Agricultural activities emit ammonia (NH₃) and particulate matter (PM), yet Korean air-quality monitoring is concentrated in urban areas. We analyzed daily PM₁₀, PM_2.5, NO₂, SO₂, and NH₃ at four rice-paddy sites (March 2023–June 2025) across five calendar-defined cultivation stages, which are confounded with season. We therefore separated winter from early spring, calculated increments relative to paired urban stations, and compared recorded operation days with other days of the same month. PM₁₀ was highest in the preparation stage (46.08 vs. 23.97 μg/m³ in growth management; η² = 0.092), and the March–April maximum persisted after winter was separated and after meteorological adjustment. However, at the three sites without a persistent local coarse-particle source, the early-spring PM₁₀ increment over urban stations was the lowest of all periods (−7.62 μg/m³), while local PM₁₀ and PM_2.5 increments peaked in July–September. In exploratory contrasts, the PM_2.5 increment was 2.3 μg/m³ higher on tillage/leveling days (p = 0.005; q = 0.063). NH₃ was governed mainly by site (partial η²: site 0.256, site × stage 0.145, stage 0.033; site means 8.6–37.0 ppb). The calendar-stage PM contrast largely reflects the regional seasonal cycle, whereas agricultural influences appear as operation day and warm-season local increments and as site-specific NH₃ levels.

Fine particulate matter (PM_2.5) in agricultural areas forms partly from gaseous precursors (NH₃, NO₂, and SO₂), yet in open fields ambient concentrations are strongly modulated by wind-driven dilution, which can obscure precursor–PM_2.5 relationships. This study compared concentration- and loading-rate-based interpretations of these relationships in Korean agricultural areas, evaluating the loading rate—concentration-scaled by a wind-speed-based air exchange rate—as a complementary indicator of pollutant loading at the monitoring site. Using a 547-day record (January 2024–June 2025) from eight monitoring sites, we examined hourly network-average concentrations and loading rates across four subsets (entire period, summer, non-summer, and high-PM events) with Pearson correlation, principal component analysis, and multiple linear regression. Concentration-based correlations among PM_2.5 and its precursors were weak and inconsistent, whereas loading-rate-based ones were strong and uniform (r ≥ 0.816). The stronger loading-rate correlations may reflect the dominant influence of wind speed. Diurnal patterns reinforced this point: concentrations were bimodal and fell during the windy daytime, while loading rates were unimodal and rose over the same hours, showing that lower concentrations need not imply reduced pollutant loading. NH₃ peaked earlier and higher in summer than the other precursors, suggesting a time-lagged rather than simultaneous link with PM_2.5. Multiple linear regression confirmed strong co-variation (R = 0.977–0.997) but was limited by multicollinearity, so its coefficients reflect relative co-variation rather than source contributions. The loading rate is therefore best used not on its own but as a complementary indicator of pollutant loading alongside concentration.