Project context and background information

Pest management is an important aspect of our current existence as climate change impacts organism ranges, species are introduced into new habitats where they become invasive, and human activity in general shifts the world around us. My research so far has included the study of both native and invasive pests, and for this project I wish to focus on an ever-present Canadian forestry pest: the spruce budworm, and its historical management.

Spruce budworm has gained popularity in the public sphere following multiple news reports highlighting the problem of frequent outbreaks. At this year’s Humber Valley Strawberry Festival in Pasadena, NL, spruce budworm moths were so numerous that their gathering resembled a blizzard:

News Articles

Moth madness at Pasadena festival symptom of spruce budworm battle in N.L.

Moth blizzard at the Humber Valley Strawberry Festival, photo submitted by Melanie Wells.

Moth blizzard at the Humber Valley Strawberry Festival, photo submitted by Melanie Wells.

Though events like this are rare to see, they have drawn interest outside of the scientific community in the role of spruce budworm as a pest and integral part of Canadian forest ecosystems. News articles on their role, like the pest itself, make their rounds every few years. One such recent example which draws on current expert knowledge about the proliferation and management of the pest is the anchor article for this project:

Go inside the outbreak ravaging Gros Morne’s forests

A spruce budworm caterpillar in Gros Morne National Park. Photo credit: taken from Lindsay Bird's CBC News article.

A spruce budworm caterpillar in Gros Morne National Park. Photo credit: taken from Lindsay Bird’s CBC News article.

Lindsay Bird, CBC News, November 23, 2025. Read the article here.

This project examines quantitative methods used in forest pest management through historical spruce budworm insecticide-application records from New Brunswick.

Goals and Research Questions

Overall, the research questions aimed to be answered by this analysis are:

  1. How did recorded insecticide use in New Brunswick change from 1952 to 1993?
  2. How did the recorded composition of applications by active ingredient change over that period?
  3. How well does a linear model summarize the trend in annual recorded chemical mass, and do more flexible polynomial models improve that summary without overfitting?

In examining these research questions, the project will focus on the following three analytical goals:

  1. Describe the application records, including the meaning and units of the variables used.
  2. Summarize how recorded chemical mass and the mix of active ingredients changed over time.
  3. Fit and evaluate a linear baseline and compare it with more flexible polynomial models of annual recorded chemical mass.

Data and variables

The application and treated-area records are taken from Heartz et al. (2023), with data supplied by the Government of New Brunswick. The application table contains records of insecticide use; the areas table contains treated-area and spray-technology information. The tables are joined using app_area, an identifier shared between them.

Variable Meaning in this analysis Units or notes
year Year of the application record Calendar year
app_area Identifier linking an application to its treated-area record Dataset identifier
act_ing Active ingredient recorded for an application BT is the dataset code for Btk
rate_kgha Application rate for ingredients measured by mass kg/ha
vol_lha Spray-mix carrier volume per area L/ha
area_ha Treated area, calculated from the areas table Hectares; source area converted from m²
app_type Operational or experimental classification of an area Category in source data
guidance Spray-navigation or guidance method recorded for an area Text category
app_sys Spray-delivery system recorded for an area Text category

Annual chemical mass is estimated by multiplying rate_kgha by area_ha and summing by year. Btk is excluded from that mass calculation because its rate is recorded in BIU/ha, not kg/ha. Carrier volume is calculated separately.

The source uses -9999 and "unknown" as missing-value codes for some fields. The code below converts those codes to R missing values (NA). No application records were recorded for 1959 because of a pause in the program, and records for 1969 and 1980 are incomplete.

Project and Pest Importance

Mature spruce budworm larva feeding on white spruce. Photo: Government of Canada.

Mature spruce budworm larva feeding on white spruce. Photo: Government of Canada.

Adult female (upper) and male (lower) spruce budworm moths, pinned. Photo: Jean-Paul Laplante, via Government of Canada; Spruce budworm.

Adult female (upper) and male (lower) spruce budworm moths, pinned. Photo: Jean-Paul Laplante, via Government of Canada; Spruce budworm.

The eastern spruce budworm, Choristoneura fumiferana (Lepidoptera: Tortricidae), is a moth native to North America whose larvae feed on the needles and buds of spruce and fir trees. It is regarded as the most widespread and damaging native defoliator of coniferous trees in North America (Nealis, 2024).

The insect completes one life cycle per year. Larvae emerge in spring and feed almost exclusively on current-year growth such as buds and new needles, constructing silk shelters as the host’s shoots expand. Adults emerge in mid-summer and can be carried many kilometres by wind and air currents (Nealis, 2024). Budworm populations undergo large oscillations, with outbreaks recurring at roughly 35-year intervals and often lasting more than a decade in eastern Canada.

The species inhabits spruce–fir forests from east of the continental divide to the Atlantic coast, occurring in every Canadian province and territory except Nunavut. Defoliation has also been recorded as far north as central Alaska and as far south as the northeastern United States (Nealis, 2024). Repeated defoliation reduces radial growth regardless of tree species or size and can eventually kill trees (Zhang et al., 2014). Defoliation also creates secondary problems, such as attacks by other pests, infection by pathogens, and increased fire risk and habitat destruction. Apart from habitat destruction and ecological damage, spruce budworm also causes significant economic losses by affecting the timber harvest, as shown by Chang et al. (2012) to cause more than $3.3 billion in losses in New Brunswick alone. Historic outbreaks such as the 1967–1992 Quebec outbreak have shown even larger losses, with 238 million m³ of timber loss and an estimated value of $12.5 billion (MacLean, 2015).

Large-scale aerial spraying of forested land in Canada began in 1945. In New Brunswick alone, between 1952 and 1993, 97% of the province’s 6.2 million hectares of forest land received at least one insecticide application, the majority aimed at spruce budworm. These early pesticides were efficient, nonselective, and persistent, and caused serious environmental impacts, public backlash, and bioaccumulation issues (Heartz et al., 2023). The best-known, DDT, was applied from 1952 to 1968 and still persists in treated soils and adjacent water bodies, with measurable ecosystem effects today. DDT was replaced in the 1970s by fenitrothion, and later aminocarb, phosphamidon, and Bacillus thuringiensis ssp. kurstaki (Btk). These post-DDT products were nonpersistent and are unlikely to be found within the ecosystem today (Heartz et al., 2023). Btk and tebufenozide have been shown to effectively reduce spruce budworm populations (Fuentealba et al., 2019; Doucet et al., 2007), while not having negative effects on non-target caterpillars and parasitoids at low budworm caterpillar densities (Glaus et al., 2023).

