Search

Gradient Boosting Decision Trees (GBDTs) have become very successful in recent years, with many awards in machine learning and data mining competitions. There have been several recent studies on how to train GBDTs in the federated learning setting. In this paper, we focus on horizontal federated learning, where data samples with the same features are distributed among multiple parties. However, existing studies are not efficient or effective enough for practical use. They suffer either from the inefficiency due to the usage of costly data transformations such as secret sharing and homomorphic encryption, or from the low model accuracy due to differential privacy designs. In this paper, we study a practical federated environment with relaxed privacy constraints. In this environment, a dishonest party might obtain some information about the other parties' data, but it is still impossible for the dishonest party to derive the actual raw data of other parties. Specifically, each party boosts a number of trees by exploiting similarity information based on locality-sensitive hashing. We prove that our framework is secure without exposing the original record to other parties, while the computation overhead in the training process is kept low. Our experimental studies show that, compared with normal training with the local data of each party, our approach can significantly improve the predictive accuracy, and achieve comparable accuracy to the original GBDT with the data from all parties., Comment: Accepted to AAAI-20

Federated learning enables multiple parties to collaboratively learn a model without exchanging their data. While most existing federated learning algorithms need many rounds to converge, one-shot federated learning (i.e., federated learning with a single communication round) is a promising approach to make federated learning applicable in cross-silo setting in practice. However, existing one-shot algorithms only support specific models and do not provide any privacy guarantees, which significantly limit the applications in practice. In this paper, we propose a practical one-shot federated learning algorithm named FedKT. By utilizing the knowledge transfer technique, FedKT can be applied to any classification models and can flexibly achieve differential privacy guarantees. Our experiments on various tasks show that FedKT can significantly outperform the other state-of-the-art federated learning algorithms with a single communication round.

In this paper, we present a novel parallel implementation for training Gradient Boosting Decision Trees (GBDTs) on Graphics Processing Units (GPUs). Thanks to the excellent results on classification/regression and the open sourced libraries such as XGBoost, GBDTs have become very popular in recent years and won many awards in machine learning and data mining competitions. Although GPUs have demonstrated their success in accelerating many machine learning applications, it is challenging to develop an efficient GPU-based GBDT algorithm. The key challenges include irregular memory accesses, many sorting operations with small inputs and varying data parallel granularities in tree construction. To tackle these challenges on GPUs, we propose various novel techniques including (i) Run-length Encoding compression and thread/block workload dynamic allocation, (ii) data partitioning based on stable sort, and fast and memory efficient attribute ID lookup in node splitting, (iii) finding approximate split points using two-stage histogram building, (iv) building histograms with the aware of sparsity and exploiting histogram subtraction to reduce histogram building workload, (v) reusing intermediate training results for efficient gradient computation, and (vi) exploiting multiple GPUs to handle larger data sets efficiently. Our experimental results show that our algorithm named ThunderGBM can be 10x times faster than the state-of-the-art libraries (i.e., XGBoost, LightGBM and CatBoost) running on a relatively high-end workstation of 20 CPU cores. In comparison with the libraries on GPUs, ThunderGBM can handle higher dimensional problems which the libraries become extremely slow or simply fail. For the data sets the existing libraries on GPUs can handle, ThunderGBM achieves up to 10 times speedup on the same hardware, which demonstrates the significance of our GPU optimizations. Moreover, the models trained by ThunderGBM are identical to those trained by XGBoost, and have similar quality as those trained by LightGBM and CatBoost.

Citrus wastes contain large amounts of pectin as the main constituent in addition to cellulose, lignin and hemicellulose. The thermal degradation behaviors and products distribution from the fast pyrolysis of citrus wastes are different from lignocellulosic biomass due to the high content of pectin. In view of this, the detailed decomposition behaviors of pectin are investigated by using density functional theory (DFT). The calculation results indicate that the most energetically favored initial reaction of pectin is the cleavage of α-1,4-glycoside bond (C1 O bond) with an energy barrier of 126.4 kJ/mol. In addition, compared with the reaction energy barriers, it can be seen that the formation of CO, CH4 and furan are more difficult than that of CO2, methanol and furfural, respectively. The comprehensive kinetic analysis indicates that Path-A2-2 should be the most favorable pathway for the formation of acetaldehyde. The rate-determining steps of the formation of furan and furfural are dehydration reactions while decarbonylation reaction is the result of the rate-determining step of the formation of methyl 2-furoate and propanone.

Federated learning enables multiple parties to collaboratively train a machine learning model without communicating their local data. A key challenge in federated learning is to handle the heterogeneity of local data distribution across parties. Although many studies have been proposed to address this challenge, we find that they fail to achieve high performance in image datasets with deep learning models. In this paper, we propose MOON: model-contrastive federated learning. MOON is a simple and effective federated learning framework. The key idea of MOON is to utilize the similarity between model representations to correct the local training of individual parties, i.e., conducting contrastive learning in model-level. Our extensive experiments show that MOON significantly outperforms the other state-of-the-art federated learning algorithms on various image classification tasks., Comment: Accepted by CVPR 2021

To investigate the impact of air pollutant control policies on future PM2.5 concentrations and their source contributions in China, we developed four future scenarios for 2030 based on a 2013 emission inventory, and conducted air quality simulations for each scenario using the chemical transport model GEOS-Chem (version 9.1.3). Two energy scenarios i.e., current legislation (CLE) and with additional measures (WAM), were developed to project future energy consumption, reflecting, respectively, existing legislation and implementation status as of the end of 2012, and new energy-saving policies that would be released and enforced more stringently. Two end-of-pipe control strategies, i.e., current control technologies (until 2017) and more stringent control technologies (until 2030), were also developed. The combinations of energy scenarios and end-of-pipe control strategies constitute four emission scenarios (2017-CLE, 2030-CLE, 2017-WAM, and 2030-WAM) evaluated in simulations. PM2.5 concentrations at national level were estimated to be 57 μg/m3 in the base year 2013, and 58 μg/m3, 42 μg/m3, 42 μg/m3, and 30 μg/m3 under the 2017-CLE, 2030-CLE, 2017-WAM, and 2030-WAM scenarios in 2030, respectively. Large PM2.5 reductions between 2013 and 2030 were estimated for heavily polluted regions (Sichuan Basin, Middle Yangtze River, North China). The energy-saving policies show similar effects to the end-of-pipe emission control measures, but the relative importance of these two groups of policies varies in different regions. Absolute contributions to PM2.5 concentrations from most major sources declined from 2017-CLE to 2030-WAM. With respect to fractional contributions, most coal-burning sectors (including power plant, industrial and residential coal burning) increased from 2017-CLE to 2030-WAM, due to larger reductions from non-coal sources, including transportation and biomass open burning. Residential combustion and open burning had much lower fractional contribution to ambient PM2.5 concentrations in the 2017-WAM/2030-WAM compared to the 2017-CLE/2030-CLE scenarios. Fractional contributions from transportation were reduced dramatically in 2030-CLE and 2030-WAM compared to 2017-CLE/2017-WAM, due to the enforcement of stringent end-of-pipe emission controls. Across all scenarios, coal combustion remained the single largest contributor to PM2.5 concentrations in 2030. Reducing PM2.5 emissions from coal combustion remains a strategic priority for air quality management in China.

