Be Weatherwise or Otherwise

[Take Home Exercise 3]

Author

FirGhaz

Published

April 2, 2025

3.1 Context Setting

Past climate trends over Singapore have shown an increase in surface air temperatures and the frequency of heavy rainfall over the past few decades. However, the climate system is complex and the past trends and the magnitude of the change will not necessarily continue into the future. Climate projections using tools like climate modelling is thus necessary to project the future climate for planning and adapting to climate change.

3.2 Overview

According to an office report as shown in the info-graphic above, Daily mean temperature are projected to increase by 1.4 to 4.6, and the contrast between the wet months (November to January) and dry month (February and June to September) is likely to be more pronounced.

We will apply visual interactivity and visualizing uncertainty methods to validate the claims presented above.

3.3 The Task

In this Take Home Exercise 3, we will be creating an analystics-driven data visualisation on;

  1. historical daily temperature and (b) rainfall data from the Meteorological Service Singapore website,

  2. records of a month of the year 1983, 1993, 2003, 2013 and 2023 and create an analytics-driven data visualisation and,

  3. apply appropriate interactive techniques to enhance the user experience in data discovery and/or visual story-telling.

3.4 Preparing Data and R Packages

In this exercise, the author has decided to select the December month data set from Changi Meteorological records from the years 1983, 1993, 2003, 2013 and 2023. The data is complete for the rainfall records; however, for the temperature records, the data started from 2003 onwards. To that end, the mean temperature will be utilised for data visualisation and analysis. Other Statistical Summary will be used too.

For the purpose of this exercise, the following R packages will be used, they are:

  • tidyverse, a family of R packages for data science process,
  • plotly for creating interactive plot,
  • gganimate for creating an animation plot,
  • DT for displaying interactive html tables,
  • `crosstalk` for for implementing cross-widget interactions (currently, linked brushing and filtering), and
  • `ggdist` for visualizing distribution and uncertainty.
Code 
pacman::p_load(ggplot2, ggbeeswarm, gghalves, ungeviz, plotly, crosstalk,
               DT, ggdist, ggridges, GGally, parallelPlot,
               colorspace, gganimate, tidyverse, ggthemes, ggiraph)

The historical records for the year will be imported and merged.

Code 
library(readr)
library(dplyr)
#| warning = FALSE
#| echo = FALSE
data_1983 <- read_csv("data/DAILYDATA_S24_198312.csv", locale = locale(encoding = "Latin1"))
Rows: 31 Columns: 13
── Column specification ────────────────────────────────────────────────────────
Delimiter: ","
chr  (1): Station
dbl (12): Year, Month, Day, Daily Rainfall Total (mm), Highest 30 Min Rainfa...

ℹ Use `spec()` to retrieve the full column specification for this data.
ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.
Code 
data_1993 <- read_csv("data/DAILYDATA_S24_199312.csv", locale = locale(encoding = "Latin1"))
Rows: 31 Columns: 13
── Column specification ────────────────────────────────────────────────────────
Delimiter: ","
chr  (1): Station
dbl (12): Year, Month, Day, Daily Rainfall Total (mm), Highest 30 Min Rainfa...

ℹ Use `spec()` to retrieve the full column specification for this data.
ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.
Code 
data_2003 <- read_csv("data/DAILYDATA_S24_200312.csv", locale = locale(encoding = "Latin1"))
Rows: 31 Columns: 13
── Column specification ────────────────────────────────────────────────────────
Delimiter: ","
chr  (1): Station
dbl (12): Year, Month, Day, Daily Rainfall Total (mm), Highest 30 Min Rainfa...

ℹ Use `spec()` to retrieve the full column specification for this data.
ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.
Code 
data_2013 <- read_csv("data/DAILYDATA_S24_201312.csv", locale = locale(encoding = "Latin1"))
Rows: 31 Columns: 13
── Column specification ────────────────────────────────────────────────────────
Delimiter: ","
chr  (1): Station
dbl (12): Year, Month, Day, Daily Rainfall Total (mm), Highest 30 Min Rainfa...

