Forecasting air pollution is considered as an essential key for early warning and control management of air pollution, especially in emergency situations, where big amounts of pollutants are quickly released in the air, causing considerable damages. Predicting pollution in such situations is particularly challenging due to the strong dynamic of the phenomenon and the various spatio-temporal factors affecting air pollution dispersion. In addition, providing uncertainty estimates of prediction makes the forecasting model more trustworthy, which helps decision-makers to take appropriate actions with more confidence regarding the pollution crisis. In this study, we propose a multi-point deep learning model based on convolutional long short term memory (ConvLSTM) for highly dynamic air quality forecasting. ConvLSTM architectures combines long short term memory (LSTM) and convolutional neural network (CNN), which allows to mine both temporal and spatial data features. In addition, uncertainty quantification methods were implemented on top of our model's architecture and their performances were further excavated. We conduct extensive experimental evaluations using a real and highly dynamic air pollution data set called Fusion Field Trial 2007 (FFT07). The results demonstrate the superiority of our proposed deep learning model in comparison to state-of-the-art methods including machine and deep learning techniques. Finally, we discuss the results of the uncertainty techniques and we derive insights.

The use of low-cost Wireless Sensor Networks (WSNs) for air quality monitoring has recently attracted a great deal of interest. Indeed, the cost-effectiveness of emerging sensors and their small size allow for dense deployments and hence improve the spatial granularity. However, these sensors offer a low accuracy and their measurement errors may be significant due to the underlying sensing technologies. The main aim of this work is to reconsider and compare some regression approaches to assimilation ones while taking into account the intrinsic characteristics of dense deployment of low cost WSN for air quality monitoring (high density, numerical model errors and sensing errors). For that, we propose a general framework that allows the comparison of different strategies based on numerical simulations and an adequate estimation of the simulation error covariances as well as the sensing errors covariances. While considering the case of Lyon city and a widely used numerical model, we characterize the simulation errors, conduct extensive simulations and compare several regression and assimilation approaches. The results show that from a given sensing error threshold, regression methods present an optimal sensor density from which the mapping quality decreases. Results also show that the Random Forest method is often the best regression approach but still less efficient than the BLUE assimilation approach when using adequate correction parameters.

Monitoring air pollution plumes in emergency situations (industrial accidents, natural disasters, deliberate terrorist releases, etc.) becomes an issue of utmost importance in our society given the dramatic effects that the released pollutants can cause. Considering these situations, the pollution plume is strongly dynamic leading to a fast dispersion of pollutants in the atmosphere. Thus, the need for real-time response is very strong and a solution to get precise mapping of pollution dispersion is required to mitigate risks. However, monitoring and forecasting air quality in real time in such situations remains a highly challenging endeavour. In this paper, we suggest a systemic approach for monitoring dynamic air pollution based on aerial sensing (sensors mounted on UAVs). The proposed framework consists of a cycle with feedback loop which will constantly combine a spatio-temporal forecasting model based on a convolutional long short term memory (ConvLSTM) network with a data assimilation technique to get accurate pollution maps, while adjusting at each time the trajectories of drones following uncertainty forecasts. Our solution was evaluated and validated using a highly dynamic real world data set namely Fusion Field Trial 2007 (FFT07). The proposed strategy, together with the obtained evaluation results, are presented, and carefully analyzed.