Constructing transferable and interpretable machine learning models for black carbon concentrations.

Black carbon (BC) has received increasing attention from researchers due to its adverse health effects. However, in-situ BC measurements are often not included as a regulated variable in air quality monitoring networks. Machine learning (ML) models have been studied extensively to serve as virtual sensors to complement the reference instruments. This study evaluates and compares three white-box (WB) and four black-box (BB) ML models to estimate BC concentrations, with the focus to show their transferability and interpretability. We train the models with the long-term air pollutant and weather measurements in Barcelona urban background site, and test them in other European urban and traffic sites. Despite the difference in geographical locations and measurement sites, BC correlates the strongest with particle number concentration of accumulation mode (PNacc, r = 0.73-0.85) and nitrogen dioxide (NO2, r = 0.68-0.85) and the weakest with meteorological parameters. Due to its similarity of correlation behaviour, the ML models trained in Barcelona performs prominently at the traffic site in Helsinki (R2 = 0.80-0.86; mean absolute error MAE = 3.90-4.73 %) and at the urban background site in Dresden (R2 = 0.79-0.84; MAE = 4.23-4.82 %). WB models appear to explain less variability of BC than BB models, long short-term memory (LSTM) model of which outperforms the rest of the models. In terms of interpretability, we adopt several methods for individual model to quantify and normalize the relative importance of each input feature. The overall static relative importance commonly used for WB models demonstrate varying results from the dynamic values utilized to show local contribution used for BB models. PNacc and NO2 on average have the strongest absolute static contribution; however, they simultaneously impact the estimation positively and negatively at different sites. This comprehensive analysis demonstrates that the possibility of these interpretable air pollutant ML models to be transfered across space and time.

Exploring the discrepancy between top-down and bottom-up approaches of fine spatio-temporal vehicular CO2 emission in an urban road network.

Road transport emissions of high spatial and temporal resolution are useful for greenhouse gas emission assessment in local action plans. However, estimating these high-resolution emissions is not straightforward, and different indirect approaches exist. The main aim of this study is to examine the differences in CO2 emissions obtained with different methods within a street canyon network in Helsinki, Finland, where a mobile laboratory campaign to quantify traffic emissions has been conducted. We compared three aerodynamic resistance based top-down methods (MOST1, MOST2 and BHT) and three activity based bottom-up microscopic emission models (NGM, HBEFAv4.2 and PHEMlight). The resulted CO2 fluxes using different methods could vary a few orders of magnitude. The combination of MOST1 and NGM model leads to the smallest discrepancy (sMAPE = 16.90 %) and the highest correlation coefficient (r = 0.78) among the rest. We evaluated the discrepancies in terms of different spatial (microenvrionments, local climate zones LCZs and grid sizes) and temporal features (seasons and periods of day). Measurements taken in LCZs of open high-rise regions and microenvironments of main road tend to have larger discrepancies between the two approaches. Using a coarser grid would lead to a relatively small discrepancy and high correlation in the wintertime, yet a loss in distinctive spatial variation. The discrepancies were also elevated on winter evenings. Among all explanatory variables, relative humidity shows the strongest relative importance for the discrepancy of the two approaches, followed by LCZs. Therefore, we stress the importance of choosing a suitable model for vehicular CO2 emission calculation based on meteorological conditions and LCZs. Such model comparison made on a local scale directly supports environmental organisations and cities' climate action plans where detailed information of CO2 emissions are needed.

Improving the current air quality index with new particulate indicators using a robust statistical approach.

To convey the severity of ambient air pollution level to the public, air quality index (AQI) is used as a communication tool to reflect the concentrations of individual pollutants on a common scale. However, due to the enhanced air pollution control in recent years, air quality has improved, and the roles of some air pollutant species included in the existing AQI as urban air pollutants have diminished. In this study, we suggest the current AQI should be revised in a way that new air pollution indicators would be considered so that it would better represent the health effects caused by local combustion processes from traffic and residential burning. Based on the air quality data of 2017-2019 in three different sites in Helsinki metropolitan area, we assumed the statistical distributions of the current indicators (NO2 and PM2.5) and the proposed particulate indicators (BC, LDSA and PNC) were related as they have similar sources in urban regions despite the varying correlations between the current and proposed indicators (NO2: r = 0.5-0.85, PM2.5: r = 0.28-0.72). By fitting the data to an optimal distribution function, together with expert opinions, we improved the current Finnish AQI and determined the AQI breakpoints for the proposed indicators where this robust statistical approach is transferrable to other cities. The addition of the three proposed indicators to the current AQI would decrease the number of good air quality hours in all three environments (largest decrease in urban traffic site, ~22 %). The deterioration of air quality class appeared more severe during peak hours in the urban traffic site due to vehicular emission and evenings in the detached housing site where domestic wood combustion often takes place. The introduction of the AQI breakpoints of the three new indicators serve as a first step of improving the current AQI before further air quality guideline levels are updated.

