Example: CSV I/O

Reading data from text files

A successful import produces the right rows, columns, types, and units. After the small example.txt exercise, we will now follow an official dataset from its source table to an R object used in the exploration example.

Source and downloads

Our source is Destatis’s Survey of land area by actual use, 2024, published on November 21, 2025. Worksheet csv-33111-03 contains 17 records: Germany and its 16 federal states, all at December 31, 2024. It has 64 columns, including identifiers, descriptions, total area in hectares, and land-use percentages. These are state-level observations for one date, not districts or a time series.

Use the Data/ folder you created on the file-organization page. Download the teaching CSV and save it there, keeping its filename. Also available are the original Destatis workbook, column dictionary and source crosswalk, and source and preparation notes. The CSV retains all 17 source records and 64 columns. Its headers, survey description and region names are English teaching translations; numerical values, identifiers and empty cells are preserved. The dictionary maps every English column name to its original German source header and explains its meaning and units. The original workbook remains in German.

The official publication page and methods report provide the publisher’s definitions.

Import and inspect

This CSV starts with its header: inspect the file before choosing skip. It uses semicolons, decimal commas, and UTF-8 text. Read the columns as text initially so that identifiers such as "01" keep their leading zero:

raw <- read.csv2("Data/destatis-land-use-2024.csv",
                 colClasses = "character", na.strings = "",
                 fileEncoding = "UTF-8", check.names = FALSE)
dim(raw)

## [1] 17 64

raw[1:3, c("Reference_year", "State_ID", "State_name")]

##   Reference_year State_ID         State_name
## 1           2024        0            Germany
## 2           2024       01 Schleswig-Holstein
## 3           2024       02            Hamburg

read.csv2() is a convenience form of read.table() with header = TRUE, sep = ";", and dec = ",". colClasses = "character" postpones numeric conversion; na.strings = "" imports empty cells as missing values, and check.names = FALSE retains the teaching CSV’s exact column names. These settings describe this file; the timber CSV in Assignment 03 has a different header arrangement and decimal separator.

Use the dictionary to distinguish measurements from metadata. Reference_year is the reporting year, State_ID the state identifier, and State_name the reporting area’s name. A code identifies an area; it is not a measured amount.

Select columns and explain their meaning

Keep raw unchanged. Select a small set of columns and assign shorter names for analysis. The names on the left below are our analysis names; the strings on the right are exact headers in the English teaching CSV:

columns <- c(
  Year = "Reference_year",
  ID = "State_ID",
  Place = "State_name",
  Area_ha = "Total_area_ha",
  Settlement = "Settlement_pct_10000000",
  Transport = "Transport_pct_20000000",
  Vegetation = "Vegetation_pct_30000000",
  Agriculture = "Agriculture_pct_31000000",
  Forest = "Forest_pct_32000000",
  Water = "Water_pct_40000000"
)
land_use <- raw[unname(columns)]
names(land_use) <- names(columns)

unname(columns) supplies the original headers for column selection; names(columns) supplies the shorter analysis names. Selecting a data frame with one index, as in raw[unname(columns)], selects columns and keeps all rows.

Area_ha is total area in hectares (1 ha = 10,000 m²). The other six measurement columns are percentages of each area’s total area, not hectares. For example, 9.4 means 9.4%, rather than a fraction of 0.094. The eight-digit suffixes identify land-use categories.

Settlement includes recreational uses and excludes the separately reported Transport category. It is not a measure of sealed soil. Agriculture and Forest are components of the broader Vegetation category. Thus these columns describe different levels of a classification.

Convert only the numeric fields, keeping ID and Place as text. setdiff() selects all column names except those two; type.convert() interprets the selected text columns using decimal commas:

numeric_columns <- setdiff(names(land_use), c("ID", "Place"))
land_use[numeric_columns] <- type.convert(
  land_use[numeric_columns], dec = ",", as.is = TRUE
)
str(land_use)