In recent years, “early intervention strategies” (EIS) have been examined as a way to make these insecticides more effective while reducing their overall use. These strategies track hotspots of overwintering larvae and apply insecticide preventively, before the larvae can defoliate trees. These were found to be highly effective and further the goals of reducing pesticide use (MacLean et al., 2019). Other strategies for population control which have proven to be effective are pheromone traps which prevent mating (Roscoe et al., 2022), and the manipulation of endophytic fungi communities naturally present within spruce and fir trees (Ullah et al., 2025).

Overall, issues with classical pesticide persistence, public opinion, and increased understanding of ecological services have urged alternative strategies and chemicals to be used for population control of spruce budworm.

Dataset Analysis

Preliminary: importing and checking the data

# Load the packages used for data import, transformation, and plotting.
library(tidyverse)
library(foreign)
# Import the application records from the project's data folder.
apps <- read.csv("data/NBHistoricPesticideApplications.csv") |>
  mutate(across(where(is.character), trimws))

# Check the table structure and recorded categories.
glimpse(apps)
## Rows: 7,205
## Columns: 10
## $ app_ID    <chr> "1952_001_1", "1953_001_1", "1953_002_1", "1953_002_2", "195…
## $ app_area  <chr> "1952_001", "1953_001", "1953_002", "1953_002", "1953_003", …
## $ year      <int> 1952, 1953, 1953, 1953, 1953, 1953, 1953, 1953, 1953, 1953, …
## $ insect    <chr> "esb", "esb", "esb", "esb", "esb", "esb", "esb", "esb", "esb…
## $ act_ing   <chr> "DDT", "DDT", "DDT", "DDT", "DDT", "DDT", "DDT", "DDT", "DDT…
## $ rate_kgha <dbl> 1.12, 0.56, 0.56, 0.56, 0.56, 0.56, 0.56, 0.56, 0.56, 0.56, …
## $ vol_lha   <dbl> 9.34, 4.68, 4.68, 4.68, 4.68, 4.68, 4.68, 4.68, 4.68, 4.68, …
## $ form_type <chr> "oil", "oil", "oil", "oil", "oil", "oil", "oil", "oil", "oil…
## $ tank_mix  <dbl> 0.11915, 0.11966, 0.11966, 0.11966, 0.11966, 0.11966, 0.1196…
## $ brand     <chr> "unknown", "unknown", "unknown", "unknown", "unknown", "unkn…
count(apps, act_ing, sort = TRUE)
##                      act_ing    n
## 1               Fenitrothion 3577
## 2                        DDT 1365
## 3                         BT  808
## 4                  Aminocarb  749
## 5               Phosphamidon  581
## 6                Trichlorfon   90
## 7                  Malathion   10
## 8                 Dimethoate    9
## 9                 Methoprene    8
## 10 Fenitrothion+Phosphamidon    4
## 11               Mexacarbate    2
## 12                  Methomyl    1
## 13                 Pyrethrin    1
count(apps, form_type)
##                                             form_type    n
## 1                                             585 oil    2
## 2                                ATLOX 3409 and water  381
## 3                             CYCLOSOL 63 and 585 oil    6
## 4                DOWANOL TPM and ATLOX 3409 and water   88
## 5                    DOWANOL and ATLOX 3409 and water 2086
## 6  DOWANOL and CYCLOSOL 63 OR CYCLOSOL 63 and 585 oil  146
## 7                              Invadine JVC and water   27
## 8                                            emulsion   97
## 9                                                 oil 1719
## 10                                            panasol    3
## 11                                         summer oil    2
## 12                                          undiluted  343
## 13                                            unknown  783
## 14                                              water 1480
## 15                                    wettable powder   42
count(apps, brand, sort = TRUE)
##             brand    n
## 1         unknown 4562
## 2       Sumathion  950
## 3    Matacil 180F  517
## 4       Dipel 132  405
## 5         Matacil  223
## 6  Thuricide 48LV  121
## 7        Dimecron  116
## 8           Dylox   90
## 9   Futura XLV HP   90
## 10     Futura XLV   40
## 11       NOVABAC3   23
## 12          Dipel   16
## 13       Rogor 40    9
## 14         Futura    7
## 15  Sumathion 20F    6
## 16       CGA13353    4
## 17        GS42710    4
## 18      Noviathon    4
## 19  Sumathion mix    4
## 20      Foray 48B    3
## 21      Dipel 176    2
## 22      Foray 76B    2
## 23     Matacil wp    2
## 24        Zectran    2
## 25      Dipel 45B    1
## 26 Dipel ABG-6103    1
## 27        Lannate    1
# Replace the source's missing-value codes with R's standard missing value, NA.
apps <- apps |>
  mutate(
    across(c(rate_kgha, vol_lha), ~ na_if(.x, -9999)),
    across(c(form_type, brand), ~ na_if(.x, "unknown"))
  )

# Identify Btk's different rate units for the mass calculation.
apps <- apps |>
  mutate(unit = if_else(act_ing == "BT", "BIU/ha", "kg/ha"))

across() applies an operation to selected columns, na_if() replaces a specified code with NA, and if_else() assigns values according to a condition. See the dplyr reference.

# Read the DBF file and convert area from square metres to hectares.
areas <- read.dbf(
  "data/NBHistoricPesticideAreas.dbf",
  as.is = TRUE
) |>
  rename_with(tolower) |>
  mutate(area_ha = shape_area / 10000) |>
  select(app_area, area_ha, app_type, aircraft, guidance, app_sys, num_apps)