As societal concerns on data privacy recently increase, we have witnessed data silos among multiple parties in various applications. Federated learning emerges as a new learning paradigm that enables multiple parties to collaboratively train a machine learning model without sharing their raw data. Vertical federated learning, where each party owns different features of the same set of samples and only a single party has the label, is an important and challenging topic in federated learning. Communication costs among different parties have been a major hurdle for practical vertical learning systems. In this paper, we propose a novel communication-efficient vertical federated learning algorithm named FedOnce, which requires only one-shot communication among parties. To improve model accuracy and provide privacy guarantee, FedOnce features unsupervised learning representations in the federated setting and privacy-preserving techniques based on moments accountant. The comprehensive experiments on 10 datasets demonstrate that FedOnce achieves close performance compared to state-of-the-art vertical federated learning algorithms with much lower communication costs. Meanwhile, our privacy-preserving technique significantly outperforms the state-of-the-art approaches under the same privacy budget.

The interactions between aerosols and ice clouds represent one of the largest uncertainties in global radiative forcing from pre-industrial time to the present. In particular, the impact of aerosols on ice crystal effective radius (Rei), which is a key parameter determining ice clouds' net radiative effect, is highly uncertain due to limited and conflicting observational evidence. Here we investigate the effects of aerosols on Rei under different meteorological conditions using 9-year satellite observations. We find that the responses of Rei to aerosol loadings are modulated by water vapor amount in conjunction with several other meteorological parameters. While there is a significant negative correlation between Rei and aerosol loading in moist conditions, consistent with the "Twomey effect" for liquid clouds, a strong positive correlation between the two occurs in dry conditions. Simulations based on a cloud parcel model suggest that water vapor modulates the relative importance of different ice nucleation modes, leading to the opposite aerosol impacts between moist and dry conditions. When ice clouds are decomposed into those generated from deep convection and formed in situ, the water vapor modulation remains in effect for both ice cloud types, although the sensitivities of Rei to aerosols differ noticeably between them due to distinct formation mechanisms. The water vapor modulation can largely explain the difference in the responses of Rei to aerosol loadings in various seasons. A proper representation of the water vapor modulation is essential for an accurate estimate of aerosol–cloud radiative forcing produced by ice clouds.

As China's ecological environmental problem becomes severe, corporate environmental behavior (CEB) has become the focus of a range of stakeholders, policy makers and the whole society since the operating activities of companies is the main source of environmental pollution. This study used a questionnaire survey and structural equation model (SEM) to examine the relative importance of external and internal pressures (EIP) on driving CEB. Data was obtained from 702 papermaking companies in China. The mechanism of EIP effects on CEB was tested employing SEM through path analysis, using AMOS 21.0. The results showed that government pressure, economic pressure and internal pressure played significant positive roles in environmental behavior of papermaking companies in China. Economic pressure was the most important factor on environmental behaviors (including defensive, accommodative and proactive behaviors) of papermaking companies. Furthermore, the role of internal pressure on CEB (defensive and proactive) was second only to economic pressure, while the role of government pressure on CEB was much weaker than our expectation. However, the role of social pressure, which was imposed by general public and environmental Non-Governmental Organizations (NGOs), was not significant. Thus, policy makers should emphasize on the interactions of economic mechanism, government regulation and internal incentive mechanism to stimulate firms to adopt active environmental behavior. More concerns from the general public shall be addressed to enhance their driving effect on CEB in Chinese papermaking industry.

A set of parameterizations is developed for spectral single-scattering properties of clean and black carbon (BC)-contaminated snow based on geometric-optics surface wave (GOS) computations, which explicitly resolves BC–snow internal mixing and various snow grain shapes. GOS calculations show that, compared with nonspherical grains, volume-equivalent snow spheres show up to 20% larger asymmetry factors and hence stronger forward scattering, particularly at wavelengths 99%. Based on GOS calculations, BC–snow internal mixing enhances the snow single-scattering coalbedo at wavelengths 99%. Overall, in addition to snow grain size, both BC–snow internal mixing and snow grain shape play critical roles in quantifying BC effects on snow optical properties. The present parameterizations can be conveniently applied to snow, land surface, and climate models including snowpack radiative transfer processes.

Aerosol-cloud interactions (aerosol indirect effects) play an important role in regional meteorological variations, which could further induce feedback on regional air quality. While the impact of aerosol-cloud interactions on meteorology and climate has been extensively studied, their feedback on air quality remains unclear. Using a fully coupled meteorology-chemistry model, we find that increased aerosol loading due to anthropogenic activities in China substantially increases column cloud droplet number concentration and liquid water path (LWP), which further leads to a reduction in the downward shortwave radiation at surface, surface air temperature and planetary boundary layer (PBL) height. The shallower PBL and accelerated cloud chemistry due to larger LWP in turn enhance the concentrations of particulate matter with diameter less than 2.5 μm (PM2.5) by up to 33.2 μg m−3 (25.1%) and 11.0 μg m−3 (12.5%) in January and July, respectively. Such a positive feedback amplifies the changes in PM2.5 concentrations, indicating an additional air quality benefit under effective pollution control policies but a penalty for a region with a deterioration in PM2.5 pollution. Additionally, we show that the cloud processing of aerosols, including wet scavenging and cloud chemistry, could also have substantial effects on PM2.5 concentrations.