Short-Term and Long-Term COVID-19 Pandemic Forecasting Revisited with the Emergence of OMICRON Variant in Jordan.

Three simple approaches to forecast the COVID-19 epidemic in Jordan were previously proposed by Hussein, et al.: a short-term forecast (STF) based on a linear forecast model with a learning database on the reported cases in the previous 5-40 days, a long-term forecast (LTF) based on a mathematical formula that describes the COVID-19 pandemic situation, and a hybrid forecast (HF), which merges the STF and the LTF models. With the emergence of the OMICRON variant, the LTF failed to forecast the pandemic due to vital reasons related to the infection rate and the speed of the OMICRON variant, which is faster than the previous variants. However, the STF remained suitable for the sudden changes in epi curves because these simple models learn for the previous data of reported cases. In this study, we revisited these models by introducing a simple modification for the LTF and the HF model in order to better forecast the COVID-19 pandemic by considering the OMICRON variant. As another approach, we also tested a time-delay neural network (TDNN) to model the dataset. Interestingly, the new modification was to reuse the same function previously used in the LTF model after changing some parameters related to shift and time-lag. Surprisingly, the mathematical function type was still valid, suggesting this is the best one to be used for such pandemic situations of the same virus family. The TDNN was data-driven, and it was robust and successful in capturing the sudden change in +qPCR cases before and after of emergence of the OMICRON variant.

COVID-19 Pandemic Development in Jordan-Short-Term and Long-Term Forecasting.

In this study, we proposed three simple approaches to forecast COVID-19 reported cases in a Middle Eastern society (Jordan). The first approach was a short-term forecast (STF) model based on a linear forecast model using the previous days as a learning data-base for forecasting. The second approach was a long-term forecast (LTF) model based on a mathematical formula that best described the current pandemic situation in Jordan. Both approaches can be seen as complementary: the STF can cope with sudden daily changes in the pandemic whereas the LTF can be utilized to predict the upcoming waves' occurrence and strength. As such, the third approach was a hybrid forecast (HF) model merging both the STF and the LTF models. The HF was shown to be an efficient forecast model with excellent accuracy. It is evident that the decision to enforce the curfew at an early stage followed by the planned lockdown has been effective in eliminating a serious wave in April 2020. Vaccination has been effective in combating COVID-19 by reducing infection rates. Based on the forecasting results, there is some possibility that Jordan may face a third wave of the pandemic during the Summer of 2021.

Sensitivity Analysis for Predicting Sub-Micron Aerosol Concentrations Based on Meteorological Parameters.

Sub-micron aerosols are a vital air pollutant to be measured because they pose health effects. These particles are quantified as particle number concentration (PN). However, PN measurements are not always available in air quality measurement stations, leading to data scarcity. In order to compensate this, PN modeling needs to be developed. This paper presents a PN modeling framework using sensitivity analysis tested on a one year aerosol measurement campaign conducted in Amman, Jordan. The method prepares a set of different combinations of all measured meteorological parameters to be descriptors of PN concentration. In this case, we resort to artificial neural networks in the forms of a feed-forward neural network (FFNN) and a time-delay neural network (TDNN) as modeling tools, and then, we attempt to find the best descriptors using all these combinations as model inputs. The best modeling tools are FFNN for daily averaged data (with R 2 = 0.77 ) and TDNN for hourly averaged data (with R 2 = 0.66 ) where the best combinations of meteorological parameters are found to be temperature, relative humidity, pressure, and wind speed. As the models follow the patterns of diurnal cycles well, the results are considered to be satisfactory. When PN measurements are not directly available or there are massive missing PN concentration data, PN models can be used to estimate PN concentration using available measured meteorological parameters.

Input-Adaptive Proxy for Black Carbon as a Virtual Sensor.

Missing data has been a challenge in air quality measurement. In this study, we develop an input-adaptive proxy, which selects input variables of other air quality variables based on their correlation coefficients with the output variable. The proxy uses ordinary least squares regression model with robust optimization and limits the input variables to a maximum of three to avoid overfitting. The adaptive proxy learns from the data set and generates the best model evaluated by adjusted coefficient of determination (adjR2). In case of missing data in the input variables, the proposed adaptive proxy then uses the second-best model until all the missing data gaps are filled up. We estimated black carbon (BC) concentration by using the input-adaptive proxy in two sites in Helsinki, which respectively represent street canyon and urban background scenario, as a case study. Accumulation mode, traffic counts, nitrogen dioxide and lung deposited surface area are found as input variables in models with the top rank. In contrast to traditional proxy, which gives 20-80% of data, the input-adaptive proxy manages to give full continuous BC estimation. The newly developed adaptive proxy also gives generally accurate BC (street canyon: adjR2 = 0.86-0.94; urban background: adjR2 = 0.74-0.91) depending on different seasons and day of the week. Due to its flexibility and reliability, the adaptive proxy can be further extend to estimate other air quality parameters. It can also act as an air quality virtual sensor in support with on-site measurements in the future.