# Check the imported fields and recorded categories.
glimpse(areas)
## Rows: 4,545
## Columns: 7
## $ app_area <chr> "1952_001", "1953_001", "1953_002", "1953_003", "1953_004", "…
## $ area_ha  <dbl> 78954.7743, 219536.9236, 16547.0359, 320161.9453, 2888.2381, …
## $ app_type <chr> "operational", "operational", "operational", "operational", "…
## $ aircraft <chr> "Boeing Stearman", "Boeing Stearman", "Boeing Stearman", "Boe…
## $ guidance <chr> "visual maps and spray pilot ground sighting", "visual maps a…
## $ app_sys  <chr> "boom and nozzle", "boom and nozzle", "boom and nozzle", "boo…
## $ num_apps <int> 1, 1, 2, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 2, 1, 1, 1, 1, 1…
count(areas, aircraft)
##                                                                                                                         aircraft
## 1                                                                                                                Boeing Stearman
## 2                                                                                            Boeing Stearman and Grumman Avenger
## 3                                                       Boeing Stearman and Grumman Avenger and Grumman AgCat and Snow Commander
## 4                                                                                                                Grumman Avenger
## 5                                                                                  Grumman Avenger and Cessna AgTrucks and DC-6B
## 6                                                                                                      Grumman Avenger and DC-6B
## 7                                  Grumman Avenger and DC-6B and Thrush Commander and Ayres Turbo Thrush and Bell 206 helicopter
## 8                                                                              Grumman Avenger and DC-6B and other single engine
## 9                                                                                              Grumman Avenger and Grumman AgCat
## 10                                                      Grumman Avenger and Grumman AgCat and Cessna 188 and Bell 206 helicopter
## 11                         Grumman Avenger and Grumman AgCat and Cessna AgTruck and Piper Pawnee Brave and Hiller 12E helicopter
## 12 Grumman Avenger and Grumman AgCat and Cessna AgTruck and Piper Pawnee Brave and Hiller 12E helicopter and Bell 206 helicopter
## 13                                 Grumman Avenger and Grumman AgCat and Cessna AgTruck and Thrush Commander and M-18 helicopter
## 14                                                                         Grumman Avenger and Grumman AgCat and Cessna AgTrucks
## 15                                                     Grumman Avenger and Grumman AgCat and Cessna AgTrucks and M-18 helicopter
## 16      Grumman Avenger and Grumman AgCat and DC-6B and Cessna AgTruck and Thrush Commander and Piper Pawnee and M-18 helicopter
## 17                               Grumman Avenger and Grumman AgCat and Piper Pawnee and DC-6B and Cessna 188 and M-18 helicopter
## 18                                                                                          Grumman Avenger and Thrush Commander
## 19                                                                       Grumman Avenger and Thrush Commander and Cessna AgWagon
## 20                                                                           Grumman Avenger and Thrush Commander and helicopter
## 21                                                                                                                  Piper Pawnee
## 22                                                                                                                       unknown
##      n
## 1  132
## 2  724
## 3  354
## 4  430
## 5   28
## 6   68
## 7   46
## 8   79
## 9   87
## 10 307
## 11 651
## 12 571
## 13  37
## 14 117
## 15 108
## 16  80
## 17 329
## 18 120
## 19 138
## 20  75
## 21   2
## 22  62
count(areas, app_type)
##           app_type    n
## 1     experimental  311
## 2 experimental-JDI    3
## 3      operational 4231
# Check for duplicate area IDs before joining the tables.
areas |>
  count(app_area) |>
  filter(n > 1)
## [1] app_area n       
## <0 rows> (or 0-length row.names)

read.dbf() imports the treated-area table; as.is = TRUE retains its text fields as character data. shape_area is converted from square metres to hectares using 10,000 m² per hectare. The foreign package documentation describes the read.dbf() function.

# Join area and technology fields to each application by the shared area ID.
joined_data <- apps |>
  left_join(areas, by = "app_area")

# Check the joined records, unmatched areas, and year range.
joined_data |>
  summarise(
    n_applications = n(),
    n_areas        = n_distinct(app_area),
    n_missing_area = sum(is.na(area_ha)),
    year_min       = min(year, na.rm = TRUE),
    year_max       = max(year, na.rm = TRUE)
  )
##   n_applications n_areas n_missing_area year_min year_max
## 1           7205    4545              0     1952     1993

The left_join() function matches rows by the key supplied in by, while retaining all rows in the first table. See the dplyr join documentation. The duplicate-key check is important because repeated app_area values in areas could multiply rows in the joined data.

Annual mass and carrier volume

# Estimate annual chemical mass and carrier volume from rate and treated area.
amount_year <- joined_data |>
  filter(!is.na(area_ha)) |>
  mutate(
    kg = if_else(unit == "kg/ha", rate_kgha * area_ha, NA_real_),
    L  = vol_lha * area_ha
  ) |>
  group_by(year) |>
  summarise(
    total_kg = if (all(is.na(kg))) NA_real_ else sum(kg, na.rm = TRUE),
    total_L  = if (all(is.na(L))) NA_real_ else sum(L, na.rm = TRUE),
    .groups = "drop"
  )

# Check annual totals and coverage before plotting or modelling.
glimpse(amount_year)
## Rows: 41
## Columns: 3
## $ year     <int> 1952, 1953, 1954, 1955, 1956, 1957, 1958, 1960, 1961, 1962, 1…
## $ total_kg <dbl> 88429.35, 511658.28, 273536.19, 270733.25, 595504.41, 1241607…
## $ total_L  <dbl> 737437.6, 4276001.3, 2285981.0, 2262556.4, 4976715.4, 1037629…
# Check which years are absent from the annual summary.
setdiff(1952:1993, amount_year$year)
## [1] 1959
# Review the years with a documented pause or incomplete records.
amount_year |>
  filter(year %in% c(1959, 1969, 1980))
## # A tibble: 2 × 3
##    year total_kg  total_L
##   <int>    <dbl>    <dbl>
## 1  1969  595087. 3147791.
## 2  1980  605612. 2385209.

group_by() groups records by year and summarise() creates one row per year. The conditional sums retain NA when every value for that annual measure is missing, rather than presenting that year as a zero total. See the dplyr reference and R’s ?sum help page.

Model fitting and comparison

# Use the diagnostic helper from the submitted Chapter 6 Walkthrough.
diag_plots <- function(model) {
  op <- par(mfrow = c(2, 2), mar = c(4, 4, 2, 1))
  on.exit(par(op))
  plot(model)
}

The diagnostic helper is adapted from the Chapter 6 course walkthrough. The plot.lm() method produces diagnostic plots for linear-model objects; see the R documentation.

# Keep positive finite annual totals because log(0) is undefined.
# Center year so the polynomial terms are easier to fit numerically.
model_data <- amount_year |>
  filter(total_kg > 0, is.finite(total_kg)) |>
  mutate(year_c = year - mean(year))

# Fit polynomial models of increasing complexity to log annual mass.
# Each model remains linear in its estimated coefficients.
m_linear <- lm(log(total_kg) ~ year_c, data = model_data)

m_quadratic <- lm(
  log(total_kg) ~ year_c + I(year_c^2),
  data = model_data
)

m_cubic <- lm(
  log(total_kg) ~ year_c + I(year_c^2) + I(year_c^3),
  data = model_data
)

m_quartic <- lm(
  log(total_kg) ~ year_c + I(year_c^2) + I(year_c^3) + I(year_c^4),
  data = model_data
)