This paper presents and characterizes an Open Application Repository for Federated Learning (OARF), a benchmark suite for federated machine learning systems. Previously available benchmarks for federated learning have focused mainly on synthetic datasets and use a limited number of applications. OARF mimics more realistic application scenarios with publicly available data sets as different data silos in image, text and structured data. Our characterization shows that the benchmark suite is diverse in data size, distribution, feature distribution and learning task complexity. The extensive evaluations with reference implementations show the future research opportunities for important aspects of federated learning systems. We have developed reference implementations, and evaluated the important aspects of federated learning, including model accuracy, communication cost, throughput and convergence time. Through these evaluations, we discovered some interesting findings such as federated learning can effectively increase end-to-end throughput., Comment: ACM Transactions on Intelligent Systems and Technology, Vol. 13, No. 4, Article 63

Support Vector Machines (SVMs) can solve structured multi-output learning problems such as multi-label classification, multiclass classification and vector regression. SVM training is expensive especially for large and high dimensional datasets. The bottleneck of the SVM training often lies in the kernel value computation. In many real-world problems, the same kernel values are used in many iterations during the training, which makes the caching of kernel values potentially useful. The majority of the existing studies simply adopt the LRU (least recently used) replacement strategy for caching kernel values. However, as we analyze in this paper, the LRU strategy generally achieves high hit ratio near the final stage of the training, but does not work well in the whole training process. Therefore, we propose a new caching strategy called EFU (less frequently used) which replaces the less frequently used kernel values that enhances LFU (least frequently used). Our experimental results show that EFU often has 20\% higher hit ratio than LRU in the training with the Gaussian kernel. To further optimize the strategy, we propose a caching strategy called HCST (hybrid caching for the SVM training), which has a novel mechanism to automatically adapt the better caching strategy in the different stages of the training. We have integrated the caching strategy into ThunderSVM, a recent SVM library on many-core processors. Our experiments show that HCST adaptively achieves high hit ratios with little runtime overhead among different problems including multi-label classification, multiclass classification and regression problems. Compared with other existing caching strategies, HCST achieves 20\% more reduction in training time on average., Accepted by IEEE Transactions on Neural Networks and Learning Systems (TNNLS)

Federated learning has been a hot research topic in enabling the collaborative training of machine learning models among different organizations under the privacy restrictions. As researchers try to support more machine learning models with different privacy-preserving approaches, there is a requirement in developing systems and infrastructures to ease the development of various federated learning algorithms. Similar to deep learning systems such as PyTorch and TensorFlow that boost the development of deep learning, federated learning systems (FLSs) are equivalently important, and face challenges from various aspects such as effectiveness, efficiency, and privacy. In this survey, we conduct a comprehensive review on federated learning systems. To achieve smooth flow and guide future research, we introduce the definition of federated learning systems and analyze the system components. Moreover, we provide a thorough categorization for federated learning systems according to six different aspects, including data distribution, machine learning model, privacy mechanism, communication architecture, scale of federation and motivation of federation. The categorization can help the design of federated learning systems as shown in our case studies. By systematically summarizing the existing federated learning systems, we present the design factors, case studies, and future research opportunities., Accepted to IEEE Transactions on Knowledge and Data Engineering (TKDE)

O-acetyl-4-O-methylglucurono-xylan is selected as a model compound because of its abundant O-acetyl and 4-O-methylglucuronic acid groups as side chains of hemicellulose. The detailed decomposition pathways of O-acetyl-4-O-methylglucurono-xylan are investigated by using density functional theory (DFT) and transition state theory. In addition, the possible pyrolysis pathways of 4-O-methylglucuronic acid, based on the Mayer bond order values, are predicted. The results indicate that the most energetically favored initial reaction of O-acetyl-4-O-methylglucurono-xylan is the cleavage of the 4-O-methylglucuronic acid unit. Furfural can be obtained through the ring-opening of 4-O-methylglucuronic acid in three different pathways. The O-methyl group is predominantly responsible for the generation of CH3OH. In addition, the formation pathways of a special furan-derived product 2-hydroxymethylene-tetrahydrofuran-3-one are first validated by DFT calculation. The rate-determining steps to form 2-hydroxymethylene-tetrahydrofuran-3-one are the cyclization reaction and enol–keto tautomerization. Anhydroxylopyranose and dianhydroxylopyranose can be produced through intramolecular dehydration and glycosidic bond cleavage reactions.

We investigate the sensitivity of black carbon (BC) in the Arctic, including BC concentration in snow (BCsnow, ng g−1) and surface air (BCair, ng m−3), as well as emissions, dry deposition, and wet scavenging using the global three-dimensional (3-D) chemical transport model (CTM) GEOS-Chem. We find that the model underestimates BCsnow in the Arctic by 40 % on average (median = 11.8 ng g−1). Natural gas flaring substantially increases total BC emissions in the Arctic (by ∼ 70 %). The flaring emissions lead to up to 49 % increases (0.1–8.5 ng g−1) in Arctic BCsnow, dramatically improving model comparison with observations (50 % reduction in discrepancy) near flaring source regions (the western side of the extreme north of Russia). Ample observations suggest that BC dry deposition velocities over snow and ice in current CTMs (0.03 cm s−1 in the GEOS-Chem) are too small. We apply the resistance-in-series method to compute a dry deposition velocity (vd) that varies with local meteorological and surface conditions. The resulting velocity is significantly larger and varies by a factor of 8 in the Arctic (0.03–0.24 cm s−1), which increases the fraction of dry to total BC deposition (16 to 25 %) yet leaves the total BC deposition and BCsnow in the Arctic unchanged. This is largely explained by the offsetting higher dry and lower wet deposition fluxes. Additionally, we account for the effect of the Wegener–Bergeron–Findeisen (WBF) process in mixed-phase clouds, which releases BC particles from condensed phases (water drops and ice crystals) back to the interstitial air and thereby substantially reduces the scavenging efficiency of clouds for BC (by 43–76 % in the Arctic). The resulting BCsnow is up to 80 % higher, BC loading is considerably larger (from 0.25 to 0.43 mg m−2), and BC lifetime is markedly prolonged (from 9 to 16 days) in the Arctic. Overall, flaring emissions increase BCair in the Arctic (by ∼ 20 ng m−3), the updated vd more than halves BCair (by ∼ 20 ng m−3), and the WBF effect increases BCair by 25–70 % during winter and early spring. The resulting model simulation of BCsnow is substantially improved (within 10 % of the observations) and the discrepancies of BCair are much smaller during the snow season at Barrow, Alert, and Summit (from −67–−47 % to −46–3 %). Our results point toward an urgent need for better characterization of flaring emissions of BC (e.g., the emission factors, temporal, and spatial distribution), extensive measurements of both the dry deposition of BC over snow and ice, and the scavenging efficiency of BC in mixed-phase clouds. In addition, we find that the poorly constrained precipitation in the Arctic may introduce large uncertainties in estimating BCsnow. Doubling precipitation introduces a positive bias approximately as large as the overall effects of flaring emissions and the WBF effect; halving precipitation produces a similarly large negative bias.

We perform a comprehensive intercomparison of the geometric-optics surface-wave (GOS) approach, the superposition T-matrix method, and laboratory measurements for optical properties of fresh and coated/aged black carbon (BC) particles with complex structures. GOS and T-matrix calculations capture the measured optical (i.e., extinction, absorption, and scattering) cross sections of fresh BC aggregates, with 5–20% differences depending on particle size. We find that the T-matrix results tend to be lower than the measurements, due to uncertainty in theoretical approximations of realistic BC structures, particle property measurements, and numerical computations in the method. On the contrary, the GOS results are higher than the measurements (hence the T-matrix results) for BC radii 100 nm. We find good agreement (differences 100 nm. We find small deviations (≤10%) in asymmetry factors computed from the two methods for most BC coating structures and sizes, but several complex structures have 10–30% differences. This study provides the foundation for downstream application of the GOS approach in radiative transfer and climate studies.

