Scholarlink

<p>Ammonia (NH<sub>3</sub>) is commonly known as precursors of fine particulate matter (PM). NH<sub>3</sub> is emitted from livestock and fertilization and the potential reduction of this pollutant is recognized as playing a relevant role in the reduction of atmospheric concentration of PM. To more accurately estimate its impact, geographically and temporally precise data on NH<sub>3</sub> emissions are crucial. In this paper we implement a space (municipalities) and time (hourly) varying emission estimate of NH<sub>3</sub>. The time resolution is obtained by the application of Machine Learning (ML) on in-situ measurements sites allowing to discriminate the effect of the atmospheric turbulence from the real activity level on manure management and fertilization. The space resolution is obtained by applying the ML to the obtained time-profiles considering to the whole domain the most relevant variables for NH<sub>3</sub> emissions. The estimated dynamic emission inventory has been validated by applying the emission profiles to a Chemical and Transport Model (CTM) drastically improving its performance in predicting NH<sub>3</sub> atmospheric concentrations. The proposed approach combines ML techniques and deterministic models showing how the predicted emission rates can be assimilated to the modelling chains maintaining all the physical constraints.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Dynamic emission inventory of ammonia in northern Italy by machine learning

  • 期刊论文

  • 发表日期：

    2025年6月24日

查看全文
  • Alessandro Marongiu,
  • Loris Colombo,
  • Anna Gilia Collalto

摘要

Ammonia (NH3) is commonly known as precursors of fine particulate matter (PM). NH3 is emitted from livestock and fertilization and the potential reduction of this pollutant is recognized as playing a relevant role in the reduction of atmospheric concentration of PM. To more accurately estimate its impact, geographically and temporally precise data on NH3 emissions are crucial. In this paper we implement a space (municipalities) and time (hourly) varying emission estimate of NH3. The time resolution is obtained by the application of Machine Learning (ML) on in-situ measurements sites allowing to discriminate the effect of the atmospheric turbulence from the real activity level on manure management and fertilization. The space resolution is obtained by applying the ML to the obtained time-profiles considering to the whole domain the most relevant variables for NH3 emissions. The estimated dynamic emission inventory has been validated by applying the emission profiles to a Chemical and Transport Model (CTM) drastically improving its performance in predicting NH3 atmospheric concentrations. The proposed approach combines ML techniques and deterministic models showing how the predicted emission rates can be assimilated to the modelling chains maintaining all the physical constraints.