Getting Started with rmet"

In rmet: Download and Read Brazilian Meteorological Data from INMET

knitr::opts_chunk$set(
  collapse  = TRUE,
  comment   = "#>",
  fig.width = 7,
  fig.height = 4,
  fig.align = "center",
  dpi = 150,
  warning = FALSE,
  message = FALSE
)

Brazil's National Institute of Meteorology (INMET) maintains a network of automatic weather stations that record hourly observations across the country. The raw data are distributed as annual ZIP archives containing CSV files, one per station, with formatting inconsistencies that vary across years (e.g., different date formats, trailing semicolons, phantom columns, and mixed encodings). rmet automates the full pipeline: downloading those archives, resolving their structural quirks, and returning a clean, analysis-ready data.frame.

Installation

install.packages("rmet")

# Development version:
# install.packages("pak")
# pak::pak("rodrigosqrt3/rmet")

library(rmet)

Step 1 — Browse the station catalogue

inmet_stations() downloads the current official automatic-station catalogue and caches it locally. Stale catalogues are periodically refreshed. If the service is unavailable, the package falls back to an older cache or to a smaller bundled catalogue. You can filter by state, operational status, provider, or station name before downloading weather archives.

stations <- inmet_stations(state = "RS", status = "Operante")
head(stations[, c("station_code", "station_name", "latitude",
                  "longitude", "elevation", "status")])

Station A801 is the Porto Alegre automatic station, located at roughly −30.05 °S, −51.17 °W, 46.97 m above sea level.

Step 2 — Download annual archives

inmet_download() fetches the annual ZIP files from INMET's servers and saves them to a persistent local cache. Downloads are resumable: if a connection drops, re-running the same call picks up where it left off.

# Downloads the 2023 archive (~70 MB) to the default cache directory.
# Safe to re-run — skips files that are already complete.
inmet_download(2023)

You can inspect what is already cached at any time:

inmet_cache_status()

Step 3 — Read data into R

inmet_read() parses the ZIP archives directly — no manual extraction needed. It normalises column names, converts measurement columns to numeric, and parses timestamps. Timestamps remain in UTC by default; pass an Olson time-zone name such as tz = "America/Sao_Paulo" when another display is required.

df <- inmet_read(
  years      = 2023,
  stations   = "A801",
  start_date = "2023-01-01",
  end_date   = "2023-12-31"
)

The examples below use the package's built-in sample dataset, which contains one year of hourly observations from station A801 (Porto Alegre, 2023).

df <- rmet_example
str(df[, c("datetime", "station_code", "state",
           "temp_dry_c", "precip_mm", "humid_rel_pct",
           "wind_speed_ms", "pressure_station_hpa")])

Exploring the data

Daily aggregation

Hourly data are often most useful after aggregating to daily summaries.

df$date <- as.Date(df$datetime)

daily <- aggregate(
  cbind(temp_dry_c, precip_mm, humid_rel_pct, wind_speed_ms) ~ date,
  data = df,
  FUN  = function(x) mean(x, na.rm = TRUE)
)

daily$temp_max <- aggregate(temp_dry_c ~ date, df, max, na.rm = TRUE)$temp_dry_c
daily$temp_min <- aggregate(temp_dry_c ~ date, df, min, na.rm = TRUE)$temp_dry_c
daily$precip   <- aggregate(precip_mm  ~ date, df, sum, na.rm = TRUE)$precip_mm
daily$month    <- as.integer(format(daily$date, "%m"))

Temperature series

col_max  <- "#E74C3C"
col_min  <- "#3498DB"
col_band <- "#F1948A"

oldpar <- par(mar = c(4, 4.5, 3, 1), family = "sans", cex.lab = 0.95, cex.axis = 0.85)

plot(
  daily$date, daily$temp_max,
  type = "n",
  ylim = range(c(daily$temp_min, daily$temp_max), na.rm = TRUE),
  xlab = "",
  ylab = "Temperature (\u00b0C)",
  main = "Daily Temperature Range \u2014 Porto Alegre (A801), 2023",
  axes = FALSE
)

polygon(
  c(daily$date, rev(daily$date)),
  c(daily$temp_max, rev(daily$temp_min)),
  col    = adjustcolor(col_band, alpha.f = 0.35),
  border = NA
)

lines(daily$date, daily$temp_max, col = col_max, lwd = 1.2)
lines(daily$date, daily$temp_min, col = col_min, lwd = 1.2)

axis(1,
     at     = seq(min(daily$date), max(daily$date), by = "2 months"),
     labels = format(seq(min(daily$date), max(daily$date), by = "2 months"), "%b"),
     las    = 1)
axis(2, las = 1)
box(col = "grey80")

legend(
  "topright",
  legend = c("Daily max", "Daily min"),
  col    = c(col_max, col_min),
  lwd    = 2,
  bty    = "n",
  cex    = 0.85
)

par(oldpar)

Precipitation

col_prec <- "#3498DB"

oldpar <- par(mar = c(4, 4.5, 3, 1), family = "sans", cex.lab = 0.95, cex.axis = 0.85)

barplot(
  daily$precip,
  col    = adjustcolor(col_prec, alpha.f = 0.75),
  border = NA,
  space  = 0,
  xlab   = "",
  ylab   = "Precipitation (mm)",
  main   = "Daily Precipitation \u2014 Porto Alegre (A801), 2023",
  axes   = FALSE
)

axis(2, las = 1)

