# A tibble: 246 × 14
day month year temperature rh ws rain ffmc dmc dc isi bui
<chr> <chr> <chr> <chr> <chr> <chr> <chr> <chr> <chr> <chr> <chr> <chr>
1 01 06 2012 29 57 18 0 65.7 3.4 7.6 1.3 3.4
2 02 06 2012 29 61 13 1.3 64.4 4.1 7.6 1 3.9
3 03 06 2012 26 82 22 13.1 47.1 2.5 7.1 0.3 2.7
4 04 06 2012 25 89 13 2.5 28.6 1.3 6.9 0 1.7
5 05 06 2012 27 77 16 0 64.8 3 14.2 1.2 3.9
6 06 06 2012 31 67 14 0 82.6 5.8 22.2 3.1 7
7 07 06 2012 33 54 13 0 88.2 9.9 30.5 6.4 10.9
8 08 06 2012 30 73 15 0 86.6 12.1 38.3 5.6 13.5
9 09 06 2012 25 88 13 0.2 52.9 7.9 38.8 0.4 10.5
10 10 06 2012 28 79 12 0 73.2 9.5 46.3 1.3 12.6
# ℹ 236 more rows
# ℹ 2 more variables: fwi <chr>, classes <chr>Statistical Analysis of Environmental Data
CIFAG Webinar
2026-04-11
About me
- Founder, EU StudyAssist
- Data scientist
- Educator
How I Got Here
- 2014-2019: Bachelors - Forestry
- 2020-2021 : Teaching Assistant
- 2021-2023: MSc. Forestry
- 2023-: Data Science Consultancy
- 2024-: EU StudyAssist
Important
Visit www.eustudyassist.com to know more about EU StudyAssist
What Is This Talk About?
In this talk we will …
- go through the typical life cycle of environmental data.
- do some data exploration for environmental data, check trends, relationship and more.
- model relationships with Linear Regression.
- diagnose model reliability.
Important
While R is used in this talk, the focus is not just on the statistical tool or a single technique.
Environmental Data Characteristics
Environmental datasets are unique and often challenging:
- Expensive to acquire: inventory data; crop measurements
- Missing Data: Sensors fail; weather happens; inaccessible!
- Measurement Bias: Different instruments, different results.
- Temporal Dependence: Observations can be time-series (e.g., daily sensor data).
- Spatial Autocorrelation: Sites closer together tend to share similar properties.
The Data
- Algeria wildfire dataset.
- Occurrence of wild fire
- Estimate the danger of wildfire occurring.
variables of interest
- occurrence of a wildfire during the summer (fire/no fire)
- FWI of forest
Note
Forest Weather Index (FWI) is a global index that estimates wildfire danger by calculating fuel moisture and fire behavior based on temperature, relative humidity, wind speed, and precipitation.
Data Science/Analysis Workflow
STEP I: Data Import
Exploratory Data Analysis (EDA)
- Cleaning: Handling missing values; unit conversions; general cleaning.
- Transformation: filtering data, summarizing, and working on data as a group.
- Visualization: Trends, distributions, and outliers.
- Correlation: Do variables move together?
Data exploration is a thorough repetitive process
STEP II: Tidy
- Confirm property of your data
- Get quick summary of data
- Try to identify discrepancies in your data
- Remove or impute rows
- Ensure variables have the right data types
- Check for outliers
STEP III: Transform
STEP IV: Visualize
A working plan
- focus on variables of interest first
- focus on other variables next
- visualize relationships between variables
STEP IV: Visualize …
Response variable (fire occurrence)
- Univariate plot
- Categorical data
Code
algeria_tbl |>
ggplot(aes(classes)) +
geom_bar(fill = "#AA4243") +
labs(
y = "Count",
title = "Frequency of wildfire occurrence class in Algeria",
subtitle = "Wildfire occurs more often than not in Algeria's Forest",
caption = "Data source: UCI Machine Learning Repository | Adu O. M."
) +
scale_y_continuous(
breaks = seq(0, 150, 30),
limits = c(0, 150)
) +
theme_light(
base_size = 24,
base_family = "bsk"
) +
coord_cartesian(expand = FALSE) +
theme(
plot.title.position = "plot",
plot.title = element_text(
colour = "#AA4203",
family = title_font,
size = 48,
margin = margin(b = 5, unit = "pt")
),
axis.title.x = element_blank(),
plot.subtitle = element_text(color = "#dc730e"),
axis.text = element_text(color = "#AA4203"),
axis.title.y = element_text(color = "#AA4203")
)
STEP IV: Visualize …
Response Variable (FWI)
- Univariate plot
- Continuous data
Code
algeria_tbl |>
ggplot(aes(fwi)) +
geom_histogram(
stat = "density",
col = "#dc730e"
) +
geom_density(
linewidth = 1.3,
col = "#AA4203"
) +
theme_clean(
base_size = 32,
base_family = main_font_2
) +
scale_y_continuous(
breaks = seq(0, .1, .02),
limits = c(0, .1)
) +
labs(
x = "Forest Weather Index",
y = "Density",
title = "Distribution of FWI",
subtitle = "The distribution shows a long right tail; modeling this distribution might require transformation",
caption = "Data source: UCI Machine Learning Repository | Adu O. M."