We investigate the snow albedo variation in spring over the southern Tibetan Plateau induced by the deposition of light-absorbing aerosols using remote sensing data from moderate resolution imaging spectroradiometer (MODIS) aboard Terra satellite during 2001–2012. We have selected pixels with 100 % snow cover for the entire period in March and April to avoid albedo contamination by other types of land surfaces. A model simulation using GEOS-Chem shows that aerosol optical depth (AOD) is a good indicator for black carbon and dust deposition on snow over the southern Tibetan Plateau. The monthly means of satellite-retrieved land surface temperature (LST) and AOD over 100 % snow-covered pixels during the 12 years are used in multiple linear regression analysis to derive the empirical relationship between snow albedo and these variables. Along with the LST effect, AOD is shown to be an important factor contributing to snow albedo reduction. We illustrate through statistical analysis that a 1-K increase in LST and a 0.1 increase in AOD indicate decreases in snow albedo by 0.75 and 2.1 % in the southern Tibetan Plateau, corresponding to local shortwave radiative forcing of 1.5 and 4.2 W m−2, respectively.

The Gradient Boosting Decision Tree (GBDT) is a popular machine learning model for various tasks in recent years. In this paper, we study how to improve model accuracy of GBDT while preserving the strong guarantee of differential privacy. Sensitivity and privacy budget are two key design aspects for the effectiveness of differential private models. Existing solutions for GBDT with differential privacy suffer from the significant accuracy loss due to too loose sensitivity bounds and ineffective privacy budget allocations (especially across different trees in the GBDT model). Loose sensitivity bounds lead to more noise to obtain a fixed privacy level. Ineffective privacy budget allocations worsen the accuracy loss especially when the number of trees is large. Therefore, we propose a new GBDT training algorithm that achieves tighter sensitivity bounds and more effective noise allocations. Specifically, by investigating the property of gradient and the contribution of each tree in GBDTs, we propose to adaptively control the gradients of training data for each iteration and leaf node clipping in order to tighten the sensitivity bounds. Furthermore, we design a novel boosting framework to allocate the privacy budget between trees so that the accuracy loss can be further reduced. Our experiments show that our approach can achieve much better model accuracy than other baselines.

To investigate the impact of air pollutant control policies on future PM

A theoretical black carbon (BC) aging model is developed to account for three typical evolution stages, namely, freshly emitted aggregates, coated BC by soluble material, and BC particles undergoing further hygroscopic growth. The geometric-optics surface-wave (GOS) approach is employed to compute the BC single-scattering properties at each aging stage, which are subsequently compared with laboratory measurements. Theoretical calculations are consistent with measurements in extinction and absorption cross sections for fresh BC aggregates, but overestimate the scattering cross sections for BC mobility diameters of 155, 245, and 320 nm, because of uncertainties associated with theoretical calculations for small particles as well as laboratory scattering measurements. The measured optical cross sections for coated BC by sulfuric acid and for those undergoing further hygroscopic growth are captured by theoretical calculations using a concentric core-shell structure, with differences of less than 20 %. This suggests that the core-shell shape represents the realistic BC coating morphology reasonably well in this case, which is consistent with the observed strong structure compaction during aging. We find that the absorption and scattering properties of fresh BC aggregates vary by up to 60 % due to uncertainty in the BC refractive index, which, however, is a factor of two smaller in the case of coated BC particles. Sensitivity analyses on the BC morphology show that the optical properties of fresh BC aggregates are more sensitive to fractal dimension than primary spherule size. The absorption and scattering cross sections of coated BC particles vary by more than a factor of two due to different coating structures. We find an increase of 20–250 % in absorption and a factor of 3–15 in scattering during aging, significantly depending on coating morphology and aging stages. Applying the aging model to CalNex 2010 field measurements, we show that the resulting BC direct radiative forcing (DRF) first increases from 1.5 to 1.7 W m-2 and subsequently decreases to 1.0 W m-2 during the transport from the Los Angeles Basin to downwind regions, as a result of the competition between absorption enhancement due to coating and dilution of BC concentration. The BC DRF can vary by up to a factor of two due to differences in BC coating morphology. Thus, an accurate estimate of BC DRF requires the incorporation of a dynamic BC aging process that accounts for realistic morphology in climate models, particularly for the regional analysis with high atmospheric heterogeneity.

We estimate black carbon (BC) emissions in the western United States for July–September 2006 by inverting surface BC concentrations from the Interagency Monitoring of Protected Visual Environments (IMPROVE) network using a global chemical transport model (GEOS-Chem) and its adjoint. Our best estimate of the BC emissions is 49.9 Gg at 2° × 2.5° (a factor of 2.1 increase) and 47.3 Gg at 0.5° × 0.667° (1.9 times increase). Model results now capture the observed major fire episodes with substantial bias reductions (~ 35 % at 2° × 2.5° and ~ 15 % at 0.5° × 0.667°). The emissions are ~ 20–50 % larger than those from our earlier analytical inversions (Mao et al., 2014). The discrepancy is especially drastic in the partitioning of anthropogenic versus biomass burning emissions. The August biomass burning BC emissions are 4.6–6.5 Gg and anthropogenic BC emissions 8.6–12.8 Gg, varying with the model resolution, error specifications, and subsets of observations used. On average both anthropogenic and biomass burning emissions in the adjoint inversions increase 2-fold relative to the respective {a priori} emissions, in distinct contrast to the halving of the anthropogenic and tripling of the biomass burning emissions in the analytical inversions. We attribute these discrepancies to the inability of the adjoint inversion system, with limited spatiotemporal coverage of the IMPROVE observations, to effectively distinguish collocated anthropogenic and biomass burning emissions on model grid scales. This calls for concurrent measurements of other tracers of biomass burning and fossil fuel combustion (e.g., carbon monoxide and carbon isotopes). We find that the adjoint inversion system as is has sufficient information content to constrain the total emissions of BC on the model grid scales.