# Month labels on x axis
month_starts <- which(!duplicated(format(daily$date, "%m")))
axis(1,
     at     = month_starts,
     labels = format(daily$date[month_starts], "%b"),
     tick   = FALSE)

box(col = "grey80")

par(oldpar)

Monthly temperature boxplots

col_box <- "#E74C3C"

df$month <- as.integer(format(df$datetime, "%m"))
month_labels <- c("Jan","Feb","Mar","Apr","May","Jun",
                  "Jul","Aug","Sep","Oct","Nov","Dec")

oldpar <- par(mar = c(4, 4.5, 3, 1), family = "sans", cex.lab = 0.95, cex.axis = 0.85)

boxplot(
  temp_dry_c ~ month,
  data     = df,
  col      = adjustcolor(col_box, alpha.f = 0.40),
  border   = col_box,
  names    = month_labels,
  xlab     = "Month",
  ylab     = "Dry-bulb temperature (\u00b0C)",
  main     = "Temperature Seasonality \u2014 Porto Alegre (A801), 2023",
  outline  = FALSE,
  whisklty = 1,
  medlwd   = 2,
  axes     = FALSE
)

axis(1, at = 1:12, labels = month_labels, las = 1)
axis(2, las = 1)
box(col = "grey80")

par(oldpar)

Wind speed distribution

col_wind <- "#2C3E50"

ws <- df$wind_speed_ms[!is.na(df$wind_speed_ms) & df$wind_speed_ms >= 0]

oldpar <- par(mar = c(4, 4.5, 3, 1), family = "sans", cex.lab = 0.95, cex.axis = 0.85)

h <- hist(ws, breaks = 30, plot = FALSE)

plot(h,
     col    = adjustcolor(col_wind, alpha.f = 0.55),
     border = "white",
     xlab   = "Wind speed (m/s)",
     ylab   = "Frequency",
     main   = "Wind Speed Distribution \u2014 Porto Alegre (A801), 2023",
     axes   = FALSE)

abline(v   = mean(ws),
       col = "#E74C3C",
       lwd = 2,
       lty = 2)

axis(1, las = 1)
axis(2, las = 1)
box(col = "grey80")

legend("topright",
       legend = paste0("Mean: ", round(mean(ws), 1), " m/s"),
       col    = "#E74C3C",
       lwd    = 2,
       lty    = 2,
       bty    = "n",
       cex    = 0.85)

par(oldpar)

Filtering and subsetting

inmet_read() applies filters to each station file before combining all stations and years, reducing the size of the final object held in memory.

# Only summer months (December–February), temperature and humidity
summer <- inmet_read(
  years      = 2023,
  stations   = "A801",
  start_date = "2023-12-01",
  end_date   = "2023-12-31",
  variables  = c("temp_dry_c", "humid_rel_pct")
)

# Multiple stations across Rio Grande do Sul
rs_stations <- inmet_read(
  years    = 2023,
  stations = c("A801", "A802", "A827"),
  variables = c("temp_dry_c", "precip_mm")
)

Extracting CSV files

If you need the raw CSV files for use in other tools, inmet_extract() copies them out of the ZIP archives to a directory of your choice.

inmet_extract(
  years   = 2023,
  out_dir = file.path(tempdir(), "inmet_csv")
)

Cache management

# See what is cached and how large each file is
inmet_cache_status()

# Remove all cached archives to free disk space
inmet_cache_clear()

Variable reference

The table below lists all meteorological columns returned by inmet_read().

| Column | Description | Unit | |---|---|---| | temp_dry_c | Dry-bulb (air) temperature | °C | | temp_max_c | Maximum air temperature reported by the source format | °C | | temp_min_c | Minimum air temperature reported by the source format | °C | | temp_max_prev_c | Maximum air temperature in the previous hour | °C | | temp_min_prev_c | Minimum air temperature in the previous hour | °C | | temp_dew_c | Dew-point temperature | °C | | temp_dew_max_c | Maximum dew-point temperature | °C | | temp_dew_min_c | Minimum dew-point temperature | °C | | precip_mm | Total precipitation | mm | | pressure_station_hpa | Atmospheric pressure at station level | hPa | | pressure_max_hpa | Maximum pressure (previous hour) | hPa | | pressure_min_hpa | Minimum pressure (previous hour) | hPa | | humid_rel_pct | Relative humidity | % | | humid_rel_max_pct | Maximum relative humidity (previous hour) | % | | humid_rel_min_pct | Minimum relative humidity (previous hour) | % | | wind_speed_ms | Wind speed | m/s | | wind_dir_deg | Wind direction | degrees | | wind_gust_ms | Wind gust speed | m/s | | radiation_kjm2 | Global solar radiation | kJ/m² |

Data source

All data are provided by INMET and are freely available. The package does not redistribute raw data; it only automates retrieval and parsing of files hosted by INMET. Automatic-station observations are raw measurements and should be subjected to quality-control rules appropriate to the intended analysis.

rmet documentation built on Sept. 21, 2026, 5:08 p.m.