Search

Weighted positive matrix factorization (PMF) has been used by scientists to find small sets of underlying factors in environmental data. However, as the size of the data has grown, increasing computational costs have made it impractical to use traditional methods for this factorization. In this paper, we present a new weighting method to dramatically decrease computational costs for these traditional algorithms. We then apply this weighting method with the Randomized Hierarchical Alternating Least Squares (RHALS) algorithm to a large environmental dataset, where we show that interpretable factors can be reproduced using these methods. We show this algorithm results in a computational speedup of 38, 67, and 634 compared to the Multiplicative Update (MU), deterministic Hierarchical Alternating Least Squares (HALS), and non-negative Alternating Least Squares (ALS) algorithms, respectively. We also investigate rotational ambiguity in the solution, and present a simple “pulling” method to rotate a set of factors. This method is shown to find alternative solutions, and in some cases, lower the weighted residual error of the algorithm.

To improve air quality, knowledge of the sources and locations of air pollutant emissions is critical. However, for many global cities, no previous estimates exist of how much exposure to fine particulate matter (PM2.5), the largest environmental cause of mortality, is caused by emissions within the city vs. outside its boundaries. We use the Intervention Model for Air Pollution (InMAP) global-through-urban reduced complexity air quality model with a high-resolution, global inventory of pollutant emissions to quantify the contribution of emissions by source type and location for 96 global cities. Among these cities, we find that the fraction of PM2.5 exposure caused by within-city emissions varies widely (µ=51%; σ=23%) and is not well-explained by surrounding population density. The list of most-important sources also varies by city. Compared to a more mechanistically detailed model, InMAP predicts urban measured concentrations with less bias but more error. Predictive accuracy in urban areas is not particularly high with either model, suggesting an opportunity for improving global urban air emission inventories. We expect the results herein can be useful as a screening tool for policy options and in many cases may be robust enough to inform policy action to improve public health.

Here we present results from an evaluation of model simulations from the International Hemispheric Transport of Air Pollution Phase II (HTAPII) and Chemistry Climate Model Initiative (CCMI) model inter-comparison projects against a comprehensive series of ground based, aircraft and satellite observations of ozone mixing ratios made at various locations across India. The study focuses on the recent past (observations from 2008–2013, models from 2008–2010) as this is most pertinent to understanding the health impacts of ozone. To our understanding this is the most comprehensive evaluation of these models' simulations of ozone across the Indian sub-continent to date. This study highlights some significant successes and challenges that the models face in representing the oxidative chemistry of the region. The multi-model range in area weighted surface ozone over the Indian subcontinent is 37.26–56.11 ppb, whilst the population weighted range is 41.38–57.5 ppb. When compared against surface observations from the Modelling Atmospheric Pollution and Networking (MAPAN) network of eight semi-urban monitoring sites spread across India, we find that the models tend to simulate higher ozone than that which is observed. However, observations of NOx and CO tend to be much higher than modelled mixing-ratios, suggesting that the underlying emissions used in the models do not characterise these regions accurately and/or that the resolution of the models is not adequate to simulate the photo-chemical environment about these surface observations. Empirical Orthogonal Function (EOF) analysis is used in order to identify the extent to which the models agree with regards to the spatio-temporal distribution of the tropospheric ozone column, derived using OMI-MLS observations. We show that whilst the models agree with the spatial pattern of the first EOF of observed tropospheric ozone column, most of the models simulate a peak in the first EOF seasonal cycle represented by principle component 1, which is later than the observed peak . This suggest a widespread systematic bias in the timing of emissions or some other unknown seasonal process. In addition to evaluating modelled ozone mixing ratios, we explore modelled emissions of NOx, CO, VOCs, and the ozone response to the emissions. We find a high degree of variation in emissions from non-anthropogenic sources (e.g. lightning NOx and biomass burning CO) between models. Total emissions of NOx and CO over India vary more between different models in the same MIP than the same model used in different MIPs, making it impossible to diagnose whether differences in modelled ozone are due to emissions or model processes. We therefore recommend targeted experiments to pinpoint the exact causes of discrepancies between modelled and observed ozone and ozone precursors for this region. To this end, a higher density of long term monitoring sites measuring not only ozone but also ozone precursors including speciated VOCs, located in more rural regions of the Indian sub-continent, would enable improvements in assessing the biases in models run at the resolution found in HTAPII and CCMI.

