Budworms, beetles, and wildfire: disturbance interactions influence the likelihood of insect-caused disturbances at a subcontinental scale
Data files
Aug 26, 2024 version files 7.38 GB
-
dfb.ovr.csv
-
JoE.Rmd
-
outbreak_dist.csv
-
README.md
-
us.shape.gpkg
-
us1.gpkg
-
wls.dat.csv
-
wsb.ovr.csv
Abstract
Irruptive forest insects are a leading biotic disturbance across temperate and boreal forests. Outbreaks of forest insects are becoming more frequent and extensive due to anthropogenic drivers (e.g., climate and land-use), perhaps increasing the likelihood that forests will experience multiple insect-caused disturbances.
Across the fire-prone Douglas-fir forests of western North America, recent outbreaks of the western spruce budworm and Douglas-fir beetle have impacted large expanses of forests, with a higher degree of overlap than expected in some ecoregions. Outbreaks of both insects are positively related to host availability and exhibit density-dependent population dynamics that are affected by climate and weather.
Here, we leverage data from aerial detection surveys, estimates of host availability, climate and weather, and categorized fire severity to describe the spatial overlap between western spruce budworm and Douglas-fir beetle and assess: 1) how climate and host availability influence the biogeography of outbreaks; 2) how weather incites outbreaks; and finally, 3) how prior disturbances (fire and biotic) affect the subsequent outbreak likelihood of western spruce budworm and Douglas-fir beetle.
Models demonstrate that western spruce budworm and Douglas-fir beetle share similar predisposing drivers of outbreaks. Outbreaks of both insects were more likely to occur following warm weather, but only beetle outbreaks were more likely following drought. When controlling for differences in outbreak distribution and inciting factors, results indicate that both prior fire and interspecific disturbance altered the likelihood of subsequent insect-caused disturbance.
Specifically, Douglas-fir beetle outbreaks were more likely to occur for several years following low severity fire, but less likely otherwise. Prior defoliation, especially longer duration defoliation, increased the likelihood of beetle outbreak within stands and across the landscape. On the contrary, western spruce budworm outbreaks were less likely to occur following fire, while prior beetle activity dampened budworm outbreak likelihood for several years, then eventually increased outbreak likelihood.
Synthesis: Biotic-biotic disturbance interactions have the potential to amplify the incidence of insect-caused disturbance across sub-continental scales. Our findings highlight the need for future work on mechanistic linkages between biotic disturbance agents as well as the ramifications for forest trajectories and function.
README: Budworms, beetles, and wildfire: disturbance interactions influence the likelihood of insect-caused disturbances at a subcontinental scale.
https://doi.org/10.5061/dryad.bcc2fqzns
Description of the data and file structure
Overview—We compiled spatially and temporally explicit data (Table 1) to describe the drivers of Douglas-fir beetle and western spruce budworm outbreaks and test for potential disturbance interactions among wildfire, Douglas-fir beetle outbreaks, and western spruce budworm outbreaks. Because the outbreak distribution of insect herbivores is constrained by bioclimatic factors, we first developed a model to describe how host availability and climate predispose forests to outbreaks of Douglas-fir beetle or western spruce budworm (Outbreak Distribution). Second, because inter-annual variation in weather is known to influence insect outbreaks, we developed a model to describe the inciting factors (i.e., weather) of outbreaks (Inciting Factors). Third, to understand disturbance interactions, we developed models to specifically test how prior wildfire, interspecific insect outbreaks, and conspecific population pressure affected the likelihood of Douglas-fir beetle and western spruce budworm (Disturbance Interactions).
Script Overview
- Load necessary packages, tuning machine learning models
Imports formatted data (included in files; contact Michael.Howe@usda.gov for scripts processing raw data). Each model is contained in a function.
- Model 1: Outbreak distribution (machine learning model; exports model)
- Model 2: Inciting factors (machine learning model; exports model)
- Model 3: Disturbance Interactions (generalized linear model; exports model)
Imports model results from all models (requires correct path structure), and makes all figures. Figures are separated into sections.
Additional figures based on imported model results.
Files and variables
Usage notes: All files were created in R and can be imported in R (including spatial data). Spatial data can also be opened in QGIS.
File: us.shape.gpkg
Description: Spatial file (EPSG:3005) used for mapping model results.
File: us1.gpkg
Description: Spatial file (EPSG:3005) containing the spatial grid information (g1k).
