rmet 0.4.0

New features

• inmet_monthly() aggregates inmet_daily() output into monthly values per station (mean and absolute extremes of temperature, precipitation and rain days, humidity, wind, radiation), with a min_days completeness rule.

• inmet_et0() adds the FAO-56 Penman-Monteith reference evapotranspiration (et0_mm) to daily data; it reproduces the FAO-56 worked example 18. With method = "hargreaves" it needs only temperature, and estimate_radiation = TRUE fills missing radiation from the temperature range. inmet_design_et0() gives the design value (a percentile, by default 0.8) per group.

• inmet_pet_thornthwaite() adds monthly potential evapotranspiration from temperature, and inmet_water_balance() computes the Thornthwaite-Mather soil water balance (storage, actual evapotranspiration, deficit and excess).

• inmet_fill() fills short gaps of hourly or daily data by linear interpolation and the rest with the typical value of the day of the year, marking every filled value in <variable>_filled. Rainfall and wind direction cannot be filled.

• inmet_indices() computes yearly or monthly climate indices from daily data (ETCCDI-style temperature and precipitation indices, dry and wet spells, growing degree days), inmet_spi() the Standardized Precipitation Index, and inmet_normals() and inmet_anomaly() climatological normals and anomalies.

• inmet_nearest() accepts many locations at once (for example, municipality centroids), inmet_stations_within() selects the stations inside a bounding box or an sf polygon, and inmet_interpolate() estimates a variable at other locations by inverse-distance weighting.

• plot() draws rmet_hourly, rmet_daily and rmet_monthly data, and inmet_convert_radiation() converts between radiation units.

• inmet_derive() adds vapour pressure, vapour pressure deficit, wind components (wind_u_ms, wind_v_ms) and the heat index to hourly data, ignoring values flagged by inmet_qc().

• inmet_availability() lists the stations and periods in the cached archives from the file names alone, without reading any data.

• inmet_cache_dir() gives the default cache directory, which can now be changed for a whole session with options(rmet.cache_dir = ) or the environment variable RMET_CACHE_DIR. All dest_dir and cache_dir arguments use it by default.

• The results of inmet_read(), inmet_get(), inmet_daily() and inmet_monthly() have the classes rmet_hourly, rmet_daily and rmet_monthly (all also rmet_data and data.frame), with a compact print(), a summary() per station and as.data.frame(). They remain ordinary data frames, and the example dataset has the new class.

• New vignette: “Monthly Values, Evapotranspiration and Derived Variables”.

Improvements

• inmet_daily(), inmet_qc() and inmet_completeness() return an empty result for the empty data frame that inmet_read() and inmet_get() give when no data is found, instead of failing with a message about datetime.
• The requests for the station catalogue and for the server version of an archive are tried up to three times before giving up.
• inmet_nearest() reports stations without coordinates that it ignores.

Performance

• The data rows of the station files are now parsed by compiled C code (without new dependencies) from the raw bytes read from the archive. Reading a full year (567 files, 4.2 million rows) takes 21 s instead of 134 s, and 120 stations take 4.8 s instead of 22 s, with identical results: the new parser was compared with the previous one on 125 real files from every kind of year (2000 to 2026), 853,000 rows, and agrees exactly.

rmet 0.3.0

Important corrections

• The bundled station catalogue was wrong in rmet 0.1.0 to 0.2.0. Only 8 of its 117 rows matched INMET’s official catalogue: 67 codes do not exist and 42 pointed to a different station in another state (for example, A801 was listed as Sao Luis, MA instead of Porto Alegre, RS). In 0.1.0 it was the only source used by inmet_stations(); in 0.2.0 it was used offline, with source = "bundled", and in examples. It is now a snapshot of the official catalogue (669 stations, all columns), regenerated from the INMET API by data-raw/stations.R, and tests check it against known stations. Results obtained from the old catalogue should be re-checked.
• Station start_date and end_date were one day early: the API timestamps (e.g. 2000-09-21T21:00-03:00) were truncated instead of converted. They are now UTC calendar dates (2000-09-22 for A801).
• The example dataset rmet_example was built with rmet 0.1.0 and had the dew-point extremes in the temp_max_c and temp_min_c columns, and timestamps in Sao Paulo time. It has been regenerated with the current parser (UTC timestamps, temp_dew_max_c and temp_dew_min_c).
• Hours of the current year were duplicated. INMET’s current-year archive holds cumulative versions of each station file (for example January to August and January to September 2026), and every version was read: for A801 in 2026, 5,747 of 12,214 rows were duplicates, which doubles any sum over those hours. Each station and hour is now returned once, from the most recent version (INMET occasionally revises values between versions). As a consequence, inmet_read() output is now sorted by station code and time.
• station_name was NA for files whose header reads ESTAC?O:, a damaged spelling present in INMET’s 2019 archive.
• Date columns are now parsed with a format chosen by exact pattern. Trying formats in turn could read a day-first date such as 04-01-2020 as year 4.
• Cached archives of the current and previous year were never refreshed, although INMET keeps appending data to them. inmet_download() (and inmet_get()) now check the server version of these archives and download updates (check_updates = TRUE).
• The vignette computed daily values with as.Date(datetime), i.e. UTC days, and daily extremes from instantaneous temperatures. It now uses inmet_daily().

