R或Matlab中的季节性箱形图

Question

DATE        obs1        obs2    obs3    
1981-01-01  2032.409    3142.46 1741.143
1981-01-02  2023.687    3870.04 1735.256
1981-01-03  2014.274    4126.25 1728.556
1981-01-04  2005.795    2615.91 1722.985
1981-01-05  2000.674    2940.83 1722.317
1981-01-06  1998.477    3258.69 1723.937
1981-01-07  1997.014    3371.6  1724.104
1981-01-08  1995.576    3184.13 1722.624
1981-01-09  1993.706    3540.76 1719.592
1981-01-10  1991.286    3312.43 1715.156
1981-01-11  1988.633    3028.65 1710.141
1981-01-12  1986.147    3212.79 1705.183
1981-01-13  1984.229    3193.23 1700.789
1981-01-14  1982.756    3294.52 1697.785
1981-01-15  1981.548    3553.78 1696.068
1981-01-16  1980.561    3492.28 1694.544
1981-01-17  1979.792    2452.09 1692.36
1981-01-18  1979.224    1873.82 1689.525
1981-01-19  1978.845    3218.28 1686.452

I need to plot seasonal(winter, spring, summer and Fall) box plot in R for the daily data as shown above. I have 10 years of data in above format for different stations. The plot should be in one figure, with multiple box plots in each season.

Answer 1

A solution using tidyverse and lubridate . tidyverse includes dplyr and tidyr to perform data wrangling, and ggplot2 to create the plot. lubridate is to handle the dates in the data frame.

Since the dataset you provided is not very useful, as it only contains a few records only from January thus impossible to create a boxplot showing the seasonal difference, I decided to create a new example data frame. The structure of my example data frame is similar to your dataset, which should give you some hints as a starting point for your real-world problem.

# Set the seed for reproducibility
set.seed(123)

library(tidyverse)
library(lubridate)

# Create example data frame
dt <- data_frame(DATE = seq(ymd("1980-01-01"), ymd("1989-12-31"), by = 1)) %>%
  mutate(obs1 = rnorm(nrow(.), mean = 0, sd = 1),
         obs2 = rnorm(nrow(.), mean = 1, sd = 2),
         obs3 = rnorm(nrow(.), mean = 2, sd = 3))

head(dt)
# # A tibble: 6 x 4
#         DATE        obs1       obs2      obs3
#       <date>       <dbl>      <dbl>     <dbl>
# 1 1980-01-01 -0.56047565  0.7874145 2.7827006
# 2 1980-01-02 -0.23017749  0.1517417 8.5720252
# 3 1980-01-03  1.55870831  0.7193725 1.3293478
# 4 1980-01-04  0.07050839  0.5454177 0.3253155
# 5 1980-01-05  0.12928774  1.4101239 6.1245771
# 6 1980-01-06  1.71506499 -0.6491910 4.5034395

tail(dt)
# # A tibble: 6 x 4
#         DATE       obs1       obs2       obs3
#       <date>      <dbl>      <dbl>      <dbl>
# 1 1989-12-26 -0.3629796  0.6750946  0.8586325
# 2 1989-12-27  0.1102218  2.8572337  9.8541328
# 3 1989-12-28 -0.2700741  1.7614026  1.9109596
# 4 1989-12-29  0.6920973  0.5275611 -0.4756240
# 5 1989-12-30  0.9282803  1.3811225  1.5222535
# 6 1989-12-31  0.5931301 -1.6638739  4.1157087

The example data frame contains records for 10 years with 3 observation groups. The values from each column are in normal distribution with different means and standard deviations.

The first step is to process the data frame by converting the dataset from wide format to long format and add a column showing season information.

dt2 <- dt %>% 
  # Convert data frame from lwide format to long format
  gather(Observation, Value, -DATE) %>%
  # Remove "obs" in the Observation column
  mutate(Observation = str_replace(Observation, "obs", "")) %>%
  # Convert the DATE column to date class
  mutate(DATE = ymd(DATE)) %>%
  # Create Month column
  mutate(Month = month(DATE)) %>%
  # Create Season column
  mutate(Season = case_when(
    Month %in% c(12, 1, 2)      ~ "winter",
    Month %in% c(3, 4, 5)       ~ "spring",
    Month %in% c(6, 7, 8)       ~ "summer",
    Month %in% c(9, 10, 11)     ~ "fall",
    TRUE                        ~ NA_character_
  ))

After that, we can use ggplot2 to create the boxplot. Notice that I use stat_summary to add a red line to each group to represent the mean.

# Create a boxplot using ggplot2
# Specify the aesthetics
ggplot(dt2, aes(x = Season, y = Value, fill = Observation)) + 
  # Specify the geom to be boxplot
  geom_boxplot() +                
  # Add a red line to the mean
  stat_summary(aes(ymax = ..y.., ymin = ..y..),
               fun.y = "mean", 
               geom = "errorbar",              # Use geom_errorbar to add line as mean
               color = "red", 
               width = 0.7, 
               position = position_dodge(width = 0.75), # Add the line to each group
               show.legend = FALSE)

Answer 2

Ok... the first step is to build up a function that can detect the season in which a given date falls. Fortunately, I already developed one long time ago that can also handle seasons in the southern hemisphere (which are reversed).

The function doesn't implement any sanity check, because I was using it with already sanitized datasets, but it should not be hard for you to eventually implement some (unless you decide to sanitize your datasets before using it). It works in a vectorized way to maximize the performance of calculations within Matlab.

