Machine-learning models to replicate large-eddy simulations of air pollutant concentrations along boulevard-type streets

Abstract

Abstract. Running large-eddy simulations (LESs) can be burdensome and computationally too expensive from the application point of view, for example, to support urban planning. In this study, regression models are used to replicate modelled air pollutant concentrations from LES in urban boulevards. We study the performance of regression models and discuss how to detect situations where the models are applied outside their training domain and their outputs cannot be trusted. Regression models from 10 different model families are trained and a cross-validation methodology is used to evaluate their performance and to find the best set of features needed to reproduce the LES outputs. We also test the regression models on an independent testing dataset. Our results suggest that in general, log-linear regression gives the best and most robust performance on new independent data. It clearly outperforms the dummy model which would predict constant concentrations for all locations (multiplicative minimum RMSE (mRMSE) of 0.76 vs. 1.78 of the dummy model). Furthermore, we demonstrate that it is possible to detect concept drift, i.e. situations where the model is applied outside its training domain and a new LES run may be necessary to obtain reliable results. Regression models can be used to replace LES simulations in estimating air pollutant concentrations, unless higher accuracy is needed. In order to have reliable results, it is however important to do the model and feature selection carefully to avoid overfitting and to use methods to detect the concept drift.

Learning Calibration Functions on the Fly: Hybrid Batch Online Stacking Ensembles for the Calibration of Low-Cost Air Quality Sensor Networks in the Presence of Concept Drift

“…Given a stream of input-target pairs (X, y) the concept is defined as the joint probability distribution P t (X, y) at time t and can be expressed as follows: P t (X, y) = P t (y, X) * P t (X) where P t (y, X) is the posterior probability of the target y given input X and P t (X) is the prior distribution of the input space [28]. A concept drift occurs when P t+1 (X, y) = P t (X, y) and can be explained as the spatially or temporally related changes in the characteristics of features X, which affect the performance of a model that infers the target y, therefore leading to the need of its recalibration [29]. Such a drift can be further divided into three categories according to the factor of the above equation that shifts: (1) P t+1 (X) = P t (X), called a virtual drift, because it does not cause performance degradation, and may be associated with seasonal changes of the features; (2) P t+1 (y, X) = P t (y, X), called a real drift as it effectively alters the distribution between input-target pairs; and (3) both real and virtual drifts occurring at the same time.…”