# Calculate PRESS using the method shown in the Week 6.1 model-selection lecture.
calc_press <- function(model) {
  predictive_residuals <-
    residuals(model) / (1 - lm.influence(model)$hat)

  sum(predictive_residuals^2)
}

# Compare candidate models fitted to the same annual observations.
model_candidates <- list(
  Linear = m_linear,
  Quadratic = m_quadratic,
  Cubic = m_cubic,
  Quartic = m_quartic
)

model_comparison <- tibble(
  model = names(model_candidates),
  r_squared = map_dbl(model_candidates, ~ summary(.x)$r.squared),
  adjusted_r_squared = map_dbl(
    model_candidates,
    ~ summary(.x)$adj.r.squared
  ),
  AIC = map_dbl(model_candidates, AIC),
  PRESS = map_dbl(model_candidates, calc_press)
) |>
  mutate(delta_AIC = AIC - min(AIC)) |>
  arrange(AIC)

knitr::kable(
  model_comparison,
  digits = 3,
  caption = "Fit and prediction-error statistics for the candidate polynomial models."
)
Fit and prediction-error statistics for the candidate polynomial models.
model r_squared adjusted_r_squared AIC PRESS delta_AIC
Quartic 0.520 0.467 94.884 25.169 0.000
Cubic 0.486 0.444 95.753 25.748 0.869
Quadratic 0.396 0.364 100.349 27.077 5.465
Linear 0.147 0.125 112.499 36.314 17.614
# Select the candidate with the lowest AIC.
# Review PRESS and diagnostic plots before interpreting the selected fit.
final_name <- model_comparison$model[1]
m_final <- model_candidates[[final_name]]

# Display the baseline and selected-model summaries.
summary(m_linear)
## 
## Call:
## lm(formula = log(total_kg) ~ year_c, data = model_data)
## 
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -2.5296 -0.6023  0.1205  0.4359  1.7827 
## 
## Coefficients:
##             Estimate Std. Error t value Pr(>|t|)    
## (Intercept) 12.56789    0.14200  88.509   <2e-16 ***
## year_c      -0.03047    0.01175  -2.592   0.0133 *  
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 0.9092 on 39 degrees of freedom
## Multiple R-squared:  0.147,  Adjusted R-squared:  0.1251 
## F-statistic: 6.721 on 1 and 39 DF,  p-value: 0.01335
summary(m_final)
## 
## Call:
## lm(formula = log(total_kg) ~ year_c + I(year_c^2) + I(year_c^3) + 
##     I(year_c^4), data = model_data)
## 
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -1.8023 -0.3302  0.1494  0.5023  1.2696 
## 
## Coefficients:
##               Estimate Std. Error t value Pr(>|t|)    
## (Intercept)  1.291e+01  2.060e-01  62.666  < 2e-16 ***
## year_c       2.380e-02  2.302e-02   1.034  0.30808    
## I(year_c^2)  8.235e-04  2.996e-03   0.275  0.78501    
## I(year_c^3) -2.182e-04  7.954e-05  -2.744  0.00941 ** 
## I(year_c^4) -1.230e-05  7.616e-06  -1.615  0.11495    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 0.7095 on 36 degrees of freedom
## Multiple R-squared:  0.5205, Adjusted R-squared:  0.4672 
## F-statistic: 9.768 on 4 and 36 DF,  p-value: 1.864e-05
# Create prediction years across the observed study period.
year_grid <- seq(
  min(model_data$year),
  max(model_data$year),
  length.out = 200
)

prediction_grid <- data.frame(
  year = year_grid,
  year_c = year_grid - mean(model_data$year)
)

# Predict log mass for the linear baseline.
linear_predictions <- data.frame(
  year = prediction_grid$year,
  fit = as.numeric(predict(m_linear, newdata = prediction_grid))
)

# Predict log mass and confidence limits for the selected model.
final_prediction_matrix <- predict(
  m_final,
  newdata = prediction_grid,
  interval = "confidence"
)

final_predictions <- data.frame(
  year = prediction_grid$year,
  fit = final_prediction_matrix[, "fit"],
  lwr = final_prediction_matrix[, "lwr"],
  upr = final_prediction_matrix[, "upr"]
)

# Convert selected-model predictions back to kilograms for Figure 1.
final_predictions$fitted_kg <- exp(final_predictions$fit)

linear_label <- paste0(
  "Linear baseline (R² = ",
  round(summary(m_linear)$r.squared, 3),
  ")"
)

final_label <- paste0(
  final_name,
  " (R² = ",
  round(summary(m_final)$r.squared, 3),
  ")"
)

The Week 6.1 lecture presents adjusted R², AIC, and PRESS as model-comparison measures. AIC and PRESS should be considered alongside the diagnostic plots rather than maximizing ordinary R².

ggplot(model_data, aes(x = year, y = log(total_kg))) +
  geom_point() +
  geom_ribbon(
    data = final_predictions,
    aes(x = year, ymin = lwr, ymax = upr),
    inherit.aes = FALSE,
    fill = "#D97928",
    alpha = 0.18
  ) +
  geom_line(
    data = linear_predictions,
    aes(x = year, y = fit, colour = linear_label),
    linewidth = 1
  ) +
  geom_line(
    data = final_predictions,
    aes(x = year, y = fit, colour = final_label),
    linewidth = 1
  ) +
  scale_colour_manual(
    name = "Model",
    values = setNames(
      c("#3778BF", "#D97928"),
      c(linear_label, final_label)
    )
  ) +
  labs(
    x = "Year",
    y = "Log of annual recorded chemical mass (kg)",
    title = "Linear baseline and AIC-selected polynomial model"
  )
**Figure 0:** Annual recorded chemical mass for years with positive finite totals, plotted on the log scale used for modelling. The blue line shows the linear baseline and the orange line shows the model with the lowest AIC among the candidate polynomial models. In the plotted comparison, the quartic curve follows the mid-period rise and late-period decline more closely than the linear trend, while showing a sharp drop at the end of the record.

Figure 0: Annual recorded chemical mass for years with positive finite totals, plotted on the log scale used for modelling. The blue line shows the linear baseline and the orange line shows the model with the lowest AIC among the candidate polynomial models. In the plotted comparison, the quartic curve follows the mid-period rise and late-period decline more closely than the linear trend, while showing a sharp drop at the end of the record.

ggplot(amount_year, aes(x = year, y = total_kg)) +
  geom_col(fill = "grey75") +
  geom_line(
    data = final_predictions,
    aes(x = year, y = fitted_kg),
    colour = "firebrick",
    linewidth = 1
  ) +
  scale_y_continuous(labels = scales::comma) +
  labs(
    x = "Year",
    y = "Active ingredient applied (kg)",
    title = "Total chemical insecticide applied per year, New Brunswick (1952–1993)"
  )
**Figure 1:** Annual mass of recorded chemical active ingredients applied in New Brunswick, 1952–1993. Bars show annual totals in kilograms; the red curve shows fitted values from the model with the lowest AIC among the candidate polynomial models, transformed back to kilograms. The quartic fit captures a mid-period plateau and subsequent decline, while annual totals retain substantial variation. Btk is excluded because its application rate is recorded in BIU/ha.

