安庆市颗粒物浓度演变规律与气象驱动因素分析

An Analysis of the Spatiotemporal Evolution Patterns of Particulate Matter Concentration and Its Meteorological Driving Factors in Anqing City

摘要: 为探究安庆市复合型大气污染的时空演变特征及其气象驱动机制，并揭示其对农业生产的影响，本研究基于2023~2025年逐小时PM_2.5浓度及同期气象观测数据，综合运用数理统计、多尺度分析、K-means聚类、随机森林特征重要性评估、滞后相关分析及部分依赖图等方法，系统分析了颗粒物浓度的时空分布规律、气象驱动机制及其与农业生产的关联。结果表明：安庆市PM_2.5浓度呈现显著的季节差异和双峰型日变化特征，冬季为污染高发期，主城区及工业影响区浓度明显高于郊区农业生产区域，对农田光照和作物生长的影响尤为突出；K-means聚类将污染过程划分为低、中、高三类典型情景，高污染情景表现出明显长尾分布，对农业区的影响更为持久，可能导致水稻、小麦等主要作物光合作用下降和叶片污染。随机森林模型显示，相对湿度和风速是影响PM_2.5浓度的最主要气象驱动因子，二者贡献度显著高于其他因子；滞后相关分析揭示气象因子对PM_2.5浓度存在3~12 h的明显滞后效应。部分依赖图进一步表明，相对湿度超过70%、风速低于1.5 m/s时，PM_2.5浓度呈显著非线性增加，且高湿度与低风速具有协同加剧效应，会显著加重农业区域的污染累积。随机森林模型在预测性能上优于决策树和GBDT，为最优模型。本研究解析了气象要素对颗粒物污染的非线性响应与滞后叠加机制，明确了不同污染情景下PM_2.5对农业生产的潜在影响，包括作物产量降低和品质受损，可为安庆市农业区重污染精准预警、农田防护措施制定和精细化管控提供科学依据。

Abstract: To investigate the spatiotemporal evolution characteristics of composite air pollution and its meteorological driving mechanisms in Anqing City, and to reveal its impact on agricultural production, this study utilized hourly PM_2.5 concentration data and concurrent meteorological observations from 2023 to 2025. Multiple methods including mathematical statistics, multi-scale analysis, K-means clustering, random forest feature importance assessment, lagged correlation analysis, and partial dependence plots were comprehensively employed to systematically analyze the spatiotemporal distribution patterns of particulate matter concentrations, meteorological driving mechanisms, and their associations with agricultural production. The results show that PM_2.5 concentrations in Anqing City exhibit significant seasonal differences and a typical bimodal diurnal variation pattern, with winter being the high-pollution season. Concentrations in the main urban area and industrial influence zones are markedly higher than those in suburban agricultural production areas, exerting particularly pronounced effects on farmland light availability and crop growth. K-means clustering divided the pollution processes into three typical scenarios: low, medium, and high pollution. The high-pollution scenario displays an obvious long-tail distribution and has a more persistent impact on agricultural areas, potentially leading to reduced photosynthesis and leaf contamination in major crops such as rice and wheat. The random forest model indicates that relative humidity and wind speed are the most important meteorological driving factors affecting PM_2.5 concentrations, with their contributions significantly higher than other factors. Lagged correlation analysis reveals a clear lagged effect of meteorological factors on PM_2.5 concentrations within 3~12 hours. Partial dependence plots further demonstrate that when relative humidity exceeds 70% and wind speed drops below 1.5 m/s, PM_2.5 concentrations increase significantly in a nonlinear manner, with a synergistic aggravating effect between high humidity and low wind speed, which substantially intensifies pollution accumulation in agricultural regions. The random forest model outperforms decision trees and GBDT in predictive performance and is identified as the optimal model. This study quantitatively elucidates the nonlinear response and lagged superposition mechanisms of meteorological elements on particulate pollution, and clarifies the potential impacts of PM_2.5 on agricultural production under different pollution scenarios—including reduced crop yield and quality degradation—thereby providing scientific support for precise early warning of heavy pollution in agricultural areas of Anqing City, the formulation of farmland protection measures, and refined control strategies.