadsheet provided by: Goepel, K.D. (2018). Implementation of an Online Software Tool for the Analytic Hierarchy Process (AHP-OS). International Journal of the Analytic Hierarchy Process, Vol. 10 Issue 3 2018, pp 469-487, https://doi.org/10.13033/ijahp.v10i3.590. A blank copy of the interactive spreadsheet is available at 'data/AHP/AHPcalc-2018-09-15.xlsx' as well as a document explaining how the interactive spreadsheet works available at 'data/AHP/AHPcalc-v2018- 09-15.pdf'. It is important to note here that, while the interactive spreadsheet has the utility to calculate AHP weights, we did not use any of these values and only used the spreadsheet for its utility and ease in describing how to rank pairwise comparisons for participants in our survey. The final full results of the AHP survey can de found at 'data/AHP_survey_responses.xlsx'.

Forest Vegetation Simulator (FVS)

Forest Vegetation Simulator (FVS) -- The Forest Vegetation Simulator is an external program, and, at this time, cannot be interfaced directly through R. We have included two guides to walk you through the FVS steps with screenshots and descriptions, one for the base runs and one for the climate scenario runs. The guides are availabe at 'data/FVS/FVS_Base_Runs_Guide.docx' and 'data/FVS/FVS_Climate_Runs_Guide.docx'. It is important to note that the FVS program is frequently updated. This project was run on a version with release date 2021-06-30 and subsequent releases may change the look and configuration of the program. Additionally, FVS is simulates forest growth based on input data and configuration. While we have opted to use this program as a deterministic model, which is often done in the wildlife management field, future simulations are unlikely to yield identical results. The guides also include instructions for obtaining mountain pine beetle risk and fire risk data. These can be ignored for the purposes of reproducing this project.

For FVS base runs and climate data-- The UT Forest Inventory and Analysis (FIA) data used for this project is available at 'data/FIA/SQLite_FIADB_UT.zip'. While the guides will walk you through setting up FVS from scratch for all the runs, you may alternatively load in the pre-saved runs file available at 'data/FVS/fvsRun.zip' which should have all of the settings saved for you automatically. For Climate-FVS, you must request data from the online database using all the unique standIDs of interest. These can be obtained from shortening the standID information provided in the FIA data to the last We have included two additional coding scripts (scripts/FVSClimate_Data_Request_Prep.R and scripts/UT_FIA_Plots_Visual.R) for clarification of this process, however, the final file needed is available at 'data/FVS/FIA/UTLocs.txt' and the folder received from the FVS help desk is also available at 'data/FVS/CLIMATE/answers.zip'. To work with the Climate-FVS extension, we recommend using only the 'FVSClimAttrs.csv' files in the 'answers.zip' folder. Additionally, we have included this csv which is available at 'data/FVS/Climate/FVSClimAttrs.csv'.

Scripts

Descriptions for all files appearing in the "scripts" sub-directory. All provided code was written in R and most recently verified using R version 4.3.1. To run scripts, download the "data" and "scripts" subdirectories into the same location on your computer, unzip the "data" subdirectory, and set your R working directory to the "scripts" subdirectory. Note that cleaning scripts 2 and 3 cannot be run due to the absence of key raw data files and are provided here only as a reference.

(1). Analytical hierarchy -- We took the raw responses from the survey participants and compiled them into a separate csv file as positive-only responses. This data file can be viewed at 'data/AHP/AHP_8Resp-Positive.csv'. We then used the 'ahpsurvey' package to determine the final variable weights for each chosen habitat variable. The code for calculating habitat variable weights is available at 'scripts/1_Analytical_Hierarchy.R'.

(2-3). REFERENCE ONLY Cleaning spatial data -- We had a variety of cleaning steps for our spatial data including merging individual tiles for elevation and basal area into one raster, setting all spatial files to the same projection (WGS84 aea) and resolution (250-m), clipping all spatial data to the extent of Utah national forest administrative boundaries, and writing new cleaned raster files. We have provided these scripts for you ('scripts/2_Spatial_Cleaning_Halfbaked.R' and 'scripts/3_Spatial_Cleaning.R"), however, we are not including the raw data files for spatial layers due to the large size of the files. We have provided a word document available at 'data/spatial_data/data_source.docx' which describes the sources for original raw spatial data. All output files from the cleaning steps are available in "data/spatial_data/CLEAN'.

