airmosaic-ai

ERA5: choosing and interpreting meteorological reanalysis

English 中文

ERA5 supports meteorological background, research covariates and large-scale input fields. Choose variables by physical meaning, spatial/time support and uncertainty before aggregation or matching. Acquisition steps belong in the ERA5 skill.

Source and product identity

ERA5 is produced by ECMWF for the Copernicus Climate Change Service. The IFS model and data assimilation combine historical observations; CDS is the distribution channel. It is neither raw station data nor simple interpolation of ground observations. Individual fields may originate from analysis or forecast steps. See the official data documentation.

Retain dataset ID, single/pressure/other levels, parameter identity, grid, times and release. ERA5, ERA5-Land and daily-statistics products are distinct. A finer export grid does not create independent information or establish effective resolving power.

Uncertainty

Source Implication for research
Observation errors and changing coverage Assimilation constraints differ by region, period and variable; examine discontinuities and known issues in long records
Model physics and assimilation Systematic biases remain; compare suitable variables with independent observations in the study region
Grid representativeness Terrain, coastlines and local processes differ from station support; record interpolation, distance, elevation and mask choices
Sampling and processing UTC/local days, accumulation windows, regridding and missing steps can change statistics; convert according to each variable’s meaning

The official uncertainty guidance describes EDA spread as relative uncertainty information with incomplete representation of systematic errors. Do not treat it as a calibrated error bar for every field or assume one error percentage for all variables.

Constructing comparable variables

  1. Preserve original units, levels, UTC times, coordinates, missing flags and preliminary/final release status. Record updates such as ERA5T explicitly.
  2. Convert by quantity: temperature can change from K to °C; precipitation requires distinguishing depth, accumulation and rate first. Retain originals and formulas.
  3. Identify instantaneous, averaged and accumulated time support. Check accumulation start/reset and forecast steps; do not difference every accumulation field automatically.
  4. Form unique variable/level/time/grid keys; inspect duplicates, missing times, longitude conventions, latitude order and masks.
  5. Define UTC or local-day windows and expected sampling. Choose means, totals, extremes or time weights by quantity. State incomplete-sampling rules and sensitivity; the PM2.5 station 20-hour rule is not a meteorological-product rule.
  6. For station or model comparisons, specify spatial extraction and time alignment. Surface fields do not replace vertical profiles; ERA5 files require separate model-input adaptation.
  7. Report transformations, valid samples, grid changes, missingness and release versions. Fit bias correction or feature scaling only on training periods when used in prediction.

A single UTC field cannot establish a daily statistic. Independent observation comparison is needed to assess local scientific suitability.

AOD is a separate product choice

The ERA5 single-level catalogue form checked on 2026-10-30 had no aerosol optical-depth option. Use ERA5 for the declared meteorological variables; an AOD covariate requires its actual separate product identity, for example CAMS EAC4. Access through a related provider does not make the products interchangeable. See the AOD operation; only public catalogue metadata was checked, not a new data download.

This site is open source. Improve this page.