Figure 1: Annual mass of recorded chemical active ingredients applied in New Brunswick, 1952–1993. Bars show annual totals in kilograms; the red curve shows fitted values from the model with the lowest AIC among the candidate polynomial models, transformed back to kilograms. The quartic fit captures a mid-period plateau and subsequent decline, while annual totals retain substantial variation. Btk is excluded because its application rate is recorded in BIU/ha.

cat(
  "The linear baseline has R² = ",
  round(summary(m_linear)$r.squared, 3),
  " and adjusted R² = ",
  round(summary(m_linear)$adj.r.squared, 3),
  ". Among the candidate models, the lowest-AIC model is **",
  final_name,
  "** (AIC = ",
  round(AIC(m_final), 2),
  "; R² = ",
  round(summary(m_final)$r.squared, 3),
  "; adjusted R² = ",
  round(summary(m_final)$adj.r.squared, 3),
  "; PRESS = ",
  round(calc_press(m_final), 2),
  "). The comparison table reports fit and prediction-error statistics for all candidates. Figure 0 compares the selected curve with the linear baseline, and Figure 2 shows diagnostics for the selected model.",
  sep = ""
)

The linear baseline has R² = 0.147 and adjusted R² = 0.125. Among the candidate models, the lowest-AIC model is Quartic (AIC = 94.88; R² = 0.52; adjusted R² = 0.467; PRESS = 25.17). The comparison table reports fit and prediction-error statistics for all candidates. Figure 0 compares the selected curve with the linear baseline, and Figure 2 shows diagnostics for the selected model.

# Apply the diagnostic helper from the Chapter 6 Walkthrough.
diag_plots(m_final)
**Figure 2:** Residual-versus-fitted, Normal Q–Q, Scale–Location, and residuals-versus-leverage diagnostic plots for the model with the lowest AIC. The quartic model shows a curved residual smoother, tail departures in the Q–Q plot, and high-leverage observations, while the Scale–Location smoother is comparatively level.

Figure 2: Residual-versus-fitted, Normal Q–Q, Scale–Location, and residuals-versus-leverage diagnostic plots for the model with the lowest AIC. The quartic model shows a curved residual smoother, tail departures in the Q–Q plot, and high-leverage observations, while the Scale–Location smoother is comparatively level.

ggplot(amount_year, aes(x = year, y = total_L)) +
  geom_col(fill = "steelblue3") +
  scale_y_continuous(labels = scales::comma) +
  labs(
    x = "Year",
    y = "Carrier volume (L)",
    title = "Annual recorded spray-mix carrier volume"
  )
## Warning: Removed 1 row containing missing values or values outside the scale range
## (`geom_col()`).
**Figure 3:** Annual recorded spray-mix carrier volume applied in New Brunswick, 1952–1993, in litres. Carrier volume peaks in 1957 and 1976 and is generally lower in the later years. No carrier-volume values are available for 1974, although application records are present for that year.

Figure 3: Annual recorded spray-mix carrier volume applied in New Brunswick, 1952–1993, in litres. Carrier volume peaks in 1957 and 1976 and is generally lower in the later years. No carrier-volume values are available for 1974, although application records are present for that year.

# Count application records for each active ingredient in each year.
comp <- joined_data |>
  count(year, act_ing)
ggplot(comp, aes(x = year, y = n, fill = act_ing)) +
  geom_col() +
  labs(
    x = "Year",
    y = "Number of application records",
    fill = "Active ingredient",
    title = "Recorded active ingredients by year"
  )
**Figure 4:** Number of recorded application records per year by active ingredient, 1952–1993. DDT dominates the early records, fenitrothion is prominent in the 1970s and early 1980s, and Btk becomes prominent in the later years.

Figure 4: Number of recorded application records per year by active ingredient, 1952–1993. DDT dominates the early records, fenitrothion is prominent in the 1970s and early 1980s, and Btk becomes prominent in the later years.

# Calculate each ingredient's share of application records within each year.
comp_prop <- comp |>
  group_by(year) |>
  mutate(proportion = n / sum(n)) |>
  ungroup()
ggplot(comp_prop, aes(x = year, y = proportion, fill = act_ing)) +
  geom_col() +
  scale_y_continuous(labels = scales::percent) +
  labs(
    x = "Year",
    y = "Proportion of application records",
    fill = "Active ingredient",
    title = "Relative composition of recorded application records"
  )
**Figure 5:** Annual composition of recorded application records by active ingredient, 1952–1993. The records shift from DDT in the early period to greater representation of fenitrothion and, later, Btk.

Figure 5: Annual composition of recorded application records by active ingredient, 1952–1993. The records shift from DDT in the early period to greater representation of fenitrothion and, later, Btk.

# Technology is recorded for treated areas rather than individual applications.
# Keep one row per area, year, and technology combination to avoid repeated counts.
tech <- joined_data |>
  distinct(app_area, year, app_type, aircraft, guidance, app_sys, area_ha) |>
  mutate(
    app_type = if_else(
      str_detect(app_type, "^experimental"),
      "experimental",
      app_type
    ),
    across(c(aircraft, guidance, app_sys), ~ na_if(.x, "unknown"))
  )

# Check the number of treated-area records with guidance data by year.
tech |>
  filter(!is.na(guidance)) |>
  count(year)
##    year   n
## 1  1952   1
## 2  1953  21
## 3  1954   4
## 4  1955  42
## 5  1956  29
## 6  1957  35
## 7  1958  10
## 8  1960  15
## 9  1961  68
## 10 1962  58
## 11 1963  51
## 12 1964  65
## 13 1965 457
## 14 1966 193
## 15 1967 161
## 16 1968  56
## 17 1969  31
## 18 1970  58
## 19 1971  54
## 20 1972  98
## 21 1973 101
## 22 1974  68
## 23 1975 138
## 24 1976 119
## 25 1977  75
## 26 1978  49
## 27 1979  71
## 28 1980  46
## 29 1981  80
## 30 1982 108
## 31 1983 329
## 32 1984 571
## 33 1985 651
## 34 1986 307
# Simplify the guidance labels for the plot legend.
guid_lab <- c(
  "visual maps and spray pilot ground sighting"         = "Visual maps + ground sighting",
  "visual maps or FPL pointer system"                   = "Visual maps or FPL pointer",
  "visual maps or FPL pointer in Cessna 185 with radio" = "Visual maps/FPL + radio",
  "FPL pointer system"                                  = "FPL pointer only",
  "FPL pointer system or modified FPL system"           = "FPL or modified FPL",
  "modified FPL system"                                 = "Modified FPL only",
  "FPL pointer system or Litton for DC-6"               = "FPL pointer + Litton (DC-6)"
)

# Count treated-area records by year and guidance method.
fig_guidance <- tech |>
  filter(!is.na(guidance)) |>
  mutate(guidance = coalesce(guid_lab[guidance], guidance)) |>
  count(year, guidance)

ggplot(fig_guidance, aes(x = year, y = n, fill = guidance)) +
  geom_col() +
  labs(
    x = "Year",
    y = "Number of treated-area records",
    fill = "Guidance system",
    title = "Spray-guidance technology by year, 1952–1993"
  )
**Figure 6:** Counts of treated-area records by recorded spray-guidance system and year, 1952–1993. Early records include visual-map methods; FPL pointer systems are prominent in later records, followed by modified and combined systems.