(4). Habitat suitability model -- This script first loads in the national forest administrative boundaries file, availabel at 'data/spatial_data/CLEAN/BOUNDARIES/UtahForests.shp', and reprojects this layer to match the projection of all covariate layers. Next, the minima and maxima values are defined for each covariate. These values are based on the threshold values determined with the literature review. A vector of covariate weights is also defined with weight values coming from the analytical hierarchy process in script #1. The habitat suiability model function is defined as follows: 1) Create boolean rasters for each covariate where cells are given a '1' value if they fall within the threshold minimum and maxium or '0' if they fall outside of it 2) Multiply each covariate layer by the corresponding weight determined with AHP 3) Add all covariate rasters together to create a composite layer The final prediction rasters for each forest are saved in the folder 'data/spatial_data/OUTPUT'. Next, the composite suitability rasters for each forest need to be broken into categories for low, medium, and high suitability. We used the full Utah forests continuous composite raster and Jenks natural breaks to create these bins. This section of code takes a decent amount of time to run, so resulting breaks are saved in the script as a vector called 'breaks_jenks'. Finally, this script contains code for each forest to plot the resulting habitat suitability model and to create a bar plot that demonstrates the proportion of cells falling into each suitability category. The code for creating habitat suitability rasters is available at 'scripts/4_Habitat_Suitability_Model.R'.

(5-8). Cleaning FVS data -- These scripts load in the FVS-generated data for each forest and pull out the columns of interest for building the forecasted habitat suitability models. FVS is based on Forest Inventory and Analysis (FIA) plot data. Because FIA data collection happens in a 3-stage process, there are several instances where data is forecasted multiple times for the same plot, based on which stage of FIA data collection was being recorded. Repeat data collection for FIA plots is filtered out, leaving only the most recent FIA data collection as the retained value for forecasting. The final files for each forest can be found at 'data/FSV/CLEAN/~Base or RCP#/~Forest name'. (scripts/5_Cleaning_FVS_Base.R, 6_Cleaning_FVS_RCP45.R, 7_Cleaning_FVS_RCP60.R, and 8_Cleaning_FVS_RCP85.R)

(9-12). Connecting FIA points to FVS -- These scripts take the FVS output files and use the StandID information to connect FVS data to FIA plots. Additionally, we only are interested in forecasting for the year 2150, so these scripts also drop all other year data from the FVS output and save only the spatially explicit FVS shapefiles. Shapfiles can be found in the folder 'data/FVS/spatial_data/~Base or RCP#'. (scripts/9_Connecting_FIA_Points_to_FVS_Base.R, 10_Connecting_FIA_Points_to_FVS_RCP45.R, 11_Connecting_FIA_Points_to_FVS_RCP60.R, and 12_Connecting_FIA_Points_to_FVS_RCP85.R)

(13-16). Kriging FVS -- Because the data from FVS is only spatially explicit at the point level, we need a way to spatially interpolate the data across the full extent of Utah national forests. These scripts take the point data and use two methods to spatially interpolate, kriging for continuous data and proximity polygons (nearest neighbor distance) for categorical data. There is one script for each FVS data type (Base, RCP45, RCP60, and RCP85), and the spatially interpolated rasters can be found in 'data/FVS/interpolated_rasters'. (scripts/13_Kriging_FVS_Base.R, 14_Kriging_FVS_RCP45.R, 15_Kriging_FVS_RCP60.R, and 16_Kriging_FVS_RCP85.R)

(17). Forecasting the habitat suitability model -- Once we have spatially interpolated rasters for each FVS simulation, then we can apply the same steps for modeling habitat suitability that we did in step #4. This script uses the same threshold values, AHP-derived weights, and habitat suitability model function to create prediction rasters for each FVS simulation in the year 2150. To facilitate comparison, the same Jenks breaks are then used to classify habitat suitability into low, medium, and high. Final prediction rasters are saved in the folder 'data/FVS/FUTURE_HSM'. (scripts/17_Forecasted_Habitat_Suitability_Models.R)

(18). Forecasted consensus -- This final code takes all of the prediction rasters for the year 2150 and keeps only the cells that remain as 'high suitability' throughout all four simulations. The purpose of this code is to illustrate the areas where, regardless of potential changes in climate, high suitability may be preserved, creating a potential climate refugia. The consensus raster is available at 'data/FVS/FUTURE_HSM/consensus.tif'. (scripts/18_Forecasted_HSM_Consensus.R)