Background error covariance (BEC) is crucial in data assimilation. This paper addresses the multivariate BEC associated with black carbon, organic carbon, nitrates, sulfates, and other con- stituents of aerosol species. These aerosol species are modeled and predicted using the Model for Simulating Aerosol Interactions and Chemistry scheme (MOSAIC) in the Weather Research and Forecasting/Chemistry (WRF/Chem) model at a resolution of 4 km in Southern California. The BEC is estimated from the differences between the 36-hour and 12-hour forecasts using the NMC method. The results indicated that the maximum background error standard deviation is associated with nitrate and is larger than that of black carbon, organic carbon, and sulfate. The horizontal and vertical scale of the correlation of nitrate is much smaller than that of other species. A significant cross-correlation is found between the species of black carbon and organic carbon. The cross-correlations between nitrate and other variables are relatively smaller and exhibit a relatively smaller length scale. Single observation data assimilation experiments are performed to illustrate the effect of the BEC on analysis increments.

The accidental release of radioactive materials from nuclear power plant leads to radioactive pollution. We apply an augmented ensemble Kalman filter (EnKF) with a chemical transport model to jointly estimate the emissions of Perfluoromethylcyclohexane (PMCH), a tracer substitute for radionuclides, from a point source during the European Tracer Experiment, and to improve the forecast of its dispersion downwind. We perturb wind fields to account for meteorological uncertainties. We expand the state vector of PMCH concentrations through continuously adding an a priori emission rate for each succeeding assimilation cycle. We adopt a time-correlated red noise to simulate the temporal emission fluctuation. The improved EnKF system rapidly updates (and reduces) the excessively large initial first-guess emissions, thereby significantly improves subsequent forecasts (r = 0.83, p0.001). It retrieves 94% of the total PMCH released and substantially reduces transport error (80% average reduction of the normalized mean square error).

We decompose factors affecting China’s energy-related air pollutant (NOx, PM2.5, and SO2) emission changes into different effects using structural decomposition analysis (SDA). We find that, from 2005 to 2012, investment increased NOx, PM2.5, and SO2 emissions by 14.04, 7.82 and 15.59 Mt respectively, and consumption increased these emissions by 11.09, 7.98, and 12.09 Mt respectively. Export and import slightly increased the emissions on the whole, but the rate of the increase has slowed down, possibly reflecting the shift in China’s foreign trade structure. Energy intensity largely reduced NOx, PM2.5, and SO2 emissions by 12.49, 14.33 and 23.06 Mt respectively, followed by emission efficiency that reduces these emissions by 4.57, 9.08, and 17.25 Mt respectively. Input-output efficiency slightly reduces the emissions. At sectoral and sub-sectoral levels, consumption is a great driving factor in agriculture and commerce, whereas investment is a great driving factor in transport, construction, and some industrial subsectors such as iron and steel, nonferrous metals, building materials, coking, and power and heating supply. Energy intensity increases emissions in transport, chemical products and manufacturing, but decreases emissions in all other sectors and subsectors. Some policies arising from our study results are discussed.

Aerosol-cloud interactions (aerosol indirect effects) play an important role in regional meteorological variations, which could further induce feedback on regional air quality. While the impact of aerosol-cloud interactions on meteorology and climate has been extensively studied, their feedback on air quality remains unclear. Using a fully coupled meteorology-chemistry model, we find that increased aerosol loading due to anthropogenic activities in China substantially increases column cloud droplet number concentration and liquid water path (LWP), which further leads to a reduction in the downward shortwave radiation at surface, surface air temperature and planetary boundary layer (PBL) height. The shallower PBL and accelerated cloud chemistry due to larger LWP in turn enhance the concentrations of particulate matter with diameter less than 2.5 μm (PM2.5) by up to 33.2 μg m−3 (25.1%) and 11.0 μg m−3 (12.5%) in January and July, respectively. Such a positive feedback amplifies the changes in PM2.5 concentrations, indicating an additional air quality benefit under effective pollution control policies but a penalty for a region with a deterioration in PM2.5 pollution. Additionally, we show that the cloud processing of aerosols, including wet scavenging and cloud chemistry, could also have substantial effects on PM2.5 concentrations.

We quantify source contributions to springtime (April 2008) surface black carbon (BC) in the Arctic by interpreting surface observations of BC at five receptor sites (Denali, Barrow, Alert, Zeppelin, and Summit) using a global chemical transport model (GEOS-Chem) and its adjoint. Contributions to BC at Barrow, Alert, and Zeppelin are dominated by Asian anthropogenic sources (40–43 %) before April 18 and by Siberian open biomass burning emissions (29–41 %) afterward. In contrast, Summit, a mostly free tropospheric site, has predominantly an Asian anthropogenic source contribution (24–68 %, with an average of 45 %). We compute the adjoint sensitivity of BC concentrations at the five sites during a pollution episode (April 20–25) to global emissions from March 1 to April 25. The associated contributions are the combined results of these sensitivities and BC emissions. Local and regional anthropogenic sources in Alaska are the largest anthropogenic sources of BC at Denali (63 %), and natural gas flaring emissions in the Western Extreme North of Russia (WENR) are the largest anthropogenic sources of BC at Zeppelin (26 %) and Alert (13 %). We find that long-range transport of emissions from Beijing-Tianjin-Hebei (also known as Jing-Jin-Ji), the biggest urbanized region in Northern China, contribute significantly (~ 10 %) to surface BC across the Arctic. On average it takes ~ 12 days for Asian anthropogenic emissions and Siberian biomass burning emissions to reach Arctic lower troposphere, supporting earlier studies. Natural gas flaring emissions from the WENR reach Zeppelin in about a week. We find that episodic, direct transport events dominate BC at Denali (87 %), a site outside the Arctic front, a strong transport barrier. The relative contribution of direct transport to surface BC within the Arctic front is much smaller (~ 50 % at Barrow and Zeppelin and ~ 10 % at Alert). The large contributions from Asian anthropogenic sources are predominately in the form of ‘chronic’ pollution (~ 40 % at Barrow and 65 % at Alert and 57 % at Zeppelin) on 1–2 month timescales. As such, it is likely that previous studies using 5- or 10-day trajectory analyses strongly underestimated the contribution from Asia to surface BC in the Arctic. Both finer temporal resolution of biomass burning emissions and accounting for the Wegener-Bergeron-Findeisen (WBF) process in wet scavenging improve the source attribution estimates.

We estimate the snow albedo forcing and direct radiative forcing (DRF) of black carbon (BC) in the Tibetan Plateau using a global chemical transport model in conjunction with a stochastic snow model and a radiative transfer model. The annual mean BC snow albedo forcing is 2.9 W m−2 averaged over snow-covered plateau regions, which is a factor of 3 larger than the value over global land snowpack. BC-snow internal mixing increases the albedo forcing by 40–60% compared with external mixing, and coated BC increases the forcing by 30–50% compared with uncoated BC aggregates, whereas Koch snowflakes reduce the forcing by 20–40% relative to spherical snow grains. The annual BC DRF at the top of the atmosphere is 2.3 W m−2 with uncertainties of −70–85% in the plateau after scaling the modeled BC absorption optical depth to Aerosol Robotic Network observations. The BC forcings are attributed to emissions from different regions.