This study is based on model results from TF HTAP (Task Force on Hemispheric Transport of Air Pollution) phase II where a set of source receptor model experiments have been defined, reducing global (and regional) anthropogenic emissions by 20 % in different source regions throughout the globe, with main focus on year 2010. All the participating models use the same set of emissions. Comparisons of model results to measurements are shown for selected European surface sites and for ozone sondes, but the main focus here is on the contributions to European ozone levels from different world regions, and how and why these contributions differ depending on model. We investigate the origins by use of a novel stepwise approach combining simple tracer calculations and calculations of CO and O3. To highlight differences, we analyse the vertical transects of the mid latitude effects from the 20 % emission reductions. Based on the relative emission changes from different world regions the models agree that for ozone the contributions from the rest of the world is larger than the effects from European emissions alone, with the largest contributions from North America and East Asia. The contribution will however depend on the choice of ozone metric. There are also considerable contributions from other nearby regions to the east and from international shipping, Whereas ozone from European sources peaks in the summer months, the largest contributions from non European sources are mostly calculated for the spring months when ozone production over the polluted continents starts to increase, while at the same time the lifetime of ozone in the free troposphere is relatively long. At the surface contributions from non European sources are of similar magnitude for all European sub regions considered, defined as TF HTAP receptor regions (north west, south west, east and south east Europe).

This study quantifies future changes in tropospheric ozone (O3) using a simple parameterisation of source–receptor relationships based on simulations from a range of models participating in the Task Force on Hemispheric Transport of Air Pollutants (TF-HTAP) experiments. Surface and tropospheric O3 changes are calculated globally and across 16 regions from perturbations in precursor emissions (NOx, CO, volatile organic compounds – VOCs) and methane (CH4) abundance only, neglecting any impact from climate change. A source attribution is provided for each source region along with an estimate of uncertainty based on the spread of the results from the models. Tests against model simulations using the Hadley Centre Global Environment Model version 2 – Earth system configuration (HadGEM2-ES) confirm that the approaches used within the parameterisation perform well for most regions. The O3 response to changes in CH4 abundance is slightly larger in the TF-HTAP Phase 2 than in the TF-HTAP Phase 1 assessment (2010) and provides further evidence that controlling CH4 is important for limiting future O3 concentrations. Different treatments of chemistry and meteorology in models remain one of the largest uncertainties in calculating the O3 response to perturbations in CH4 abundance and precursor emissions, particularly over the Middle East and south Asia regions. Emission changes for the future Evaluating the CLimate and Air Quality ImPacts of Short-livEd Pollutants (ECLIPSE) scenarios and a subset of preliminary Shared Socioeconomic Pathways (SSPs) indicate that surface O3 concentrations will increase regionally by 1 to 8ppbv in 2050. Source attribution analysis highlights the growing importance of CH4 in the future under current legislation. A change in the global tropospheric O3 radiative forcing of +0.07Wm−2 from 2010 to 2050 is predicted using the ECLIPSE scenarios and SSPs, based solely on changes in CH4 abundance and tropospheric O3 precursor emissions and neglecting any influence of climate change. Current legislation is shown to be inadequate in limiting the future degradation of surface ozone air quality and enhancement of near-term climate warming. More stringent future emission controls provide a large reduction in both surface O3 concentrations and O3 radiative forcing. The parameterisation provides a simple tool to highlight the different impacts and associated uncertainties of local and hemispheric emission control strategies on both surface air quality and the near-term climate forcing by tropospheric O3.