Figure 6: Counts of treated-area records by recorded spray-guidance system and year, 1952–1993. Early records include visual-map methods; FPL pointer systems are prominent in later records, followed by modified and combined systems.

# Simplify the delivery-system labels for the plot legend.
sys_lab <- c(
  "boom and nozzle"                                   = "Boom and nozzle",
  "boom and nozzle or Micronair"                      = "Boom/nozzle or Micronair",
  "boom and nozzle or Micronair AU4000 or Sergonairs" = "Boom/nozzle, Micronair or Sergonairs",
  "Micronair AU4000"                                  = "Micronair AU4000",
  "Micronair"                                         = "Micronair"
)

# Count treated-area records by year and application system.
fig_appsys <- tech |>
  filter(!is.na(app_sys)) |>
  mutate(app_sys = coalesce(sys_lab[app_sys], app_sys)) |>
  count(year, app_sys)

ggplot(fig_appsys, aes(x = year, y = n, fill = app_sys)) +
  geom_col() +
  labs(
    x = "Year",
    y = "Number of treated-area records",
    fill = "Application system",
    title = "Spray-delivery technology by year, 1952–1993"
  )
**Figure 7:** Counts of treated-area records by recorded spray-delivery system and year, 1952–1993. Boom-and-nozzle systems dominate many earlier records, while Micronair and combined-system categories appear in later years.

Figure 7: Counts of treated-area records by recorded spray-delivery system and year, 1952–1993. Boom-and-nozzle systems dominate many earlier records, while Micronair and combined-system categories appear in later years.

Results and Conclusions

Data coverage and provenance

The joined dataset contains 7,205 insecticide applications across 4,545 treated areas, spanning 1952–1993. Each application matched a treated-area record. The dataset includes 4,231 operational areas, 311 experimental areas, and 3 experimental-JDI areas. No applications were recorded for 1959, when the program was paused, and records for 1969 and 1980 are incomplete (TBM Avengers Archive, 2010). Annual values therefore describe the digitised records available for each year.

Insecticide application over time

There is a decline in recorded chemical application mass over the 1952–1993 period (Figure 1). Annual totals vary considerably, with prominent peaks in 1957 and 1976 and generally lower totals toward the end of the period. Btk is excluded from this mass calculation because its application rate is recorded in BIU/ha rather than kg/ha. The mass trend therefore describes ingredients measured in kilograms, while the Btk records contribute to the separate application-record and carrier-volume summaries.

The linear baseline has R² = 0.147 and adjusted R² = 0.125. Among the candidate models, the lowest-AIC model is Quartic (AIC = 94.88, R² = 0.52, adjusted R² = 0.467, PRESS = 25.17). The selected model has a higher R² than the linear baseline and follows a curved temporal pattern. Figure 0 shows that the quartic fit turns sharply downward near the end of the series, which may present a flaw with the model rather than real data trends. Regardless, it is chosen here as the best fit due to fit to data and visual trends.

The diagnostics in Figure 2 show a curved residuals-versus-fitted smoother, with a dip and rise among the cluster of fitted values near the upper end. The Q–Q points depart from the reference line at the tails, particularly at the lower end. The Scale–Location smoother is comparatively level, although the points are tightly grouped at high fitted values. Observations 41 and 1 have high leverage, and observation 16 has a large negative standardized residual. These features indicate that influential observations and residual departures remain in the selected model.

Figure 3 shows annual spray-mix volume in litres. The largest peaks occur in 1957 and 1976, with lower recorded volumes in the later years. Carrier-volume records are missing for all 68 applications recorded in 1974, so that year is omitted from the volume plot; only chemical-mass records are available for 1974.

Changes in active ingredients

Figures 4 and 5 show changes in recorded active ingredients over time. Figure 4 gives the number of application records by ingredient; Figure 5 shows each ingredient’s share of records within its year. The records show DDT dominating the early period, followed by increased representation of phosphamidon and fenitrothion. Btk becomes prominent in the later years. A small number of records identify a fenitrothion–phosphamidon tank mixture. These plots describe recorded application records by ingredient.

Evolution of application and guidance technologies

Figures 6 and 7 show counts of treated-area records by guidance and delivery system, respectively. Early guidance records include visual maps and pilot ground-sighting; FPL pointer systems become prominent in later records, followed by modified or combined systems. Delivery records are dominated by boom-and-nozzle systems in many earlier years, with Micronair and combined-system categories appearing later. Some technology fields are missing or unknown, particularly in earlier years, so the figures describe the records available for each period.

Spraying records and budworm populations

Historical budworm-population data are difficult to compare directly with the application records because of differences in treatment design, location, and timescale. Gray and MacKinnon (2007) examined spruce budworm defoliation and insecticide protection in New Brunswick from 1965 to 1992, providing relevant context.

Gray and MacKinnon (2007) reported that protection peaked from 1976 to 1982, overlapping with a peak in recorded insecticide application in 1976. This overlap is consistent with reactive spraying following severe defoliation in 1974–1975, as discussed by Gray and MacKinnon, but does not establish a causal relationship. MacLean et al. (2019) discuss early intervention strategies as a means of reducing the need for repeated pesticide use.

Overall conclusions

The New Brunswick records show changes in recorded insecticide mass, carrier volume, active ingredients, and application technology between 1952 and 1993. The quartic model was selected over the linear baseline by AIC among the four candidate polynomial models, and its higher R² reflects its more flexible fit. However, the quartic model’s sharp decline near the end of the record and the diagnostic evidence of influential observations and tail departures warrant caution. The exclusion of Btk from the mass calculation, incomplete or absent records in some years, and the lack of direct measures of budworm populations or forest damage limit conclusions about total insecticide use and program effectiveness.

Glossary

Active ingredient - The chemical that produces an insecticidal effect, typically diluted by a carrier such as water.

**Aminocarb chemical structure**, source: PubChem

Aminocarb chemical structure, source: PubChem

Aminocarb (C₁₁H₁₆N₂O₂) - Carbamate class insecticide introduced in 1956 with low toxicity to mammals.

BIU/ha - Billion international units per hectare, a measure of potency.

**Boom-and-nozzle mechanism**, source: Yonas (2021).

Boom-and-nozzle mechanism, source: Yonas (2021).

Boom-and-nozzle - The conventional system of aerial pesticide application consisting of a boom (long bar) and hydraulic nozzle that forces liquid through an orifice to form droplets. Droplet size depends on nozzle type, orifice size, and pressure.