File: outbreak_dist.csv
Description:
Variables
- g1k: Spatial Index
- CMI: Hogg's Climate Moisture Index
- MAP: Mean Annual Precip. (cm)
- NFFD: Number of Frost-free days (days)
- PAS: Precip. as snow (cm)
- PPT_at: Precip. in autumn (cm)
- PPT_sp: Precip. in spring (cm)
- PPT_sm: Precip. in summer (cm)
- PPT_wt: Precip. in winter (cm)
- Tave_at: Mean temp (˚C) in autumn
- Tave_sp: Mean temp (˚C) in spring
- Tave_sm: Mean temp (˚C) in summer
- Tave_wt: Mean temp (˚C) in winter
- Tmin_wt: Min. temp (˚C) in winter
- gs.tave: Mean spring + summer temp. (˚C)
- ow.tave: Mean autumn + winter temp. (˚C)
- gs.ppt: Mean spring + summer precip. (cm)
- ow.ppt: Mean autumn + winter precip. (cm)
- max.df.ba: Max. Douglas-fir basal area (log(m2/ha+1))
- max.ba: Max. basal area of spruce & Douglas-fir (log(m2/ha+1))
- dougfir: Douglas-fir basal area (log(m2/ha+1))
- fir: Fir basal area (log(m2/ha+1))
- spruce: Spruce basal area (log(m2/ha+1))
- total: Sum Douglas-fir, Fir, and spruce basal area ((log(m2/ha+1))
- wsb: Observed western spruce budworm
- dfb: Observed Douglas-fir beetle
- df: Rounded Douglas-fir basal area
- tot: Round total basal area
File: dfb.ovr.csv
Description:
Variables
- g1k: Spatial Index
- year: Year
- obs: Observed Douglas-fir Beeetle
- fire.severity_t1: binned fire severity (low/high) in year t-1
- fire.severity_t2: binned fire severity (low/high) in year t-2
- fire.severity_t3: binned fire severity (low/high) in year t-3
- fire.severity_t4: binned fire severity (low/high) in year t-4
- fire.severity_t5: binned fire severity (low/high) in year t-5
- fire.severity_t6: binned fire severity (low/high) in year t-6
- fire.severity_t7: binned fire severity (low/high) in year t-7
- fire.severity_t8: binned fire severity (low/high) in year t-8
- fire.severity_t9: binned fire severity (low/high) in year t-9
- fire.severity_t10: binned fire severity (low/high) in year t-10
- wsb_t1: binned western spruce budworm (low/high) in year t-1
- wsb_t2: binned western spruce budworm (low/high) in year t-2
- wsb_t3: binned western spruce budworm (low/high) in year t-3
- wsb_t4: binned western spruce budworm (low/high) in year t-4
- wsb_t5: binned western spruce budworm (low/high) in year t-5
- wsb_t6: binned western spruce budworm (low/high) in year t-6
- wsb_t7: binned western spruce budworm (low/high) in year t-7
- wsb_t8: binned western spruce budworm (low/high) in year t-8
- wsb_t9: binned western spruce budworm (low/high) in year t-9
- wsb_t10: binned western spruce budworm (low/high) in year t-10
- dfb.lag_t1: binned Douglas-fir beetle within 5 km (low/high) in year t-1
- dfb.lag_t2: binned Douglas-fir beetle within 5 km (low/high) in year t-2
- dfb.lag_t3: binned Douglas-fir beetle within 5 km (low/high) in year t-3
- dfb.lag_t4: binned Douglas-fir beetle within 5 km (low/high) in year t-4
- dfb.lag_t5: binned Douglas-fir beetle within 5 km (low/high) in year t-5
- dfb.lag_t6: binned Douglas-fir beetle within 5 km (low/high) in year t-6
- dfb.lag_t7: binned Douglas-fir beetle within 5 km (low/high) in year t-7
- dfb.lag_t8: binned Douglas-fir beetle within 5 km (low/high) in year t-8
- dfb.lag_t9: binned Douglas-fir beetle within 5 km (low/high) in year t-9
- dfb.lag_t10: binned Douglas-fir beetle within 5 km (low/high) in year t-10
- cl.pred: climate prediction (model 2)
- ls.pred: landscape prediction (model 1)
File: wsb.ovr.csv
Description:
Variables
- g1k: Spatial Index
- year: Year
- obs: Observed western spruce budworm
- fire.severity_t1: binned fire severity (low/high) in year t-1
- fire.severity_t2: binned fire severity (low/high) in year t-2
- fire.severity_t3: binned fire severity (low/high) in year t-3
- fire.severity_t4: binned fire severity (low/high) in year t-4
- fire.severity_t5: binned fire severity (low/high) in year t-5
- fire.severity_t6: binned fire severity (low/high) in year t-6
- fire.severity_t7: binned fire severity (low/high) in year t-7
- fire.severity_t8: binned fire severity (low/high) in year t-8
- fire.severity_t9: binned fire severity (low/high) in year t-9
- fire.severity_t10: binned fire severity (low/high) in year t-10
- dfb_t1: binned Douglas-fir beetle (low/high) in year t-1
- dfb_t2: binned Douglas-fir beetle (low/high) in year t-2
- dfb_t3: binned Douglas-fir beetle (low/high) in year t-3
- dfb_t4: binned Douglas-fir beetle (low/high) in year t-4
- dfb_t5: binned Douglas-fir beetle (low/high) in year t-5
- dfb_t6: binned Douglas-fir beetle (low/high) in year t-6
- dfb_t7: binned Douglas-fir beetle (low/high) in year t-7
- dfb_t8: binned Douglas-fir beetle (low/high) in year t-8
- dfb_t9: binned Douglas-fir beetle (low/high) in year t-9
- dfb_t10: binned Douglas-fir beetle (low/high) in year t-10
