Abstract
<jats:p>Abstract. Ammonia (NH3) is a critical precursor to fine particulate matter (PM2.5) pollution. However, long-term observations at the county scale with high temporal resolution remain scarce. We present a four‑year (2021–2025) hourly NH3 dataset from 10 monitoring sites across Quzhou County, a typical intensive agricultural county on the North China Plain. To address missing values caused by instrumental interruptions, we evaluated three machine learning models and selected XGBoost for data imputation. The reconstructed dataset reveals that the annual mean NH3 concentration in Quzhou County exhibited an overall decreasing trend following an initial rise, increasing from 31 ppb in 2021–2022 to a peak of 36.8 ppb in 2022–2023, and subsequently declining to 26.8 ppb by 2024–2025. A persistent north–south gradient highlights substantial spatial heterogeneity, with mean concentrations ranging from 20.9 ppb at northern cropland sites to 52.8 ppb at southern livestock hotspots. Temporally, NH3 exhibits a bimodal seasonal cycle peaking in March and June, and a diurnal maximum between 07:00 and 10:00 local time. SHAP analysis identified water vapor pressure, air temperature, and wind speed as the primary meteorological controls. NH3 concentrations were elevated when vapor pressure exceeded 1.03 kPa and air temperature surpassed 12.4 °C, and were suppressed when wind speed exceeded 1.12 m s⁻¹. This dataset provides a robust observational foundation for evaluating emission reductions, validating satellite products, and informing air quality models.</jats:p>