Particulate Matter (PM10 and PM2.5) in Indian Coal Mining Regions: A Systematic Review of Measured Concentrations, Sources, Health Evidence and Mitigation

Abstract

Protocol: Particulate matter (PM10 and PM2.5) in Indian coal mining regions — a systematic review with quantitative synthesis Registration type: Retrospective (registered after screening and first-reviewer data extraction, before any inferential analysis). See "Timeline and what was already done". Reporting guidelines: PRISMA 2020 (reporting); SWiM (synthesis without meta-analysis). Contact: Sanyam Dadhich, Amrendra Kumar, Nisha Bhardwaj, Department of Mining Engineering, IIT (BHU) Varanasi. 1. Background and aim Opencast coal mining, mine fires and coal transport are major sources of particulate matter (PM) in India's coalfields, but the evidence is spread over short-term, methodologically varied studies. The aim is to synthesise measured PM10 and PM2.5 in Indian coal mining regions and to turn the extracted data into decision-support evidence for coal-sector regulators and operators, while stating how certain each conclusion is. 2. Objectives Describe the evidence base: which coalfields, mine types, site types, seasons, instruments and averaging times have measured PM. Estimate, by coalfield, the share of ambient site means above the 24-hour Indian NAAQS (2009), with uncertainty that accounts for several sites coming from one study. Identify which factors (coalfield, season, site type, size fraction, instrument class, distance from the mine) explain variation in measured PM. Compute, for each coalfield × season × site type with enough data, the proportional reduction in mean PM needed for the mean to reach the standard ("required reduction"), and compare it with the control effects measured or modelled in the literature. Rank coalfield/season/site-type combinations for monitoring and control attention, with an evidence-confidence rating for each. Summarise sources, health and ecological evidence and mitigation evidence narratively. 3. Eligibility criteria Setting: a named Indian coal (or lignite, reported separately) mining area. Exposure/measure: measured PM10, PM2.5, PM1, SPM/RSPM or respirable dust with numerical results. Data: primary field data, monitoring-network data reported by the authors, or a model built on measured data (modelled studies are reported separately and never pooled with measured data). Publication: 2000–2026, English, peer-reviewed or archival. Not eligible: non-coal contexts (other minerals, power plants only, urban/industrial only); no PM concentrations (dust-fall only, gases only, satellite AOD only, emission-factor estimates, leaf/panel deposition); reviews without data. Health, comparison with standards, sources and mitigation are extracted outcomes, not inclusion requirements. 4. Information sources and search SciSpace, Google Scholar, PubMed, Web of Science and Scopus, plus backward and forward citation searching of included papers (OpenAlex, 4 October 2026). Search strings and per-source hit counts are in the Search_Log sheet of the review workbook (deposited with this registration). Grey literature (CPCB, CMPDI, state pollution control boards) was planned but not searched; this is a stated limitation. 5. Selection and data extraction Titles/abstracts screened by the first reviewer; full texts assessed by the first reviewer. Verification (added after the initial screening): a second reviewer independently screens a stratified sample of 214 records (120 random exclusions, 54 re-screen candidates, 40 records advanced to full text; fixed random seed 20261010) blind to the first reviewer's decisions. Agreement is reported as percent agreement and Cohen's kappa per stratum. Double extraction: all included papers are extracted a second time, independently and blind to the first extraction (AI-assisted, from the full-text files), and every discrepancy in key fields is adjudicated by the second reviewer against the source paper. Key fields: coalfield, site type, season, PM fraction, instrument, averaging time, n, mean, SD/min/max, standard used. The use of AI assistance is disclosed in the paper. Agreement is reported before and after adjudication. Papers that repeat another paper's sampling campaign are flagged (Independent dataset = No) and excluded from all counts and models. Full texts that could not be obtained (30 of 103 sought) are listed and are not counted as read. 6. Variables and data classes Site-season rows carry the mean, SD, min, max, n, PM fraction, instrument, averaging time, coalfield, mine type, site type, distance, season and the standard used. Rows are classed as ambient 24-hour-type (primary analysis set), short-term continuous, work-zone/occupational, secondary network, or modelled. Classes are never pooled with each other. Units printed wrongly in source papers are corrected only when the paper's own text or ratio makes the intended unit unambiguous, and the correction is logged. 7. Reporting-quality appraisal No validated risk-of-bias tool exists for this study type. A five-domain reporting-quality screen (sampling design, measurement method, QA/QC, reporting of n and spread, independence of dataset) is applied by two reviewers on the double-extracted set; it describes reporting completeness, not accuracy of results. 8. Synthesis plan (specified before any model is run) Primary analysis set: ambient 24-hour-type site-season means for PM10 and PM2.5 from independent datasets. No inverse-variance meta-analysis, because n and SD are often unreported and averaging time, instrument and season differ. A. Exceedance: proportion of site means above the 24-hour NAAQS 2009 limit (PM10 100, PM2.5 60 µg/m³), by coalfield; 95% CI by cluster bootstrap (resampling datasets, 5,000 replicates). Coalfields with fewer than 3 independent datasets are reported descriptively without a CI. B. Explanatory model: linear mixed-effects model of log(mean) with fixed effects for coalfield, season, site type, PM fraction, instrument class and log(distance), and a random intercept for dataset. Categories with fewer than 5 rows or 3 datasets are merged into "other". Effects are reported as ratios with 95% CI. P-values, if shown, are adjusted by Benjamini–Hochberg and are descriptive, not confirmatory. C. Required reduction: R = max(0, 1 − limit/mean) for each coalfield × season × site type cell with at least 2 independent datasets and 3 rows; median and range reported. This is the reduction in the mean needed for the mean to equal the limit, not a guarantee of daily compliance. D. Priority ranking: cells ranked by median R. Each cell gets an evidence-confidence rating (high / moderate / low / very low) from the number of independent datasets, reporting-quality results, and consistency of direction across datasets. Thresholds are fixed in the analysis script before it is run. E. Scenarios and sensitivity: S0 baseline; S1 without network (CPCB-type) data; S2 without overlap-flagged papers; S3 documented gravimetric method only; S4 WHO 2021 24-hour guidelines (PM10 45, PM2.5 15 µg/m³) instead of NAAQS; S5 without fire-affected sites; S6 measurements from 2015 onward; S7 leave one coalfield out; S8 without papers rated high concern for reporting quality. Conclusions are called robust only if their direction holds in all scenarios. F. Narrative synthesis for sources, health, ecology and mitigation, by coalfield (SWiM). Causal language is not used. Software: R or Python; versions and scripts are deposited. 9. Timeline and what was already done Done before registration: searches; title/abstract screening; full-text selection; first-reviewer extraction; descriptive counts of exceedance for the baseline scenario (including S1–S3 in a draft); citation chasing. Not done before registration: the second-reviewer screening check; the second extraction; models (B); required reduction (C); ranking (D); scenarios S4–S8. 10. Amendments Any later change to this plan will be logged with date and reason and reported in the paper as a deviation from the protocol.