MapLUR

Abstract

Land-use regression (LUR) models are important for the assessment of air pollution concentrations in areas without measurement stations. While many such models exist, they often use manually constructed features based on restricted, locally available data. Thus, they are typically hard to reproduce and challenging to adapt to areas beyond those they have been developed for. In this article, we advocate a paradigm shift for LUR models: We propose the D ata-driven, O pen, G lobal (DOG) paradigm that entails models based on purely data-driven approaches using only openly and globally available data. Progress within this paradigm will alleviate the need for experts to adapt models to the local characteristics of the available data sources and thus facilitate the generalizability of air pollution models to new areas on a global scale. To illustrate the feasibility of the DOG paradigm for LUR, we introduce a deep-learning model called MapLUR. It is based on a convolutional neural network architecture and is trained exclusively on globally and openly available map data without requiring manual feature engineering. We compare our model to state-of-the-art baselines like linear regression, random forests and multi-layer perceptrons using a large data set of modeled NO 2 concentrations in Central London. Our results show that MapLUR significantly outperforms these approaches even though they are provided with manually tailored features. Furthermore, we illustrate that the automatic feature extraction inherent to models based on the DOG paradigm can learn features that are readily interpretable and closely resemble those commonly used in traditional LUR approaches.

Error Prediction of Air Quality at Monitoring Stations Using Random Forest in a Total Error Framework

“…Indeed, spatial features, such as topography, proximity to infrastructure as well as the land use in a buffer area near to air quality monitoring stations, have proven to improve the performance of deep learning-based air pollution predictions [49]. DNN LUR, such as [50] provide an approach of geo-context in an unsupervised way that outperforms other approaches such as RF. For these reasons, spatial error prediction will require the use of either Bayesian DNN or Deep Ensemble.…”

Interpretable Machine Learning Approaches for Forecasting and Predicting Air Pollution: A Systematic Review

“…This means that during the backpropagation process, only the paths that increase the activation of a particular feature or neuron are considered, and paths that reduce the activation are effectively ignored. Steininger et al (2020) used guided backpropagation to visualize the regions of an input image that a neural network is focusing on for NO2 estimation, allowing visualization of which parts of the image the model is paying attention to. This helps in understanding what aspects of the input data have the most influence on the network's decision.…”

“…Only Steininger et al (2020) employed satellite images for air pollution prediction. They used guided backpropagation (GB) as a model-specific method and found that the model prioritizes motorways, trunk roads, and primary roads in map images to predict NO2 concentration.…”

Volunteered Geographic Information

“…Due to the individual differences in daily activities, each person has very individual exposure patterns, which obviously cannot be adequately captured by measurements at a few monitoring stations in the city (improved approach; see Steininger et al 2020) but require person-specific measurements (Dias and Tchepel 2018;Hinwood et al 2007). However, air pollution data from a terrestrial monitoring station can only be considered representative of the surrounding district.…”