Here is it:

function season = GetSeason(date,southern_hemisphere)

    if (nargin == 1)
        southern_hemisphere = false;
    end

    [~,month,day] = datevec(date);
    offset = month + (day / 100);

    winter = (offset < 3.21) | (offset >= 12.22);
    spring = ~winter & (offset < 6.21);
    summer = ~winter & ~spring & (offset < 9.23);
    autumn = ~winter & ~spring & ~summer;

    offset(spring) = 0;
    offset(summer) = 1;
    offset(autumn) = 2;
    offset(winter) = 3;

    if (southern_hemisphere)
        offset = offset + 2;
    end

    season = mod(offset,4) + 1;
end

Now, the first step, within your script, is to extract your observations from a dataset file. In order to create a fully working demo for you, I created an Excel dataset. But you could also use a CSV dataset with almost no changes in the code or other file formats handled by Matlab:

% detect the dataset columns format
opts = detectImportOptions('data.xlsx');

% impose a specific format for the dataset columns
opts = setvartype(opts,{'datetime' 'double' 'double' 'double'});

% extract data in a table variable
data = readtable('data.xlsx',opts);

% sanitize the table variable removing the rows with missing or invalid values
data = rmmissing(data);

% sort the table variable rows by date (default first rows, default ascending)
data = sortrows(data);

The second test consists in getting the corresponding season for the observation dates:

seasons = GetSeason(data.Date);

The third step, assuming we are performing all this process only for the first column of observations called Obs1 :

spring_1 = data.Obs1(seasons == 1);
summer_1 = data.Obs1(seasons == 2);
autumn_1 = data.Obs1(seasons == 3);
winter_1 = data.Obs1(seasons == 4);

The fourth and final step consists in plotting one boxplot for each season in a single graph (the grouping variable groups must be passed as argument into the boxplot function in order to aknowledge the latter of how many boxes it has to draw and using which values):

groups = [
    ones(size(spring_1));
    2 * ones(size(summer_1));
    3 * ones(size(autumn_1));
    4 * ones(size(winter_1));
];

figure();
boxplot([spring_1; summer_1; autumn_1; winter_1],groups);
set(gca,'XTickLabel',{'Spring' 'Summer' 'Autumn' 'Winter'});

And here is the result:

UPDATE WITH FULL WORKING CODE FOR ALL OBSERVATIONS

opts = detectImportOptions('data.xlsx');
opts = setvartype(opts,{'datetime' 'double' 'double' 'double'});

data = readtable('data.xlsx',opts);
data = rmmissing(data);
data = sortrows(data);

seasons = GetSeason(data.Date);

spring_1 = data.Obs1(seasons == 1);
summer_1 = data.Obs1(seasons == 2);
autumn_1 = data.Obs1(seasons == 3);
winter_1 = data.Obs1(seasons == 4);
spring_2 = data.Obs2(seasons == 1);
summer_2 = data.Obs2(seasons == 2);
autumn_2 = data.Obs2(seasons == 3);
winter_2 = data.Obs2(seasons == 4);
spring_3 = data.Obs3(seasons == 1);
summer_3 = data.Obs3(seasons == 2);
autumn_3 = data.Obs3(seasons == 3);
winter_3 = data.Obs3(seasons == 4);

plot_data = [
    spring_1;
    summer_1;
    autumn_1;
    winter_1;
    spring_2; 
    summer_2; 
    autumn_2;
    winter_2;
    spring_3;
    summer_3;
    autumn_3;
    winter_3
];

plot_groups = [
    (1 * ones(size(spring_1))) (1  * ones(size(spring_1)));
    (1 * ones(size(summer_1))) (2  * ones(size(summer_1)));
    (1 * ones(size(autumn_1))) (3  * ones(size(autumn_1)));
    (1 * ones(size(winter_1))) (4  * ones(size(winter_1)));
    (2 * ones(size(spring_2))) (5  * ones(size(spring_2)));
    (2 * ones(size(summer_2))) (6  * ones(size(summer_2)));
    (2 * ones(size(autumn_2))) (7  * ones(size(autumn_2)));
    (2 * ones(size(winter_2))) (8  * ones(size(winter_2)));
    (3 * ones(size(spring_3))) (9  * ones(size(spring_3)));
    (3 * ones(size(summer_3))) (10 * ones(size(summer_3)));
    (3 * ones(size(autumn_3))) (11 * ones(size(autumn_3)));
    (3 * ones(size(winter_3))) (12 * ones(size(winter_3)))
];

labels_obs = {'' '' '' '' '' '' '' '' '' '' '' ''};
labels_season = repmat({'Spring' 'Summer' 'Autumn' 'Winter'},1,3);

figure('Units','normalized','Position',[0.05 0.1 0.9 0.8]);
boxplot(plot_data,plot_groups, ...
    'BoxStyle','outline', ...
    'FactorGap',[5 1], ...
    'Labels',{labels_obs; labels_season}, ...
    'Notch','on');

colors = repmat('wcyg',1,3);
h = findobj(gca,'Tag','Box');

for i = 1:numel(h)
    patch(get(h(i),'XData'),get(h(i),'YData'),colors(i),'FaceAlpha',0.5);
end

h = findall(allchild(findall(gca,'Type','hggroup')),'Type','text','String','');
positions = cell2mat(get(h,'pos'));
positions_new = num2cell([mean(reshape(positions(:,1),4,[]))' positions(1:4:end,2:end)],2);
set(h(1:4:end),{'Position'},positions_new,{'String'},{'Observations 3'; 'Observations 2'; 'Observations 1'})

h = findall(allchild(findall(gca,'Type','hggroup')),'Type','text','String','');
delete(h);

Result: