Fig. 3: Workflow for Training and Estimating.
From: Machine learning-enhanced high-resolution exposure assessment of ultrafine particles
a Feature selection and target outputs from National Air Pollution Monitoring Network (NABEL) data, including the training dataset (2016–2019, with 5% allocated for hyperparameter tuning) and the test dataset (2020) on Stacking Technique for Ensemble Modeling of Particle Number Concentration (Stem-PNC). b Stacking model structure with base and meta models, using cross-validation to assess generalization. c Integration of Copernicus Atmosphere Monitoring Service (CAMS) and European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis v5 (ERA5) data along with meteorological, traffic, and temporal features. d Spatial extrapolation to estimate 2D PNC distribution, leveraging high-resolution CAMS data and the trained Stem-PNC. NABEL monitoring stations include Basel-Binningen (BAS), Bern-Bollwerk (BER), Beromünster (BRM), Chaumont (CHA), Davos-Seehornwald (DAV), Dübendorf-Empa (DUE), Härkingen-A1 (HAE), Jungfraujoch (JUN), Lausanne-César-Roux (LAU), Lugano-Università (LUG), Magadino-Cadenazzo (MAG), Payerne (PAY), Rigi-Seebodenalp (RIG), Sion-Aéroport-A9 (SIO), Tänikon (TAE), and Zürich-Kaserne (ZUE).