Btk - Bacillus thuringiensis var. kurstaki, a bacterial insecticide which damages larvae of Lepidoptera when Cry crystals react with the alkaline environment of the insect’s stomach to form toxins that damage the gut.

Carbamate pesticides - Synthetic organic compounds derived from carbamic acid. They control pest populations by inhibiting acetylcholinesterase, disrupting nervous-system functions such as nerve signalling. This process is reversible, distinguishing carbamates from organophosphate pesticides.

**DDT chemical structure**, source: PubChem.

DDT chemical structure, source: PubChem.

DDT - An organochlorine insecticide first synthesized in 1874. It is slightly to moderately toxic to many classes of animals, including mammals, and can cause reproductive damage, tumours, and sometimes death.

**Dimethoate chemical structure**, source: PubChem.

Dimethoate chemical structure, source: PubChem.

Dimethoate (C₅H₁₂NO₃PS₂) - A white crystalline organophosphate pesticide. Production ceased in 1982; the United States government classifies it as an “Extremely Hazardous Substance.”

**Fenitrothion chemical structure**, source: PubChem.

Fenitrothion chemical structure, source: PubChem.

Fenitrothion (C₉H₁₂NO₅PS) - A brownish-yellow oil and organophosphate pesticide used as a selective acaricide and against insects and other herbivorous pests.

FPL pointer system - Forest Protection Limited’s navigation system used by pilots in the program.

Litton (DC-6) - An inertial navigation system fitted to Douglas DC-6 spray tankers. It determines aircraft position using accelerometers and gyroscopes.

**Malathion chemical structure**, source: PubChem.

Malathion chemical structure, source: PubChem.

Malathion (C₁₀H₁₉O₆PS₂) - A colourless liquid organophosphate pesticide used to control many insects, including head lice and fleas.

**Methomyl chemical structure**, source: PubChem.

Methomyl chemical structure, source: PubChem.

Methomyl (C₅H₁₀N₂O₂S) - A white crystalline carbamate pesticide used as a nematocide and on selected crops, including tobacco, cotton, soy, corn, alfalfa, and vegetables.

**Methoprene chemical structure**, source: PubChem.

Methoprene chemical structure, source: PubChem.

Methoprene (C₁₉H₃₄O₃) - A synthetic mimic of juvenile hormone and a growth regulator that disrupts insect development.

**Mexacarbate chemical structure**, source: PubChem.

Mexacarbate chemical structure, source: PubChem.

Mexacarbate (C₁₂H₁₈N₂O₂) - A white crystalline carbamate pesticide that is toxic on contact and when ingested.

**Micronair mechanism**, source: Micron Sprayers.

Micronair mechanism, source: Micron Sprayers.

Micronair - A wind-driven rotary atomizer that spins liquid off a rotating cage to produce a narrow, controlled droplet spectrum. It was introduced to aerial work in the late 1950s.

Organophosphate pesticides - Organophosphates inhibit acetylcholinesterase and disrupt the nervous system. Unlike carbamate effects, this inhibition is not reversible.

**Phosphamidon chemical structure**, source: PubChem.

Phosphamidon chemical structure, source: PubChem.

Phosphamidon (C₁₀H₁₉ClNO₅P) - A pale yellow liquid organophosphate pesticide used specifically as an acaricide.

**Pyrethrin chemical structure**, source: PubChem.

Pyrethrin chemical structure, source: PubChem.

Pyrethrin (C₄₃H₅₆O₈) - A natural insecticide extracted from Chrysanthemum flowers that acts as a contact poison.

Sergonair - A high-speed atomizer produced by SERG International. Public information about the product is limited.

Spray-mix volume - The total amount of diluted liquid applied per unit area.
**Tebufenozide chemical structure**, source: PubChem.

Tebufenozide chemical structure, source: PubChem.

Tebufenozide (C₂₂H₂₈N₂O₂) - A carbohydrazide that triggers premature molting in caterpillars by acting as an agonist of ecdysone.

**Trichlorfon chemical structure**, source: PubChem.

Trichlorfon chemical structure, source: PubChem.

Trichlorfon (C₄H₈Cl₃O₄P) - A white crystalline organophosphate pesticide used as a prodrug. Once absorbed or ingested, it is activated into dichlorvos (DDVP), which is commonly used for indoor pest control.

References

Peer-reviewed sources

Chang, W.-Y., Lantz, V. A., Hennigar, C. R., & MacLean, D. A. (2012). Economic impacts of forest pests: A case study of spruce budworm outbreaks and control in New Brunswick, Canada. Canadian Journal of Forest Research, 42(3), 490–505. https://doi.org/10.1139/x11-190

Doucet, D., Frisco, C., Cusson, M., Bauce, É., Palli, S. R., Tomkins, B., Arif, B., & Retnakaran, A. (2007). Diapause disruption with tebufenozide for early-instar control of the spruce budworm, Choristoneura fumiferana. Pest Management Science, 63(8), 730–736. https://doi.org/10.1002/ps.1400

Fuentealba, A., Alain, D., Hébert, C., Berthiaume, R., Quezada-García, R., & Bauce, É. (2019). Comparing the efficacy of various aerial spraying scenarios using Bacillus thuringiensis to protect trees from spruce budworm defoliation. Forest Ecology and Management, 432, 1013–1021. https://doi.org/10.1016/j.foreco.2018.10.034

Glaus, V., Nisole, A., Edwards, S., Bélanger, S., Johns, R. C., Djoumad, A., Cusson, M., Fournier, V., & Martel, V. (2023). Nontarget impacts of insecticide-based population control of eastern spruce budworm (Lepidoptera: Tortricidae) on nontarget caterpillar communities and parasitism. The Canadian Entomologist, 155. https://doi.org/10.4039/tce.2022.47

Gray, D. R., & MacKinnon, W. E. (2007). Historical spruce budworm defoliation records adjusted for insecticide protection in New Brunswick, 1965–1992. Journal of the Acadian Entomological Society, 3, 1–6. https://www.acadianes.org/journal/papers/graymackinnon.pdf

Heartz, S., MacLean, D. A., Johns, R. C., Carleton, D., Amos-Binks, L., Anderson, Q., Yu, Y., Lewis, A. D., Lavigne, D., Burant, J. B., & Edge, C. B. (2023). Historical pesticide applications for the treatment of eastern spruce budworm infestations in New Brunswick. Ecology, 104(7). https://doi.org/10.1002/ecy.4068

MacLean, D. (2015, August 15). The economic reality of spruce budworm. Healthy Forest Partnership. https://healthyforestpartnership.ca/the-economic-reality-of-spruce-budworm-august-6-2015/

MacLean, D. A., Amirault, P., Amos-Binks, L., Carleton, D., Hennigar, C., Johns, R., & Régnière, J. (2019). Positive results of an early intervention strategy to suppress a spruce budworm outbreak after five years of trials. Forests, 10(5), 448. https://doi.org/10.3390/f10050448

Nealis, V. G. (2024). Spruce budworm. In J. P. Brandt, B. I. Daigle, J.-L. St-Germain, A. C. Skinner, B. C. Callan, & V. G. Nealis (Eds.), Trees, insects, mites, and diseases of Canada’s forests. Natural Resources Canada, Canadian Forest Service.