## 'data.frame':	17 obs. of  10 variables:
##  $ Year       : int  2024 2024 2024 2024 2024 2024 2024 2024 2024 2024 ...
##  $ ID         : chr  "0" "01" "02" "03" ...
##  $ Place      : chr  "Germany" "Schleswig-Holstein" "Hamburg" "Lower Saxony" ...
##  $ Area_ha    : int  35768346 1580430 75509 4770988 41994 3411244 2111564 1985796 3574785 7054178 ...
##  $ Settlement : num  9.5 9.4 46.3 9.5 44 16.9 9.5 8.9 9.6 7.9 ...
##  $ Transport  : num  5.1 4.4 12.4 5.2 12.3 7.1 6.7 6.5 5.6 4.7 ...
##  $ Vegetation : num  83.1 81.3 33.1 83.2 32.2 74.2 82.4 83.2 83.7 85.7 ...
##  $ Agriculture: num  50.2 68.1 22.7 57.7 26.8 46.6 41.1 40.3 44.6 46 ...
##  $ Forest     : num  29.9 10.3 5.6 21.7 1.4 24.8 40.1 40.5 37.9 35.4 ...
##  $ Water      : num  2.3 4.9 8.2 2.1 11.6 1.8 1.4 1.4 1.1 1.7 ...

Inspect the result: an unexpected text value can prevent a column from becoming numeric. Missing data are not zeros. A displayed zero may also represent a small amount rounded to the source’s precision.

Check which observations belong together

Germany (code "0") is the national aggregate, not a seventeenth federal state. Select just the states before comparing their distributions:

states <- land_use[land_use$ID != "0", ]
rownames(states) <- NULL
nrow(states)

## [1] 16

anyDuplicated(states$ID)

## [1] 0

summary(states[c("Settlement", "Transport", "Agriculture", "Forest")])

##    Settlement      Transport       Agriculture        Forest
##  Min.   : 5.60   Min.   : 3.100   Min.   : 4.00   Min.   : 1.40
##  1st Qu.: 7.85   1st Qu.: 4.400   1st Qu.:40.12   1st Qu.:20.40
##  Median : 9.50   Median : 5.400   Median :46.30   Median :25.80
##  Mean   :17.01   Mean   : 6.569   Mean   :44.49   Mean   :25.61
##  3rd Qu.:15.93   3rd Qu.: 6.800   3rd Qu.:54.62   3rd Qu.:35.33
##  Max.   :55.90   Max.   :14.700   Max.   :68.10   Max.   :40.50

There should be 16 rows and no duplicated state identifiers. These checks support a meaningful import; they do not replace reading the source definitions. An unweighted mean of the state percentages gives each state equal weight. It is not Germany’s area-weighted percentage; the published national record answers that question.

Settlement, transport, vegetation, and water are the four main categories. Their rounded shares add to approximately 100%:

main_categories <- c("Settlement", "Transport", "Vegetation", "Water")
range(rowSums(states[main_categories]))

## [1]  99.9 100.1

The small deviations from 100% reflect rounding. Adding Agriculture and Forest again would double-count parts of Vegetation. The four columns used in the overview boxplots are a selected set of non-overlapping categories, not an exhaustive partition of all land.

Writing data to CSV files

Save prepared output separately from the source. row.names = FALSE avoids adding an extra index column:

dir.create("Output", showWarnings = FALSE)
write.csv(states, "Output/land-use-2024-states.csv",
          row.names = FALSE, fileEncoding = "UTF-8")

This output uses commas between fields and decimal points. CSV stores text, so re-importing it requires attention to identifiers and types again.

Alternative data I/O using RDS

The prepared file used in the next example is generated from this same CSV by the selection, renaming, conversion, and state filtering shown above. As explained in the shared project layout, this RDS stays in Data/ because it is the prepared input for exploration. The CSV export above is in Output/:

saveRDS(states, "Data/destatis-land-use-2024-states.rds", version = 2)
restored <- readRDS("Data/destatis-land-use-2024-states.rds")
identical(states, restored)

## [1] TRUE

RDS preserves the R object’s column types. It does not by itself explain the source, units, or preparation choices; retain the dictionary and source notes with the data. You have now created the RDS needed for exploration; no further download is required to continue. The supplied prepared RDS is an alternative if you need to start at the exploration page.

For an optional way to repeat the preparation, download the preparation script into Data/, then run:

source("Data/prepare-land-use-2024.R")
prepare_land_use("Data")

This example gives an initial, documented preparation workflow. Unit 04 develops missing-value handling, more extensive cleaning, and reshaping. This small state table is for Unit 03; later regression exercises use a separate regional dataset with different categories and many more observations.

The import and preparation are complete. Next, Visualization introduces plot selection with small examples before we explore these prepared state data.