airmosaic-ai

CAMS: using reanalysis and forecasts in research

English 中文

CAMS supplies atmospheric-composition context and comparison fields. Interpret EAC4 and operational forecasts as different products with different time structures. Acquisition belongs in the CAMS skill.

Source and product identity

Product Production Research meaning
Global reanalysis EAC4 CAMS/ECMWF atmospheric chemistry model with observational assimilation; distributed through ADS Model-constrained historical composition; examine changing observations, species, levels and validation
Global composition forecasts Assimilated initialization followed by forward integration Retain initialization, lead and system release; operational upgrades affect long-term consistency

Satellite assimilation does not imply a ground measurement for every PM2.5 cell. Surface mass concentration, gas mixing ratio, column total and AOD are different physical quantities. See the official forecast documentation. European regional products require their own specification.

Uncertainty and suitability

Chemical/aerosol processes, transport, initialization, boundaries and emissions introduce uncertainty; disagreement with a station cannot identify a single cause. Assimilated species, vertical sensitivities and coverage differ from surface-monitoring coverage and change over time. Grid cells differ from local stations in terrain, local pollution and boundary-layer support.

Forecast error varies with species, region, initialization and lead. Evaluate leads separately and examine system upgrades; overlapping forecasts are not independent observations. Use product-specific EAC4 validation and global-service reports, rather than applying one error percentage to every field. File checks do not establish independent observational accuracy.

Comparable fields and time series

  1. Fix product, parameter, surface/model/pressure coordinate, grid and units together.
  2. Convert PM2.5 from kg/m³ to µg/m³ by 10^9 only when the metadata declare that mass-concentration unit. Gas mass mixing ratios need different formulas and supporting density/temperature/pressure information.
  3. Retain EAC4 UTC sampling. Key forecasts by product/system, initialization, lead, variable, level and grid; compute valid_time = forecast_reference_time + leadtime explicitly.
  4. Predetermine fixed-initialization or lead-selection rules for overlapping valid times. Evaluation may use only forecasts actually available at the decision time; do not default to averaging initializations.
  5. Define day boundaries and sampling support. Three-hour instantaneous samples at 00,03,…,21 provide eight samples per UTC day, not 24 observed hours. Set coverage and weighting by field definition; station 20-hour thresholds do not apply.
  6. Separate missing files/times, nonfinite values and unit/level errors. Do not substitute another initialization without marking it or fill missing values with zero.
  7. Match station observations only after reconciling UTC/local time, averaging, concentration state, surface definition and spatial support. Fit bias correction on training data and retain independent evaluation. Species/grid/vertical conversion for a transport-model boundary is a separate method.

Do not splice EAC4 and operational forecasts into a homogeneous trend series without studying product and system differences.

This site is open source. Improve this page.