airmosaic-ai

Homogenized daily PM2.5: Bai et al. (2020)

English 中文

This product appendix explains source identity and scientific suitability. Acquisition is in the PM2.5 skill. The following release/file evidence was reviewed on 2026-10-26.

Source and coverage

Item Description
Dataset Bai et al. (2020), A homogenized daily in situ PM2.5 concentration dataset in China during 2015–2019
Publication PANGAEA DOI 10.1594/PANGAEA.917557, published 2020-05-18
Method ESSD 12, 3067–3080, DOI 10.5194/essd-12-3067-2020
License CC BY 4.0; cite both release and method paper
Provenance National station measurements → CNEMC publication → author collection/QC/processing → PANGAEA
Coverage 2015-01-01–2019-12-31; 1,309 daily series, µg/m³
Access Public ZIP from Download dataset; no account or API key

The data include gap-filling, series merging, homogeneity testing and bias adjustment. They are distinct from published hourly archives. Splicing later years does not establish a consistent processing history.

File identity

File in release Structure
PM2.5 data in China/Homogenized_PM25_data.txt First three lines: title, frequency/unit, date range; line 4 blank; line 5: Date and 1,309 series columns; daily wide table follows
PM2.5 data in China/Goelocation.txt Preserve provider spelling; two explanatory lines, blank line 3, line 4: Site_ID Latitude Longitude; location rows follow

UTF-8 whitespace-delimited text is not comma-separated CSV. Inspect headers/field counts rather than assuming fixed skips after a release change. The checked 1,309 location IDs uniquely matched concentration columns. Latitude/longitude directions were stated, but the CRS was not explicitly declared; do not label EPSG:4326 as verified.

S0001 etc. are product-series IDs, not confirmed original national station codes. No mapping to codes such as 1001A was supplied. Series merging may span relocation; present coordinates do not establish historical identity or position.

Processing uncertainty and use

Section 3 of the paper describes outlier/missing-value treatment, diurnal-cycle-constrained EOF reconstruction, reference-series change-point detection/distribution adjustment, series merging and postprocessing. Its parameters are study-specific.

Imputation improves continuity without adding independent measurements and may change peaks, variance and temporal dependence. Reference selection affects detected breaks and adjustments; homogenization does not eliminate all measurement/representativeness error. Long-series selection and relocation merging change sample composition. The product cannot independently prove its own reconstruction accuracy; comparisons with national observations must account for shared inputs and station/time overlap.

Use daily changes/comparisons only when the processing assumptions suit the question. Regulatory daily-validity review, hourly events and original missingness require their corresponding original records.

Normalization and handoff

  1. Retain DOI, acquisition date and ZIP checksum, preserving the original release.
  2. Validate headers, dates, unique series and numeric fields over the intended scope; one seven-day sample does not validate the full period.
  3. Join locations by explicit series ID with uniqueness/unmatched checks. Leave original national station codes unknown.
  4. Retain source/DOI, series_id, original date, concentration/unit, coordinates/source and product_kind=homogenized_daily in long form.
  5. The package lacks per-value imputation/adjustment flags, valid observed hours and explicit time zone/CRS. Do not mark every value unfilled, set n_valid_hours=24 or claim daily compliance with an original-observation 20-hour rule.
  6. Do not reapply hourly-to-daily aggregation. Define daily-coverage criteria for monthly/annual analysis; resolve date semantics and spatial/physical support before matching other products.
  7. Obtain original evidence if the question needs hourly completeness, original station IDs or per-value processing flags; otherwise respect the product’s limitations.

The download example records release and small-sample content checks.

This site is open source. Improve this page.