Using mudata objects
In mudata2: Interchange Tools for Multi-Parameter Spatiotemporal Data

The mudata2 package is designed to be used as little as possible. That is, if you need use data that is currently in mudata format, the functions in this package are designed to let you spend as little time as possible reading, subsetting, and inspecting your data. The steps are generally as follows:

Read in the data using read_mudata()
See what it contains using summary(), print(), distinct_locations(), and distinct_params()
Inspect the metadata using tbl_locations() and tbl_params()
Pick your parameters using select_params() or filter_params()
Pick your locations using select_locations() or filter_locations()
Extract the data using tbl_data() or tbl_data_wide()

In this vignette we will use the ns_climate dataset within the mudata2 package, which is a collection of monthly climate observations from Nova Scotia (Canada), sourced from Environment Canada using the rclimateca package.

library(mudata2)
data("ns_climate")
ns_climate

Reading an object

The ns_climate object is already an object in R, but if it wasn't, you would need to use read_mudata() to read it in. If you're curious what a mudata object looks like on disk, you could try using write_mudata() to find out. I tend to prefer writing to a directory rather than a JSON or ZIP file, but you can take your pick.

# write to directory
write_mudata(ns_climate, "ns_climate.mudata")
# write to ZIP
write_mudata(ns_climate, "ns_climate.mudata.zip")
# write to JSON
write_mudata(ns_climate, "ns_climate.mudata.json")

Then, you can read in the object using read_mudata():

# read from directory
read_mudata("ns_climate.mudata")
# read from ZIP
read_mudata("ns_climate.mudata.zip")
# read from JSON
read_mudata("ns_climate.mudata.json")

Inspecting an object

The three main ways to quickly inspect a mudata object are print() and summary(). The print() function is what you get when you type the name of the object at the prompt, and gives a short summary of the object. The output suggests a couple of other ways to inspect the object, including distinct_locations(), which returns a character vector of location identifiers, and distinct_params(), which returns a character vector of parameter identifiers.

print(ns_climate)

The summary() function provides some numeric summaries by dataset, location, and parameter if the value column of the data table is numeric (if it isn't, it provides counts instead).

summary(ns_climate)

Inspecting metadata

You can have a look at the embedded documentation using tbl_params(), and tbl_locations(), which contain any additional information about parameters and locations for which data are available. The identifiers (i.e., param and location columns) of these can be used to subset the object using select_*() functions; the tables themselves can be used to subset the object using the filter_*() functions.

# extract the parameters table
ns_climate %>% tbl_params()

# exract the locations table
ns_climate %>% tbl_locations()

Subsetting an object

You can subset mudata objects using select_params() and select_locations(), which use dplyr-like selection syntax to quickly subset mudata objects using the identifiers from distinct_locations() and distinct_params() (respectively).

# find out which parameters are available
ns_climate %>% distinct_params()

# subset by parameter
ns_climate %>% select_params(mean_temp, total_precip)

You can also use the dplyr select helpers to select related params/locations...

ns_climate %>% select_params(contains("temp"))

...and rename params/locations on the fly.

ns_climate %>% select_locations(Kentville = starts_with("KENT"))

To select params/locations based on the tbl_params() and tbl_locations() tables, you can use the filter_*() functions (note that last_year is a column in tbl_locations(), and unit is a column in tbl_params()):

# only use locations whose last data point was after 2000
ns_climate %>%
  filter_locations(last_year > 2000)

# use only params measured in mm
ns_climate %>%
  filter_params(unit == "mm")

Similarly, we can subset parameters, locations, and the data table all at once using filter_data().

library(lubridate)
# extract only June temperature from the data table
ns_climate %>%
  filter_data(month(date) == 6)

Extracting data

The data is stored in the data table (i.e., tbl_data()) in parameter-long form (that is, one row per measurement rather than one row per observation). This has advantages in that information about each measurement can be stored next to the value (e.g., standard deviation, notes, etc.), however it is rarely the form required for analysis. To extract data in parameter-long form, you can use tbl_data():

ns_climate %>% tbl_data()

To extract data in a more standard parameter-wide form, you can use tbl_data_wide():

ns_climate %>% tbl_data_wide()

The tbl_data_wide() function isn't limited to parameter-wide data - data can be anything-wide (Edzer Pebesma has a great discussion on this). Using tbl_data_wide() is identical to using tbl_data() and tidyr::spread(), with context-specific defaults.

ns_climate %>%
  select_params(mean_temp) %>%
  filter_data(year(date) == 1960) %>%
  tbl_data_wide(key = location)

Putting it all together

Using the pipe (%>%), we can string all the steps together concisely:

temp_1960 <- ns_climate %>%
  # pick parameters
  select_params(contains("temp")) %>%
  # pick locations
  select_locations(
    `Sable Island` = starts_with("SABLE"),
    `Kentville` = starts_with("KENT"),
    `Badeck` = starts_with("BADD")
  ) %>%
  # filter data table
  filter_data(year(date) == 1960) %>%
  # extract data in wide format
  tbl_data_wide()

temp_1960

We can then use this data with ggplot2 to lead us to the conclusion that three locations in the same province had more or less the same monthly temperature characteristics in 1960.

library(ggplot2)
ggplot(
  temp_1960,
  aes(
    x = date,
    y = mean_temp,
    ymin = extr_min_temp,
    ymax = extr_max_temp,
    col = location,
    fill = location
  )
) +
  geom_ribbon(alpha = 0.2, col = NA) +
  geom_line()