ℹ Use `spec()` to retrieve the full column specification for this data.
ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.
Code 
data_2023 <- read_csv("data/DAILYDATA_S24_202312.csv")
Rows: 31 Columns: 13
── Column specification ────────────────────────────────────────────────────────
Delimiter: ","
chr  (1): Station
dbl (12): Year, Month, Day, Daily Rainfall Total (mm), Highest 30 Min Rainfa...

ℹ Use `spec()` to retrieve the full column specification for this data.
ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.

3.5 Merging rainfall Records

Merging code chunks for rainfall, as shown below:

Code 
rainfall_columns <- c("Station","Day","Year","Daily Rainfall Total (mm)")

rainfall_table <- bind_rows(
  select(data_1983, all_of(rainfall_columns))%>% mutate(Year = as.character(Year)),
  select(data_1993, all_of(rainfall_columns))%>% mutate(Year = as.character(Year)),
  select(data_2003, all_of(rainfall_columns))%>% mutate(Year = as.character(Year)),
  select(data_2013, all_of(rainfall_columns))%>% mutate(Year = as.character(Year)),
  select(data_2023, all_of(rainfall_columns))%>% mutate(Year = as.character(Year)),
)

print(rainfall_table)
# A tibble: 155 × 4
   Station   Day Year  `Daily Rainfall Total (mm)`
   <chr>   <dbl> <chr>                       <dbl>
 1 Changi      1 1983                          2.8
 2 Changi      2 1983                          1.7
 3 Changi      3 1983                          5  
 4 Changi      4 1983                          8.2
 5 Changi      5 1983                          0  
 6 Changi      6 1983                          0  
 7 Changi      7 1983                          0  
 8 Changi      8 1983                         19.8
 9 Changi      9 1983                         48.1
10 Changi     10 1983                         21.1
# ℹ 145 more rows

3.5.1 Developing a Static EDA on Daily Rainfall by Day and Year

geom_point() function under the ggplot package was used to plot the total rainfall (mm) on a given day by year. In this static EDA, we can observe that:

1) majority of the days by year, total rainfall (mm) was observed to be below 50mm while,

2) in the years 1983 and 2013, where the total rainfall on day 25 and day 3 were above 100mm. We can infer that these were anomalies with significantly heavy rainfall as compared to the volumes of other rainy days by year.

Note

The Static EDA has its shortcomings in terms of enhancing data appreciation. Next, we will move onto interactivity to enhance our data visualisation.

Code 
library(ggplot2)

ggplot(data = rainfall_table, aes(x = Day, y = `Daily Rainfall Total (mm)`, color = factor(Year))) +
  geom_point() +  # Add points for the daily rainfall totals
  theme_minimal() +  # Use a minimal theme
  labs(title = "Daily Rainfall Total (mm) by Day and Year",
       x = "Day",
       y = "Daily Rainfall Total (mm)",
       color = "Year") +  # Label the axes and the legend
  scale_color_brewer(palette = "Set1")  # Use a color palette that's easily distinguishable

3.5.2 Developing Interacitvity using plotly

Leveraging on plotly the interactive scatter allows the reader to (1) hover for details (Day and Rainfall volume of the day) , (2) select (via cursor, legend or lasso) and (3) compare data on hover. For Sn (3), the function allows the reader to have a quick glance to make a quick analysis on the Daily Rainfall trends across the years. For instance, we can observe that:

  • Day 1 and Day 2 records rainfalls on all 5 selected years, therefore we can infer that the probability of raining during the 1st two days of December will be higher (of course a larger data sample size will enhance the accuracy of this inference).
  • Day 5, 7, 29 and 30 records the lowest rainfall over the 5 selected years, this can only be seen through the compare data on hover function.

3.5.3 Extending plotly further to capture Statistical Components

The code chunk below, calculates the statistical summary of the scatter points which includes; (1) Total Rainfall by day, (2) Mean Rainfall by day and (3) Median Rainfall by day across all years.

Code 
library(dplyr)

# Calculate summary statistics for each year
summary_stats_by_day <- rainfall_table %>%
  group_by(Day) %>%
  summarise(
    Total_Rainfall = sum(`Daily Rainfall Total (mm)`, na.rm = TRUE),
    Mean_Rainfall = mean(`Daily Rainfall Total (mm)`, na.rm = TRUE),
    Median_Rainfall = median(`Daily Rainfall Total (mm)`, na.rm = TRUE),
  ) %>%
  ungroup()
Code 
rainfall_table_with_stats <- rainfall_table %>%
  left_join(summary_stats_by_day, by = "Day")

library(plotly)

plot_ly(data = rainfall_table_with_stats, 
        x = ~Day, 
        y = ~`Daily Rainfall Total (mm)`,
        color = ~factor(Year),
        type = 'scatter',
        mode = 'markers',
        hoverinfo = 'text',
        text = ~paste('Day:', Day,
                      ', Year:', Year,
                      '<br>Total Yearly Rainfall (mm):', Total_Rainfall,
                      '<br>Mean Yearly Rainfall (mm):', Mean_Rainfall,
                      '<br>Median Yearly Rainfall (mm):', Median_Rainfall)) %>%
  layout(title = 'Daily Rainfall Total (mm) with Sum_Stats by Day ',
         xaxis = list(title = 'Day'),
         yaxis = list(title = 'Rainfall (mm)'))

Here with the stat summary, we can observe that individual points is supported by its stats summary of total, mean and median rainfall (mm) by day across the years.

3.6 Temperature across the years

3.6.1 Merging Temperature Records

Code 
temp_columns <- c("Station","Day","Year","Mean Temperature (°C)", "Maximum Temperature (°C)", "Minimum Temperature (°C)")

temp_table <- bind_rows(
  select(data_1983, all_of(temp_columns))%>% mutate(Year = as.character(Year)),
  select(data_1993, all_of(temp_columns))%>% mutate(Year = as.character(Year)),
  select(data_2003, all_of(temp_columns))%>% mutate(Year = as.character(Year)),
  select(data_2013, all_of(temp_columns))%>% mutate(Year = as.character(Year)),
  select(data_2023, all_of(temp_columns))%>% mutate(Year = as.character(Year)),
)

Utilising three key features of stat_halfeye(), geom_boxplot and geom_dotplot to study onto to the given data. The code chunk below shows the static plots:

Code 
ggplot(temp_table, aes(x = Year, y = `Mean Temperature (°C)`, fill = Year)) +
  stat_halfeye(
    adjust = 0.4, 
    position = position_nudge(x = 0.13), 
    color = NA, 
    alpha = 0.4
  ) +
  geom_boxplot(
    width = 0.2, 
    outlier.shape = NA, 
    position = position_dodge(width = 0.5), 
    alpha = 0.6, 
    color = "gray40", 
    fill = "grey93"
  ) +
  geom_dotplot(
    binaxis = "y", 
    stackdir = "down",
    position = position_nudge(x = -0.13),
    binwidth = 0.25, 
    dotsize = 0.25
  ) +
  scale_y_continuous(limits = c(23.5, 28.5)) +
  scale_fill_brewer(palette = "Pastel1") +
  theme_minimal(base_size = 8) +
  theme(
    legend.position = "right",
    axis.text.x = element_text(angle = 45, hjust = 1),
    legend.title = element_blank(),
    plot.title = element_text(hjust = 0.5, size = 16),
    plot.subtitle = element_text(hjust = 0.5, size = 12)
  ) +
  labs(
    y = "Mean Temperature (°C)", 
    x = "Year", 
    title = "Mean Temperature by Year",
    subtitle = "Rising Mean Temperature over the decade"
  ) +
  guides(fill = guide_legend(title.position = "top", title.hjust = 0.5))

Based on the static EDA, we can see that the overall mean temperature for the selected year have increased across four decades. Through the `stat_halfeye’ it can be seen that (1) the range of fluctuating mean temps over the years has condensed significantly, (2) peaks (Frequencies of Higher Mean Temp) are compressed to be lesser but sharper in scale (i.e: 2023 has 1 main peak as compared to 1983 - 2003, with a higher peak) and (3) the number of dotplots (with binwidth=2.5) denotes the co-occurrence of a binned temperature of similar range over the given days.

3.6.2 Adding Interactivity to the Temp Scale

Through the interactivity, the tooltip, provides precise interactivity to give the viewer more information on the points/dots denoted above. The codes below calculates other statistical components to be added back onto the temp_table.

Code 
#| warning = FALSE
library(ggplot2)
library(ggiraph)
library(RColorBrewer)
library(dplyr)

summary_stats <- temp_table %>%
  group_by(Year) %>%
  summarise(
    OverallMean = mean(`Mean Temperature (°C)`, na.rm = TRUE),
    OverallMin = min(`Mean Temperature (°C)`, na.rm = TRUE),
    OverallMax = max(`Mean Temperature (°C)`, na.rm = TRUE),
    Range = OverallMax - OverallMin  # Calculate the range
  ) %>%
  ungroup()

temp_table_2 <- temp_table %>%
  left_join(summary_stats, by = "Year")

head(temp_table_2)
# A tibble: 6 × 10
  Station   Day Year  `Mean Temperature (°C)` `Maximum Temperature (°C)`
  <chr>   <dbl> <chr>                   <dbl>                      <dbl>
1 Changi      1 1983                     26.4                       31  
2 Changi      2 1983                     24.3                       27.2
3 Changi      3 1983                     25.1                       30.2
4 Changi      4 1983                     25.2                       30.3
5 Changi      5 1983                     26                         29.8
6 Changi      6 1983                     25                         27.7
# ℹ 5 more variables: `Minimum Temperature (°C)` <dbl>, OverallMean <dbl>,
#   OverallMin <dbl>, OverallMax <dbl>, Range <dbl>
Code 
p <- ggplot(temp_table_2, aes(x = Year, y = `Mean Temperature (°C)`, fill = Year)) +
  stat_halfeye(
    adjust = 0.4, 
    position = position_nudge(x = 0.2), 
    color = NA, 
    alpha = 0.4
  ) +
  geom_boxplot (
    width = 0.2, 
    outlier.shape = NA, 
    position = position_dodge(width = 0.5), 
    alpha = 0.6, 
    color = "gray40", 
    fill = "white"
  ) +
  geom_point_interactive( 
    aes(
      tooltip = paste('Day', `Day`, '<br>Mean Temp (°C):', `Mean Temperature (°C)`), 
      data_id = Year
    ),
    position = position_nudge(x = -0.2),
    bindwidth = 2.5,
    size = 1.3,  # Adjust the size as per your preference
    alpha = 0.35
  ) +
  scale_y_continuous(limits = c(23.5, 28.5)) +
  scale_fill_brewer(palette = "Pastel1") +
  theme_minimal(base_size = 8) +
  theme(
    legend.position = "right",
    axis.text.x = element_text(angle = 45, hjust = 1),
    legend.title = element_blank(),
    plot.title = element_text(hjust = 0.5, size = 16),
    plot.subtitle = element_text(hjust = 0.5, size = 12)
  ) +
  labs(
    y = "Mean Temperature (°C)", 
    x = "Year", 
    title = "Mean Temperature by Year",
    subtitle = "Rising Mean Temperature over the decade"
  ) +
  guides(fill = guide_legend(title.position = "top", title.hjust = 0.5))

g <- girafe(                                  
 ggobj = p,                             
  width_svg = 6,                         
  height_svg = 6*0.618,
  options = list(                        
   opts_hover(css = "fill: #963000;"),  
    opts_hover_inv(css = "opacity:0.2;") 
  )                                        
 ) 
g
  1. Through the interactivity on the geom_boxplot, it can be observed that the Overall Mean and Max temperature has increased over the years with the year 2023 having the highest temperature.
  2. It can also be observed that the range is much more condensed in the Year 2023, with that we can infer that as the Max Temp rises the Minimum temperature over the years have risen as well.
  3. Through the intensity of the dotplot and the lumpiness of the halfeye, it can be seen that most 2023 has a ‘single’ condensed region unlike other years where there are mulitple peaks.

3.7 Conclusion

In Summary, the interactivity functions utilised in the Take Home Exercise allowed us to explore and gain more insights on the probability (for rainfalls) and increasing trends (for temperature). The utility of the tooltip, data_ID and functions such as comparing two or more data when hovering, enhancing data appreciation and in turns garner greater insights and inferences.