type of computer numerical model which typically simulates atmospheric chemistry and may give air pollution forecasting
CTM focuses on the stocks and flows of one or more chemical species
CTM is expected to accurately represent the entire cycle for the species of interest, including fluxes (e.g. advection), chemical production/loss, and deposition
The METER database combines public data set aggregation, crowdsourcing, and artificial intelligence to create a freely available, global repository of methane-emitting infrastructure
Existing approaches to inferring spatially fine-grained air quality information mainly fall into two categories:physical methods and data-driven approaches
Proposed a generic neural attention model based on deep neural networks for urban air quality inference. We leverage both records from monitoring stations and various urban data (e.g., meteorology, road networks, POIs), and extract important features that are correlated with air quality
Two methods have been implemented to remove the non-dust part of the PM10 observations during the dust event in order to use them as a dust proxy in a dust assimilation system
The first method uses a conventional regional chemistry transport model, LOTOS-EUROS/non-dust, which simulates the emission, transport, chemistry, and deposition of aerosols mainly related to anthropogenic activities
The second method uses a machine learning model (LSTM) that statistically describes the relations between regular PM10 concentrations (outside dust events) and available air quality and meteorological data
The best results are obtained when using a LSTM model to remove the non-dust part of the PM10 observations, with a posteriori concentrations in good agreement with the measurements
In addition to a better spatial representation of emission sources and especially hotspots, the air quality modeling results show that UrbEm outputs, when compared to a uniform spatial disaggregation, have an impact on NO2 predictions up to 70% for urban regions with complex topographies, which corresponds to a big improvement of model accuracy (FAC2 > 0.5), especially at the source-impacted sites
multi-sensor Earth observation dataset containing georeferenced images in the U.S. labeled for the presence or absence of six methane source facilities