- wsb.lag_t1: binned western spruce budworm within 5 km (low/high) in year t-1
- wsb.lag_t2: binned western spruce budworm within 5 km (low/high) in year t-2
- wsb.lag_t3: binned western spruce budworm within 5 km (low/high) in year t-3
- wsb.lag_t4: binned western spruce budworm within 5 km (low/high) in year t-4
- wsb.lag_t5: binned western spruce budworm within 5 km (low/high) in year t-5
- wsb.lag_t6: binned western spruce budworm within 5 km (low/high) in year t-6
- wsb.lag_t7: binned western spruce budworm within 5 km (low/high) in year t-7
- wsb.lag_t8: binned western spruce budworm within 5 km (low/high) in year t-8
- wsb.lag_t9: binned western spruce budworm within 5 km (low/high) in year t-9
- wsb.lag_t10: binned western spruce budworm within 5 km (low/high) in year t-10
- cl.pred: climate prediction (model 2)
- ls.pred: landscape prediction (model 1)
- ls.pred.df: landscape prediction based on Douglas-fir (only douglas-fir)
- ls.pred.tot: landscape prediction based on alternative hosts (Douglas-fir, spruce, and fir)
File: JoE.Rmd
Description: R Script. Loads formatted data, fits models (within functions), exports model results, creates manuscript figures and supplemental analyses.
File: wls.dat.csv
Description:
Variables
- g1k: Spatial Index
- year: Year
- NFFD_dev: Deviation in number of frost-free days
- gs.tave_dev: Deviation in mean spring and summer temp. (˚C)
- ow.tave_dev: Deviation in mean autumn and winter temp. (˚C)
- Tmin.wt_dev: Deviation in minimum winter temp. (˚C)
- PAS_prop: Proportional deviation in precip. as snow
- gs.ppt_prop: Proportional deviation in mean spring and summer precip.
- ow.ppt_prop: Proportional deviation in mean autumn and winter precip.
- dfb: Observed Douglas-fir beetle (per year)
- wsb: Observed western spruce budworm (per year)
- wsb.obs: Observed western spruce budworm (anytime between 1997-2022)
- dfb.obs: Observed Douglas-fir beetle (anytime between 1996-2021)
Code/software
All analyses were performed in R. Code was last evaluated in R version 4.4.1 (2024-06-14). All files are accessible in R (including spatial .gpkg).
Necessary Packages
terra, sf, stars, mlr, tidyverse
Optional Packages (help with figures/running full script)
janitor, cowplot, extrafont, pals, DALEXtra
Access information
Data was derived from the following sources:
- United States Detection Survey Data (Aerial Detection Surveys, Forest Health Protection, United States Forest Service, 1997-2022)
- Individual Tree Species Parameter Maps (Individual Tree Species Parameter Maps, Forest Health Protection, United States Forest Service, 2012)
- ClimateNA (Wang et al., 2016)
- Monitoring Trends in Burn Severity (Eidenshink et al., 2007)
Methods
Aerial surveys– To characterize the spatiotemporal patterns of Douglas-fir beetle and western spruce budworm outbreaks, we used aerial detection survey (ADS). ADS data are collected by aerial observers in fixed-wing aircrafts and include the likely damaging agent (species of insect/disease or abiotic factor), binned estimates of damage severity, and forest type. ADS data are the most accurate data describing the spatiotemporal patterns of specific forest insect outbreaks at a broad spatial extent across the western US; a recent assessment of the survey program found that 85% of randomly selected Douglas-fir beetle and 100% of western spruce budworm polygons were correctly identified based on ground-truthing in Oregon, California, and New Mexico (Coleman et al., 2018).
We summarized ADS data by calculating the presence/absence and severity of Douglas-fir beetle and western spruce budworm within a 1 km2 grid. First, we listed presence when a grid cell contained any polygons or points that listed Douglas-fir beetle or western spruce budworm as the primary agent. Observed Douglas-fir beetle activity data were shifted by one year because reported tree damage generally represents bark beetle activity in the prior year (i.e., shifted from 1997-2021 to 1996-2020) (Meddens et al., 2012). Because the effects of defoliation are more immediate and readily visible, dates of observed western spruce budworm activity were unchanged (Senf et al., 2015). Second, we estimated Douglas-fir beetle and western spruce budworm outbreak severity using the ‘histogram matching’ method, which accounts for the changes in the ADS protocol over the study period (Egan et al., 2019, 2020; Hicke et al., 2020) (Supplemental Materials 1). We considered the maximum reported ADS severity for each grid cell-year combination (low, medium, or high) and simplified severity classes into low (“low”) or high (“medium” & “high”) due to the highly uneven distribution of reported severities.