We estimate biomass burning and anthropogenic emissions of black carbon (BC) in the western US for May–October 2006 by inverting surface BC concentrations from the Interagency Monitoring of PROtected Visual Environment (IMPROVE) network using a global chemical transport model. We first use active fire counts from the Moderate Resolution Imaging Spectroradiometer (MODIS) to improve the spatiotemporal distributions of the biomass burning BC emissions from the Global Fire Emissions Database (GFEDv2). The adjustment primarily shifts emissions from late to middle and early summer (a 33% decrease in September–October and a 56% increase in June–August) and leads to appreciable increases in modeled surface BC concentrations in early and middle summer, especially at the 1–2 and 2–3 km altitude ranges. We then conduct analytical inversions at both 2° × 2.5° and 0.5° × 0.667° (nested over North America) horizontal resolutions. The a posteriori biomass burning BC emissions for July–September are 31.7 Gg at 2° × 2.5° (an increase by a factor of 4.7) and 19.2 Gg at 0.5° × 0.667° (an increase by a factor of 2.8). The inversion results are rather sensitive to model resolution. The a posteriori biomass burning emissions at the two model resolutions differ by a factor of ~6 in California and the Southwest and by a factor of 2 in the Pacific Northwest. The corresponding a posteriori anthropogenic BC emissions are 9.1 Gg at 2° × 2.5° (a decrease of 48%) and 11.2 Gg at 0.5° × 0.667° (a decrease of 36%). Simulated surface BC concentrations with the a posteriori emissions capture the observed major fire episodes at most sites and the substantial enhancements at the 1–2 and 2–3 km altitude ranges. The a posteriori emissions also lead to large bias reductions (by ~30% on average at both model resolutions) in modeled surface BC concentrations and significantly better agreement with observations (increases in Taylor skill scores of 95% at 2° × 2.5° and 42 % at 0.5° × 0.667°).

We systematically evaluate the black carbon (BC) simulations for 2006 over the Tibetan Plateau by a global 3-D chemical transport model (CTM) (GEOS-Chem) driven by GEOS-5 assimilated meteorological fields, using in situ measurements of BC in surface air, BC in snow, and BC absorption aerosol optical depth (AAOD). Using improved anthropogenic BC emission inventories for Asia that account for rapid technology renewal and energy consumption growth (Zhang et al., 2009; Lu et al., 2011) and improved global biomass burning emission inventories that account for small fires (van der Werf et al., 2010; Randerson et al., 2012), we find that model results of both BC in surface air and in snow are statistically in good agreement with observations (biases < 15%) away from urban centers. Model results capture the seasonal variations of the surface BC concentrations at rural sites in the Indo-Gangetic Plain, but the observed elevated values in winter are absent. Modeled surface-BC concentrations are within a factor of 2 of the observations at remote sites. Part of the discrepancy is explained by the deficiencies of the meteorological fields over the complex Tibetan terrain. We find that BC concentrations in snow computed from modeled BC deposition and GEOS-5 precipitation are spatiotemporally consistent with observations (r = 0.85). The computed BC concentrations in snow are a factor of 2–4 higher than the observations at several Himalayan sites because of excessive BC deposition. The BC concentrations in snow are biased low by a factor of 2 in the central plateau, which we attribute to the absence of snow aging in the CTM and strong local emissions unaccounted for in the emission inventories. Modeled BC AAOD is more than a factor of 2 lower than observations at most sites, particularly to the northwest of the plateau and along the southern slopes of the Himalayas in winter and spring, which is attributable in large part to underestimated emissions and the assumption of external mixing of BC aerosols in the model. We find that assuming a 50% increase of BC absorption associated with internal mixing reduces the bias in modeled BC AAOD by 57% in the Indo-Gangetic Plain and the northeastern plateau and to the northeast of the plateau, and by 16% along the southern slopes of the Himalayas and to the northwest of the plateau. Both surface BC concentration and AAOD are strongly sensitive to anthropogenic emissions (from China and India), while BC concentration in snow is especially responsive to the treatment of BC aerosol aging. We find that a finer model resolution (0.5° × 0.667° nested over Asia) reduces the bias in modeled surface-BC concentration from 15 to 2%. The large range and non-homogeneity of discrepancies between model results and observations of BC across the Tibetan Plateau undoubtedly undermine current assessments of the climatic and hydrological impact of BC in the region and thus warrant imperative needs for more extensive measurements of BC, including its concentration in surface air and snow, AAOD, vertical profile and deposition.

In situ measurements of carbon monoxide (CO) and ozone (O3) at the Pico Mountain Observatory (PMO) located in the Azores, Portugal, are analyzed together with results from an atmospheric chemical transport model (GEOS-Chem) and satellite remote sensing data (AIRS (Atmospheric Infrared Sounder) for CO, and TES (Tropospheric Emission Spectrometer) for O3) to examine the evolution of free-troposphere CO and O3 over the North Atlantic for 2001–2011. GEOS-Chem captured the seasonal cycles for CO and O3 well but significantly underestimated the mixing ratios of CO, particularly in spring. Statistically significant (using a significance level of 0.05) decreasing trends were found for both CO and O3 based on harmonic regression analysis of the measurement data. The best estimates of the possible trends for CO and O3 measurements are −0.31 ± 0.30 (2-σ) ppbv yr−1 and −0.21 ± 0.11 (2-σ) ppbv yr−1, respectively. Similar decreasing trends for both species were obtained with GEOS-Chem simulation results. The most important factor contributing to the decreases in CO and O3 at PMO over the past decade is the decline in anthropogenic emissions from North America, which more than compensate for the impacts from increasing Asian emissions. It is likely that climate change in the past decade has also affected the intercontinental transport of O3.

