PM и метеорология: всички сравнения, ход и покритие

# PM and Weather: Methods

Study period: 15 February 2024 to 13 October 2026. Local dates and clock labels use Europe/Sofia; UTC timestamps are retained in the measurement tables.

For 2025, station statistics retain all 34,395 original averaged export records, including boundary and irregular-duration records. For 2024 and 2026, the accepted nominal quarter-hour observations are retained, including averages derived from minute measurements. Station summaries give each retained observation equal weight. Standard deviation is sample SD (N-1), describing observation spread, not uncertainty of the mean. Measured zeros are retained; missing observations are not interpolated. Unknown valid-minute counts remain unknown.

Winter groups January, February and December of the displayed year. Incomplete months and seasons describe observed periods only. Coverage and matched-calendar comparisons use only regular nominal quarters (14-16 minutes for exports), mapped by interval midpoint. Irregular-duration records contribute to 2025 station statistics but do not create additional complete quarters. Matched-calendar comparisons use common month/day/hour/quarter-hour positions.

Concurrent PM-weather and Mladost comparisons use separate coverage criteria: 2024/2026 require at least 10 measured PM minutes per quarter, 1080 per day and 45 per hour. The 2025 export-based exception requires 72 regular quarters per day or three per hour. It does not apply to 2024 exports with unknown minute counts. Weather daily means require at least 1080 valid minutes per variable on common slots.

Partial-rank correlations control calendar month and the other weather variables. Their 95% intervals use a seven-calendar-day block bootstrap. High-PM days use a within-year 90th percentile, not a regulatory threshold. Associations do not establish causation. Lufft and Vaisala wind directions differ by about 73 degrees; sector plots alone do not identify transport origin.

Mladost comparisons use identical qualifying dates or hours at both sites. Published agency daily values are not recomputed from rounded hourly values. Hourly source rows on DST-change dates are not used because their clock labels are ambiguous. Fully-zero exclusion is a sensitivity analysis, not instrument-status quality control.