Given that an important driver of spatiotemporal patterns of insect outbreaks is the population density within the surrounding landscape (Aukema et al., 2008; Howe et al., 2021; Senf et al., 2017; Simard et al., 2012), we estimated prior conspecific population pressure at 5 km (e.g., t-1). Five kilometers exceeds the dispersal distance of individual Douglas-fir beetles, which is generally less than 400 m (Dodds & Ross, 2002), but is well within the effective distance of landscape spatial correlation (i.e., nonparametric correlation function; Supplemental Materials 1). The dispersal distance of western spruce budworm is not well known (Flower et al., 2014), although dispersal distance is suggested to be roughly 1.3 km during outbreaks (Senf et al., 2017). Budworms are also capable of much longer dispersal flights as a similar species, eastern spruce budworm (Choristoneura fumiferana [Clemens, 1865]), can disperse more than 100 km within a season (Greenbank et al., 1980). In our analysis of landscape spatial correlation (Supplemental Materials 1), western spruce budworm outbreak populations were highly correlated at distances between 0-5 km with declining correlation between 5-10 km. Specifically, we used simplified inverse sigmoidal distance weighted sum of Douglas-fir beetle and western spruce budworm presence/absence, similar to other analyses of bark beetle population pressure (Hart et al., 2017; Howe, et al., 2022; Preisler et al., 2012). Estimates of prior conspecific population pressure were log(x+1) transformed for a better distribution of values.
Fire severity—We calculated estimates of burn severity by intersecting a 30 m categorized fire severity raster from the Monitoring Trends in Burn Severity (MTBS) Project with our 1 km2 grid. For each 1 km2 cell, we summed the number of 30 m pixels within each fire severity category (“greener”, “low”, “medium”, “high”, “unburned”, “masked” [i.e., by clouds]), multiplied the area of each by 0.1 for low, 0.25 for medium, or 0.75 for high severity, which approximately correspond to the amount of overstory mortality typically associated with each burn severity class (e.g., low: < 25% mortality; medium 25-75% mortality; high > 75% mortality). For each 1 km2 cell, we then classified fire severity into “low” or “high” with a breakpoint of 0.5 (approximately less than or greater than 50% basal area loss). Fire severity breakpoints were selected to approximate the bins used in the ADS data to classify insect damage and loss of forest basal area. A comparison of our binning method relative to the 30 m categorized fire severity classes is provided (Supplemental Materials 1).
Host availability–To quantify host basal area, we aggregated 2002 and 2012 30 m rasters of Douglas-fir basal area to 1 km2 by intersecting with our 1 km2 grid. For each 1 km2 pixel, we calculated the area-weighted mean, transformed the units into m2/ha, and performed a log transform (i.e., log[m2/ha+1]) for both the 2002 and 2012 rasters. We then used the maximum estimate of Douglas-fir basal area for each grid cell because the 2012 data did not include all basal area lost due to Douglas-fir beetle outbreaks in the early 2000s. We also estimated spruce and fir basal area based on the 2012 data to examine how incorporating alternative western spruce budworm hosts altered western spruce budworm outbreak distribution (Supplemental Materials 2).
Climate—We upscaled 800 m rasters of thirty-year climate normals (1961-1990) from ClimateNA (T. Wang et al., 2016) by intersecting with the 1 km2 grid and calculated the mean across each grid cell. We used 1961-1990 as our reference time-period for climate because Douglas-fir trees are long-lived and there have been recent significant increases in temperature across the western United States. Thus, 1961-1990 represents a reference for conditions at the beginning of our study, which spans 1996-2021. In total, we considered 17 climatic normal variables (Supplemental Materials 3) for our analyses. Annual weather rasters were upscaled as above and we calculated weather as either the deviation from climatic normal over a three-year period (t-0:t-2) for temperature (˚C) or the proportional deviation from climatic normal for precipitation (proportion based on cm). Three-year weather deviations were selected for two reasons: A) visible outbreaks are the product of multi-year population expansions; and B) it can take several years for decreases in precipitation to impact tree vigor and non-structural carbohydrate pools (Peltier et al., 2023). Proportional deviation in precipitation was used to control for the large bioclimatic differences across our study area. As an example, for a given grid cell in the year 2000, we calculated the difference from the 1961-1990 climatic normal over the 1998-2000 time-period. We considered 7 weather variables for our analyses (Table 1; Supplemental Materials 3).