NOx emissions and O3 chemistry in the Los Angeles (L.A.) Basin during the CalNex-2010 field campaign (May–June 2010) have been evaluated by analyzing O3 and NOy (NO, NO2, HNO3, PAN) observations using a regional air quality model (WRF-Chem). Model simulations were conducted at 4-km spatial resolution over the basin using the Carbon-Bond Mechanism version Z (CBM-Z) and NOx emissions reduced by 24% relative to 2005 National Emissions Inventory (NEI’05), according to recent state emission statistics (BASE_NOx scenario). In addition, a 22–26% NOx emission reduction from weekday to weekend was applied. WRF-Chem reproduced the observed diurnal cycle and day-to-day variations in surface O3, Ox, HNO3 and HCHO (correlation r2 = 0.57 − 0.63; pairs of data n > 400; confidence value p

Geoengineering applications by injection of sulfate aerosols into the stratosphere are under consideration as a measure of last resort to counter global warming. Here a potential regional-scale application to offset the impacts of heat waves is critically examined. Using the Weather Research and Forecasting model with fully coupled chemistry (WRF-Chem), the effect of regional-scale sulfate aerosol emission over California in each of two days of the July 2006 heat wave is used to quantify potential reductions in surface temperature as a function of emission rates in a layer at 12 km altitude. Local meteorological factors yield geographical differences in surface air temperature sensitivity. For emission rates of approximately 30 μg m−2 s−1 of sulfate aerosols (with standard WRF-Chem size distribution) over the region, temperature decreases of around 7 °C result during the middle part of the day over the Central Valley, one of the areas hardest hit by the heat wave. Regions more ventilated with oceanic air such as Los Angeles have slightly smaller reductions. The length of the hottest part of the day is also reduced. Advection effects on the aerosol cloud must be more carefully forecast for smaller injection regions. Verification of the impacts could be done via measurements of differences in reflected and surface downward shortwave. Such regional geoengineering applications with specific near-term target effects but smaller cost and side effects could potentially provide a means of testing larger scale applications. However, design considerations for regional applications, such as a preference for injection at a level of relatively low wind speed, differ from those for global applications. The size of the required injections and the necessity of injection close to the target region raise substantial concerns. The evaluation of this regional-scale application is thus consistent with global model evaluations, emphasizing that mitigation via reduction of fossil fuels remains preferable to considering geoengineering with sulfate aerosols.

Atmospheric mineral dust particles exert significant direct radiative forcings and are important drivers of climate and climate change. We used the GEOS-Chem global three-dimensional chemical transport model (CTM) coupled with the Fu-Liou-Gu (FLG) radiative transfer model (RTM) to investigate the dust radiative forcing and heating rate based on different vertical profiles for April 2006. We attempt to actually quantify the sensitivities of radiative forcing to dust vertical profiles, especially the discrepancies between using realistic and climatological vertical profiles. In these calculations, dust emissions were constrained by observations of aerosol optical depth (AOD). The coupled calculations utilizing a more realistic dust vertical profile simulated by GEOS-Chem minimize the physical inconsistencies between 3-D CTM aerosol fields and the RTM. The use of GEOS-Chem simulated vertical profile of dust extinction, as opposed to the FLG prescribed vertical profile, leads to greater and more spatially heterogeneous changes in the estimated radiative forcing and heating rate produced by dust. Both changes can be attributed to a different vertical structure between dust and non-dust source regions. Values of the dust vertically resolved AOD per grid level (VRAOD) are much larger in the middle troposphere, though smaller at the surface when the GEOS-Chem simulated vertical profile is used, which leads to a much stronger heating rate in the middle troposphere. Compared to the FLG vertical profile, the use of GEOS-Chem vertical profile reduces the solar radiative forcing at the top of atmosphere (TOA) by approximately 0.2–0.25 W m−2 over the African and Asian dust source regions. While the Infrared (IR) radiative forcing decreases 0.2 W m−2 over African dust belt, it increases 0.06 W m−2 over the Asian dust belt when the GEOS-Chem vertical profile is used. Differences in the solar radiative forcing at the surface between the use of the GEOS-Chem and FLG vertical profiles are most significant over the Gobi desert with a value of about 1.1 W m−2. The radiative forcing effect of dust particles is more pronounced at the surface over the Sahara and Gobi deserts by using FLG vertical profile, while it is less significant over the downwind area of Eastern Asia.

[1] To improve estimates of remote contributions of dust to fine particulate matter (PM2.5) in the western United States, new dust particle size distributions (PSDs) based upon scale-invariant fragmentation theory (Kok_PSD) with constraints from in situ measurements (IMP_PSD) are implemented in a chemical transport model (GEOS-Chem). Compared to initial simulations, this leads to reductions in the mass of emitted dust particles with radii

A three-dimensional variational data assimilation (3-DVAR) algorithm for aerosols in a WRF/Chem model is presented. The WRF/Chem model uses the MOSAIC (Model for Simulating Aerosol Interactions and Chemistry) scheme, which explicitly treats eight major species (elemental/black carbon, organic carbon, nitrate, sulfate, chloride, ammonium, sodium and the sum of other inorganic, inert mineral and metal species) and represents size distributions using a sectional method with four size bins. The 3-DVAR scheme is formulated to take advantage of the MOSAIC scheme in providing comprehensive analyses of species concentrations and size distributions. To treat the large number of state variables associated with the MOSAIC scheme, this 3-DVAR algorithm first determines the analysis increments of the total mass concentrations of the eight species, defined as the sum of the mass concentrations across all size bins, and then distributes the analysis increments over four size bins according to the background error variances. The number concentrations for each size bin are adjusted based on the ratios between the mass and number concentrations of the background state. Additional flexibility is incorporated to further lump the eight mass concentrations, and five lumped species are used in the application presented. The system is evaluated using the analysis and prediction of PM2.5 in the Los Angeles basin during the CalNex 2010 field experiment, with assimilation of surface PM2.5 and speciated concentration observations. The results demonstrate that the data assimilation significantly reduces the errors in comparison with a simulation without data assimilation and improved forecasts of the concentrations of PM2.5 as well as individual species for up to 24 h. Some implementation difficulties and limitations of the system are discussed.

We systematically investigate the effects of Wegener-Bergeron-Findeisen (WBF) on BC scavenging efficiency, surface BCair, deposition flux, concentration in snow (BCsnow, ng g−1), and washout ratio using a global 3D chemical transport model (GEOS-Chem). We differentiate riming- versus WBF-dominated in-cloud scavenging based on liquid water content (LWC) and temperature. Specifically, we relate WBF to either temperature or ice mass fraction (IMF) in mixed-phase clouds. We find that at Jungfraujoch, Switzerland and Abisko, Sweden, where WBF dominates, the discrepancies of simulated BC scavenging efficiency and washout ratio are significantly reduced (from a factor of 3 to 10 % and from a factor of 4–5 to a factor of two). However, at Zeppelin, Norway, where riming dominates, simulation of BC scavenging efficiency, BCair, and washout ratio become worse (relative to observations) when WBF is included. There is thus an urgent need for extensive observations to distinguish and characterize riming- versus WBF-dominated aerosol scavenging in mixed-phase clouds and the associated BC scavenging efficiency. We find the reduction resulting from WBF to global BC scavenging efficiency varies substantially, from 8 % in the tropics to 76 % in the Arctic. The resulting annual mean BCair increases by up to 156 % at high altitudes and at northern high latitudes because of lower temperature and higher IMF. Overall, WBF halves the model-observation discrepancy (from −65 % to −30 %) of BCair across North America, Europe, China and the Arctic. Globally WBF increases BC burden from 0.22 to 0.29–0.35 mg m−2 yr−1, which partially explains the gap between observed and previous model simulated BC burdens over land. In addition, WBF significantly increases BC lifetime from 5.7 days to ~8 days. Additionally, WBF results in a significant redistribution of BC deposition in source and remote regions. Specifically, it lowers BC wet deposition (by 37–63 % at northern mid-latitudes and by 21–29 % in the Arctic) while increases dry deposition (by 3–16 % at mid-latitudes and by 81–159 % in the Arctic). The resulting total BC deposition is lower at mid-latitudes (by 12–34 %) but higher in the Arctic (by 2–29 %). We find that WBF decreases BCsnow at mid-latitudes (by ~15 %) but increases it in the Arctic (by 26 %) while improving model comparisons with observations. In addition, WBF dramatically reduces the model-observation discrepancy of washout ratios in winter (from a factor of 16 to 4). The remaining discrepancies in BCair, BCsnow and BC washout ratios suggest that in-cloud removal in mixed-phased clouds is likely still excessive over land.