This study uses multi-model ensemble results of 11 models from the 2nd phase of Task Force Hemispheric Transport of Air Pollution (HTAP II) to calculate the global sulfur (S) and nitrogen (N) deposition in 2010. Modelled wet deposition is evaluated with observation networks in North America, Europe and Asia. The modelled results agree well with observations, with 76–83 % of stations having predicted within ±50 % of observations. The results underestimate SO42−, NO3− and NH4+ wet depositions in some European and East Asian stations, but overestimate NO3− wet deposition in Eastern United States. Inter-comparison with previous projects (PhotoComp, ACCMIP and HTAP I) shows HTPA II has considerably improved the estimation of deposition at European and East Asian stations. Modelled dry deposition is generally higher than the “inferential” data calculated by observed concentration and modelled velocity in North America, but the inferential data has high uncertainty, too. The global S deposition is 84 Tg(S) in 2010, with 49 % of the deposits on continental regions and 51 % on ocean (19 % on coastal). The global N deposition consists of 59 Tg(N) oxidized nitrogen (NOy) deposition and 64 Tg(N) reduced nitrogen (NHx) deposition in 2010. 65 % of N is deposited on the continental regions and 35 % is on ocean (15 % on coastal). The estimated outflow of pollution from land to ocean is about 4 Tg(S) for S deposition and 18 Tg(N) for N deposition. Compared our results to the results in 2001 from HTAP I, we find that the global distributions of S and N depositions have changed considerably during the last 10 years. The global S deposition decreases 2 Tg(S) (3 %) from 2001 to 2010, with significant decreases in Europe (5 Tg(S) and 55 %), North America (3 Tg(S) and 29 %) and Russia (2 Tg(S) and 26 %), and increases in South Asia (2 Tg(S) and 42 %) and the Middle East (1 Tg(S) and 44% ). The global N deposition increases by 7 Tg(N) (6 %), mainly contributed by South Asia (5 Tg(N) and 39 %), East Asia (4 Tg(N) and 21 %) and Southeast Asia (2 Tg(N) and 21 %). The NHx deposition is increased with no control policy on NH3 emission in North America. On the other hand, NOy deposition starts to dominate in East Asia (especially China) due to boosted NOx emission in recent years.

In the Hemispheric Transport of Air Pollution Phase 2 (HTAP) exercise, a range of global atmospheric general circulation and chemical transport models performed coordinated perturbation experiments with 20 % reductions in emissions of anthropogenic aerosols, or aerosol precursors, in a number of source regions. Here, we compare the resulting changes in the atmospheric load and vertically resolved profiles of black carbon (BC), organic aerosols (OA) and sulfate (SO4) from 10 models that include treatment of aerosols. We use a set of temporally, horizontally and vertically resolved profiles of aerosol forcing efficiency (AFE) to estimate the impact of emission changes in six major source regions on global radiative forcing (RF) pertaining to the direct aerosol effect. Results show that mitigations in South and East Asia have substantial impacts on the radiative budget in all investigated receptor regions, especially for BC. In Russia and the Middle East, more than 80 % of the forcing for BC and OA is due to extra-regional emission reductions. Similarly, for North America, BC emissions control in East Asia is found to be more important than domestic mitigations, which is consistent with previous findings. Comparing fully resolved RF calculations to RF estimates based on vertically averaged AFE profiles allows us to quantify the importance of vertical resolution to RF estimates. We find that locally in the source regions, a 20 % emission reduction strengthens the radiative forcing associated with SO4 by 25 % when including the vertical dimension, as the AFE for SO4 is strongest near the surface. Conversely, the local RF from BC weakens by 37 % since BC AFE is low close to the ground. The influence of inter-continental transport on BC forcing, however, is enhanced by one third when accounting for the vertical aspect, because long-range transport leads primarily to aerosol changes at high altitudes, where the BC AFE is strong.