) +
coord_cartesian(expand = FALSE) +
theme(
plot.title.position = "plot",
plot.title = element_text(
colour = "#AA4203",
family = title_font_2,
size = 32,
margin = margin(b = 5, unit = "pt")
),
plot.subtitle = element_textbox_simple(
color = "#dc730e",
margin = margin(b = 5, unit = "pt")
),
axis.text = element_text(color = "#AA4203"),
axis.title.y = element_text(color = "#AA4203")
)
STEP IV: Visualize …
Explanatory Variable
STEP IV Visualize …
Response vs explanatory variable
- Bivariate plot
- Categorical vs Categorical data
Code
algeria_tbl |>
summarize(
.by = c(region, classes),
count = n()
) |>
ggplot(aes(region, count, fill = classes)) +
geom_col(position = "dodge", col = "#030303") +
labs(
x = "Region",
y = "Count",
fill = "Wildfire occurrence",
title = "Fire occurrence across Algeria's Forest",
subtitle = "Sidi Bel Abbès experiences more wildfires than Béjaïa",
caption = "Visuals by Adu Olamide M."
) +
geom_label(
aes(label = count),
position = position_dodge(width = 1),
show.legend = FALSE,
size = 4.5
) +
scale_y_continuous(
limits = c(0, 90),
breaks = seq(0, 90, 15)
) +
scale_fill_manual(
values = c("#dc730e", "#dcf3ff"),
labels = c("Fire", "No Fire")
) +
coord_cartesian(expand = FALSE) +
theme_pander(
base_size = 24,
base_family = main_font
) +
theme(
plot.title = element_text(
colour = "#AA4203",
family = title_font_2,
size = 32,
margin = margin(b = 5, unit = "pt")
),
plot.subtitle = element_textbox_simple(
color = "#dc730e",
margin = margin(b = 5, unit = "pt")
),
axis.text = element_text(color = "#AA4203"),
axis.title.y = element_text(color = "#AA4203")
)
STEP IV: Visualize …
Response vs explanatory variable
- bivariate plot
- Continuous vs continuous
Code
algeria_tbl |>
ggplot(aes(date, fwi,)) +
geom_line(
linewidth = 0.8,
col = "#AA470e"
) +
theme_pander(base_size = 24) +
labs(
x = "Date",
y = "FWI",
title = "Trend of FWI from June to October",
subtitle = "There are high spikes at least once in each month. September and August show exceptionally high spikes"
) +
coord_cartesian(expand = FALSE) +
theme (
plot.subtitle = element_textbox_simple()
)
STEP IV: Visualize …
Response vs explanatory variable
- multivariate plot
- Categorical vs continuous
Code
algeria_tbl |>
ggplot(aes(date, fwi, col = classes)) +
geom_line() +
facet_wrap(~region)+
theme_pander(
base_size = 24,
base_family = main_font
) +
labs (
y = "FWI",
title = "Trend of FWI from June to October across the forest regions in Algeria",
subtitle = "High FWI between July and September signals forests with high fuel loads"
) +
scale_color_manual(
values = c("#dc730e", "dodgerblue"),
label = c("Fire", "No fire")
) +
theme(
plot.title.position = "plot",
plot.title = element_text(
colour = "#AA4203",
family = title_font,
size = 32,
margin = margin(b = 5, unit = "pt")
),
plot.subtitle = element_textbox_simple(
color = "#cd730e",
margin = margin(b = 5, unit = "pt")
),
axis.text = element_text(color = "#AA4203"),
axis.title.y = element_text(color = "#AA4203")
)
Some factors that influence the choice of models
- data types: categorical (binomial, multinomial) / continuous / discrete (count)
- size of data
- linearity of variables
- distribution of variables (normal / non-normal data / tweedie)
- explainability vs prediction
Are the Assumptions Met?
- Linearity: Is the relationship truly a straight line?
- Independence: Are errors related?
- Normality: Are residuals bell-shaped?
- Homoscedasticity: Constant variation?
Model Interpretation
While the model captured signals well, it shows:
- heteroscedasticity (fan shape). It captures lower values more accurately than higher values.
- potential non-linear relationship. There is a pattern in the data that the current linear model is missing.
Code
algeria_tbl |>
ggplot(aes(fwi)) +
geom_histogram(
stat = "density",
col = "#dc730e"
) +
geom_density(
linewidth = 1.3,
col = "#AA4203"
) +
geom_label(
aes(
x = 15,
y = 0.02,
label = "Non-normal distribution")
) +
geom_hline(aes(yintercept = 0.074, x=25)) +
geom_textbox(
aes(x = 10, y = 0.08, label = "There are a lot of observations around point zero. Zeros affect linear models")
) +
theme_clean(
base_size = 32,
base_family = main_font_2
) +
scale_y_continuous(
breaks = seq(0, .1, .02),
limits = c(0, .1)
) +
labs(
x = "Forest Weather Index",
y = "Density",
title = "Distribution of FWI",
subtitle = "The distribution shows a long right tail; modeling this distribution might require transformation",
caption = "Data source: UCI Machine Learning Repository | Adu O. M."
) +
coord_cartesian(expand = FALSE) +
theme(
plot.title.position = "plot",
plot.title = element_text(
colour = "#AA4203",
family = title_font_2,
size = 32,
margin = margin(b = 5, unit = "pt")
),
plot.subtitle = element_textbox_simple(
color = "#dc730e",
margin = margin(b = 5, unit = "pt")
),
axis.text = element_text(color = "#AA4203"),
axis.title.y = element_text(color = "#AA4203")
)
