MP-GERE — Field-measurement dataset for "Detection-threshold drift distorts global assessments of microplastic exposure"
Description
Dataset accompanying the manuscript "Detection-threshold drift distorts global assessments of microplastic exposure" (11,268 field-standardized measurements from 10,289 publications across 102 countries).
Version: v2.1.1 (2026-09-02) — README aligned with the final manuscript title and reporting policy; data files unchanged from v2.1 (2026-08-29, production re-extraction of 10,289 papers with the GLM-5.3 extraction model).
Files
| File | Records | Description |
|---|---|---|
field_measurements_11268.jsonl |
11,268 | Normalized field-measurement records (pre-deduplication), one JSON object per line |
harmonized_field_records_10304.csv |
10,304 | Harmonized records after de-duplication (see Record counts below) |
ecological_effect_db.json |
27 | Ecological effect-size estimates from 8 independent meta-analyses (2019–2026) |
analysis_outputs/ |
— | All analysis outputs used in the manuscript (see Analysis outputs) |
Record counts — how the three numbers relate
| Stage | n | Explanation |
|---|---|---|
| Field measurements (pre-dedup) | 11,268 | All direct field measurements after excluding nanoplastic-derived records (1,899) and cross-column extraction errors (13); title count of the manuscript |
| Harmonized (post-dedup) | 10,304 | After de-duplication of records sharing study, location and concentration (harmonization key) |
| Analysis set | 9,460 | After excluding 1,808 records flagged as unreliable/ invalid/ physically impossible by the exclusion policy (below) |
All downstream statistics (exceedance, trends, spatial evidence) are computed on the analysis set (9,460) except the sensitivity exceedance (pre-deduplication, n = 6,770) and the planetary-boundary subset (n = 5,833, records with valid concentration and matrix-matched PNEC).
Column dictionary (harmonized_field_records_10304.csv)
| Column | Type | Meaning |
|---|---|---|
record_id |
str | Unique record identifier (MP-GERE-…) |
study_id / doi |
str | Source publication DOI |
year |
int | Publication year (records 1880–2026; 26 records < 1990) |
country |
str | Country of sampling, if reported |
latitude / longitude |
float | Coordinates if reported; empty otherwise |
matrix |
str | Environmental matrix: sediment, freshwater, marine_water, soil, wastewater, other (or None) |
raw_value / raw_unit |
float / str | Value and unit as reported in the source (124 distinct raw unit strings in the full corpus; 46 in the field set) |
standard_value / standard_unit |
float / str | Standardized concentration — particles·kg⁻¹ (solids) or particles·L⁻¹ (water) |
polymer_type |
str | Reported polymer (e.g. PE, mixed) |
shape |
str | Reported shape (fiber, fragment, pellet, film, mixed, …) |
size_min_um / size_max_um |
float | Particle-size range in μm (lower bound = detection threshold; 35% of records lack it) |
analytical_method |
str | Detection method (visual, FTIR, Raman, μFTIR, …) |
source_page |
int | Page of the source document where the value appears |
source_quote |
str | Verbatim evidence excerpt used for extraction verification |
confidence |
float | Extraction confidence 0–1 |
needs_manual_review |
bool | Flagged for manual review during extraction |
standardization_path |
str | Trace of the unit-standardization steps applied |
bias_tags |
str | Methodological bias tags (visual_only, small_particle_excluded, …) |
qa_flags |
str | Quality flags, |-separated (see below) |
qa_flags semantics
Flags used by the exclusion policy (common/analysis_policy.py in the code repository):
| Flag prefix | Meaning | Excluded from analysis? |
|---|---|---|
mass_conversion_blocked |
Mass-based unit could not be converted to particles without density/size assumptions | Yes (1,568 records) |
schema_error |
Failed schema validation (e.g. invalid land-use enum) | Yes |
non_mp_substance |
Record refers to a non-microplastic substance | Yes (260) |
std_dev_as_value |
Value actually a standard deviation | Yes (24) |
outlier: above maximum |
Physically impossible concentration (above matrix-specific plausibility ceiling) | Yes |
outlier: below minimum |
Concentration below the matrix-specific plausibility floor | No (retained — open-ocean abundances routinely fall below common floors; excluding them would bias exceedance upward) |
unknown_unit |
Raw unit unrecognized | Depends on downstream availability |
cross_matrix_blocked |
Matrix cross-conversion not applicable | Not excluded |
Unit standardization
All raw units (124 distinct strings in the corpus, 46 in the field set) were mapped to particles·kg⁻¹ (solids) and particles·L⁻¹ (water). Mass-to-particle conversions were applied only when polymer density and size distribution were reported; otherwise the record was flagged mass_conversion_blocked and excluded rather than estimated. Superscripts and suffixes (e.g. ×10³, items/L⁻¹) are normalized.
Analysis outputs (analysis_outputs/)
| File | Contents | Manuscript claim |
|---|---|---|
planetary_boundary_assessment.json |
PEC/PNEC assessment, CV2 (proportion of sites exceeding), per-matrix stats | 41.6% exceedance (n = 5,833; median 0.46; P95 70) |
temporal_analysis.json |
Linear mixed-effects trend model + per-matrix trends | +29.5% yr⁻¹ (n = 7,328, p < 0.0001; quadratic model) |
sensitivity_matrix.json |
Exceedance by matrix × detection threshold (coarse/fine/unknown) | 17% coarse / 54% fine / 42% overall (n = 6,770) — Table 1 |
spatial_evidence.json |
Per-country exceedance hotspots (≥3 geo-referenced records per country) | 1,984 records / 53 countries; Brazil 95%, USA 84%, China 82%, UK 81% |
size_abundance.json |
log-log OLS of concentration on minimum detected size | Slope −1.034 (n = 5,502, p < 0.0001), 10.8× per tenfold size decrease |
Exclusion of exploratory spatial-model outputs
The manuscript does not use the spatial interpolation model. Its outputs (spatial_report.json, global_exposure_surface.csv, ters_exposure_surface.tif, ters_uncertainty_surface.tif) are excluded from this release: the exposure surfaces carried placeholder uncertainty values (grid_std = 0.1) and were validated on only 15 leave-one-site-out points. The descriptive spatial evidence (spatial_evidence.json, country-level hotspots from measured coordinates) is included because it is used in the manuscript. Users needing the exploratory surfaces can regenerate them from the code repository (analysis/spatial_model.py, marked as exploratory there).
License and citation
Dataset: CC BY 4.0 (attribution to the authors of this manuscript). Underlying measurement values remain the property of their original studies (DOIs retained in each record).
Files
mp_zenodo_v2.1.zip
Additional details
Related works
- Is supplement to
- Software: 10.5281/zenodo.22255450 (DOI)