New features

• inmet_daily() builds daily summaries per station and local day: maximum and minimum temperature from the previous-hour extremes, means, totals, gusts and radiation, with a min_hours completeness rule and an n_hours column. By default (tz = "auto") each station uses its own local time.
• inmet_qc() flags implausible hourly values: out of physical range, internally inconsistent, isolated spikes, and stuck sensors. Flags are stored in <variable>_flag columns (or the values removed), summarised in the qc_summary attribute, and respected by inmet_daily() and inmet_completeness().
• inmet_completeness() reports the percentage of valid hours per station and month or year.
• inmet_timezone() returns the historical local time zone of Brazilian stations; inmet_stations() gains a time_zone column.
• inmet_nearest() finds the stations closest to a location.
• inmet_as_sf() converts stations or data to sf points (requires the suggested package sf).
• inmet_read() and inmet_get() gain drop_empty; FALSE keeps hours in which every measurement is missing.
• inmet_extract() gains stations to extract only some station files.
• inmet_cache_status() reports when each archive was downloaded.
• New vignette: “Time Zones, Missing Hours and Daily Data”.

Performance and reliability

• Station files are read directly from the archives, without extracting them to disk, and parsed with vectorised code. Reading a full year (545 stations, 4 million rows) takes about half the time of 0.2.0, with identical values.
• Downloads check the final size against the size announced by the server, so truncated files are no longer accepted, and recover when a server ignores a resume request instead of corrupting the archive.
• Files are decoded line by line as UTF-8 or Latin-1, so archives in either encoding are read correctly.
• All measurement columns are now double (some were integer before).

Internal

• Code reorganised into validate.R, parse.R, read.R, download.R, stations.R, daily.R, qc.R, completeness.R, timezone.R, spatial.R and cache.R, with comments on the INMET format. utils.R was removed.

rmet 0.2.0

New features

• Added inmet_get() as the recommended one-call download-and-read workflow.
• inmet_stations() can now retrieve and cache the complete official INMET automatic-station catalogue. New filters cover operational status and provider entity; a bundled catalogue remains available as an offline fallback.
• Station metadata now includes start_date, end_date, status, entity, wsi_code, and oscar_code. The existing start_year column is retained.
• Cache inspection reports partial downloads, and cache clearing can remove the cached station catalogue.

Correctness and reliability

• Changed the default output time zone from "America/Sao_Paulo" to "UTC". INMET historical files record UTC, while Brazilian stations span multiple time zones. Users can still request any Olson time zone explicitly.
• Fixed column matching that could classify dew-point maximum and minimum temperatures as air-temperature maximum and minimum values.
• Added support for station codes longer than four characters.
• Added strict validation for years, dates, flags, directories, and time zones.
• Added support for 2400 UTC timestamps and explicit warnings for unparseable timestamps.
• Downloads now reuse valid archives without contacting the server, write to .part files, resume interrupted transfers, validate ZIP structure, and only then publish the final archive.
• Temporary extraction now uses an isolated directory that is always cleaned. Unsafe archive paths are rejected.
• Corrected the package HTTP user agent and added HTTP timeout/error handling.

Documentation

• Made install.packages("rmet") the primary installation instruction.
• Corrected station-column names in the vignette and clarified timestamp, filtering, caching, and raw-data quality behavior.

rmet 0.1.0

• Initial CRAN release with archive download, parsing, extraction, station lookup, and cache-management helpers.