Roscoe, L. E., MacKinnon, W., Régnière, J., Forbes, G., Brophy, M., & Lamb, R. (2022). Use of a sprayable sex pheromone formulation in landscape-level control of Choristoneura fumiferana populations. Insects, 13(12), 1175. https://doi.org/10.3390/insects13121175

Ullah, A., Shah, A., Chen, S. E., Shah, A., Rodriguez-Ramos, J. C., Zaman, R., & Erbilgin, N. (2025). Alliance between conifer trees and endophytic fungi against insect defoliators. Plant, Cell & Environment, 48(7), 5236–5249. https://doi.org/10.1111/pce.15503

Zhang, X., Lei, Y., Ma, Z., Kneeshaw, D., & Peng, C. (2014). Insect-induced tree mortality of boreal forests in eastern Canada under a changing climate. Ecology and Evolution, 4(12), 2384–2394. https://doi.org/10.1002/ece3.988

Industry sources and other references

Bird, L. (2025, November 23). Go inside the outbreak ravaging Gros Morne’s forests. CBC News. https://www.cbc.ca/news/canada/newfoundland-labrador/spruce-budworm-outbreak-atlantic-voice-9.6983204

Health Canada. (2009, January 13). Archived: Bacillus thuringiensis subspecies kurstaki – Btk. https://www.canada.ca/en/health-canada/services/consumer-product-safety/reports-publications/pesticides-pest-management/fact-sheets-other-resources/bacillus-thuringiensis-subspecies-kurstaki.html

LN-3 inertial navigation system pilot’s handbook. (1963). Litton Systems Inc.

Micron Sprayers. (2023, June 14). Micronair AU4000 rotary atomiser – specialist spraying equipment. https://www.microngroup.com/micronair-au4000-rotary-atomiser/

National Center for Biotechnology Information. (2026a). PubChem compound summary for CID 25750, phosphamidon. PubChem. https://pubchem.ncbi.nlm.nih.gov/compound/Phosphamidon

National Center for Biotechnology Information. (2026b). PubChem compound summary for CID 3082, dimethoate. PubChem. https://pubchem.ncbi.nlm.nih.gov/compound/Dimethoate

National Center for Biotechnology Information. (2026c). PubChem compound summary for CID 31200, fenitrothion. PubChem. https://pubchem.ncbi.nlm.nih.gov/compound/Fenitrothion

National Center for Biotechnology Information. (2026d). PubChem compound summary for CID 4004, malathion. PubChem. https://pubchem.ncbi.nlm.nih.gov/compound/Malathion

National Center for Biotechnology Information. (2026e). PubChem compound summary for CID 4109, methomyl. PubChem. https://pubchem.ncbi.nlm.nih.gov/compound/Methomyl

National Center for Biotechnology Information. (2026f). PubChem compound summary for CID 5281555, pyrethrin II. PubChem. https://pubchem.ncbi.nlm.nih.gov/compound/Pyrethrin-II

National Center for Biotechnology Information. (2026g). PubChem compound summary for CID 5366546, methoprene. PubChem. https://pubchem.ncbi.nlm.nih.gov/compound/Methoprene

National Center for Biotechnology Information. (2026h). PubChem compound summary for CID 91773, tebufenozide. PubChem. https://pubchem.ncbi.nlm.nih.gov/compound/Tebufenozide

National Center for Biotechnology Information. (2026i). PubChem compound summary for CID 9414, mexacarbate. PubChem. https://pubchem.ncbi.nlm.nih.gov/compound/Mexacarbate

National Pesticide Information Center. (1999). DDT (general fact sheet) [Pesticide fact sheet]. U.S. Environmental Protection Agency. https://npic.orst.edu/factsheets/ddtgen.pdf

Riley, C. M. (1991). Combined effects of aircraft type and atomizer on the deposit of a simulated insecticide spray. https://www.serginternational.org/reports/1991_01_1991.htm

TBM Avengers Archive. (2010, December 20). Aerial spray program – 1958 to 1993. https://forestprotectiontbmavengers.wordpress.com/history-of-the-spray-program-in-new-brunswick-2/tbm-teams-and-pilots-in-the-aerial-spray-program-1958-to-1993/

Yonas, M. (2021). Fabrication and performance evaluation of ground wheel operated boom sprayer. International Journal of Engineering Research, 6(3), 16–23.

R, packages, and course materials

Meskin, C. (2026). Chapter 6 Walkthrough [Course R Markdown file].

Meskin, C. (2026). Week 6.1 Model Fit & Selection [Course lecture slides].

R Core Team. (n.d.). AIC: Akaike’s An Information Criterion. R stats package documentation. https://stat.ethz.ch/R-manual/R-devel/library/stats/html/AIC.html

R Core Team. (n.d.). Fitting linear models: lm. R stats package documentation. https://stat.ethz.ch/R-manual/R-devel/library/stats/html/lm.html

R Core Team. (n.d.). Plot diagnostics for an lm object: plot.lm. R stats package documentation. https://stat.ethz.ch/R-manual/R-devel/library/stats/html/plot.lm.html

R Core Team. (n.d.). Summarizing linear model fits: summary.lm. R stats package documentation. https://stat.ethz.ch/R-manual/R-devel/library/stats/html/summary.lm.html

R Core Team. (n.d.). Influence measures for linear models. R stats package documentation. https://stat.ethz.ch/R-manual/R-devel/library/stats/html/lm.influence.html

dplyr. (n.d.). dplyr function reference. https://dplyr.tidyverse.org/reference/index.html

foreign. (n.d.). read.dbf: Read a DBF file. R package documentation. https://cran.r-project.org/web/packages/foreign/refman/foreign.html

GenAI statement

Generative AI (Gemini) was used in this project to provide code for CSS formatting of images and to assist with troubleshooting when a section of code was not working after attempts by the author to consult official R documentation. Gemini suggested corrected code that was tested and annotated by the author. Gemini was also used to provide formatting guidance for the glossary. Kagi “Quick” was used to reorganize parts of the Importance and Results sections and check annotation consistency. Proposed changes were implemented only when the author determined they were relevant. The author reviewed the final work, and the preliminary prose that was reorganized was written by the author.