We examine the distribution of tropical tropospheric ozone (O3) from the Microwave Limb Sounder (MLS) and the Tropospheric Emission Spectrometer (TES) by using a global three-dimensional model of tropospheric chemistry (GEOS-Chem). MLS and TES observations of tropospheric O3 during 2005 to 2009 reveal a distinct, persistent O3 maximum, both in mixing ratio and tropospheric column, in May over the Equatorial Southern Indian Ocean (ESIO). The maximum is most pronounced in 2006 and 2008 and less evident in the other three years. This feature is also consistent with the total column O3 observations from the Ozone Mapping Instrument (OMI) and the Atmospheric Infrared Sounder (AIRS). Model results reproduce the observed May O3 maximum and the associated interannual variability. The origin of the maximum reflects a complex interplay of chemical and dynamic factors. The O3 maximum is dominated by the O3 production driven by lightning nitrogen oxides (NOx) emissions, which accounts for 62% of the tropospheric column O3 in May 2006. We find the contribution from biomass burning, soil, anthropogenic and biogenic sources to the O3 maximum are rather small. The O3 productions in the lightning outflow from Central Africa and South America both peak in May and are directly responsible for the O3 maximum over the western ESIO. The lightning outflow from Equatorial Asia dominates over the eastern ESIO. The interannual variability of the O3 maximum is driven largely by the anomalous anti-cyclones over the southern Indian Ocean in May 2006 and 2008. The lightning outflow from Central Africa and South America is effectively entrained by the anti-cyclones followed by northward transport to the ESIO.

This study used numerical experiments to investigate two important concerns in simulating the cold season snowpack: the impact of the alterations of snow albedo due to anthropogenic aerosol deposition on snowpack and the treatment of snow physics using a multi-layer snow model. The snow albedo component considered qualitatively future changes in anthropogenic emissions and the subsequent increase or decrease of black carbon deposition on the Sierra Nevada snowpack by altering the prescribed snow albedo values. The alterations in the snow albedo primarily affect the snowpack via surface energy budget with little impact on precipitation. It was found that a decrease in snow albedo (by as little as 5–10% of the reference values) due to an increase in local emissions enhances snowmelt and runoff (by as much as 30–50%) in the early part of a cold season, resulting in reduced snowmelt-driven runoff (by as much as 30–50%) in the later part of the cold season, with the greatest impacts at higher elevations. An increase in snow albedo associated with reduced anthropogenic emissions results in the opposite effects. Thus, the most notable impact of the decrease in snow albedo is to enhance early-season snowmelt and to reduce late-season snowmelt, resulting in an adverse impact on warm season water resources in California. The timing of the sensitivity of snow water equivalent (SWE), snowmelt, and runoff vary systematically according to terrain elevation; as terrain elevation increases, the peak response of these fields occurs later in the cold season. The response of SWE and surface energy budget to the alterations in snow albedo found in this study shows that the effects of snow albedo on snowpack are further enhanced via local snow-albedo feedback. Results from this experiment suggest that a reduction in local emissions, which would increase snow albedo, could alleviate the early snowmelt and reduced runoff in late winter and early spring caused by global climate change, at least partially. The most serious uncertainties associated with this part of the study are a quantification of the relationship between the amount of black carbon deposition and snow albedo—a subject of future study. The comparison of the spring snowpack simulated with a single- and multi-layer snow model during the spring of 1998 shows that a more realistic treatment of snow physics in a multi-layer snow model could improve snowpack simulations, especially during spring when snow ablation is significant, or in conjunction with climate change projections.

Forest fires are an important source to carbonaceous aerosols in the Western United States (WUS). We quantify the relative contribution of biomass burning to black carbon (BC) in the WUS mountain ranges by analyzing surface BC observations for 2006 from the Interagency Monitoring of PROtected Visual Environment (IMPROVE) network using the GEOS-Chem global chemical transport model. Observed surface BC concentrations show broad maxima during late June to early November. Enhanced potassium concentrations and potassium/sulfur ratios observed during the high-BC events indicate a dominant biomass burning influence during the peak fire season. Model surface BC reproduces the observed day-to day and synoptic variabilities in regions downwind of but near urban centers. Major discrepancies are found at elevated mountainous sites during the July-October fire season when simulated BC concentrations are biased low by a factor of two. We attribute these low biases largely to the underestimated (by more than a factor of two) and temporally misplaced biomass burning emissions of BC in the model. Additionally, we find that the biomass burning contribution to surface BC concentrations in the USA likely was underestimated in a previous study using GEOS-Chem (Park et al., 2003), because of the unusually low planetary boundary layer (PBL) heights in the GEOS-3 meteorological reanalysis data used to drive the model. PBL heights from GEOS-4 and GEOS-5 reanalysis data are comparable to those from the North American Regional Reanalysis (NARR). Model simulations show slightly improved agreements with the observations when driven by GEOS-5 reanalysis data, but model results are still biased low. The use of biomass burning emissions with diurnal cycle, synoptic variability, and plume injection has relatively small impact on the simulated surface BC concentrations in the WUS.

The column abundance of NO3 was measured over Table Mountain Facility, CA (34.4° N, 117.7° W) from May 2003 through September 2004, using lunar occultation near full moon with a grating spectrometer. The NO3 column retrieval was performed with the differential optical absorption spectroscopy (DOAS) technique using both the 623 and 662 nm NO3 absorption bands. Other spectral features such as Fraunhofer lines and absorption from water vapor and oxygen were removed using solar spectra obtained at different airmass factors. We observed a seasonal variation, with nocturnally averaged NO3 columns between 5−7 × 1013 molec cm−2 during October through March, and 5−22 × 1013 molec cm−2 during April through September. A subset of the data, with diurnal variability vastly different from the temporal profile obtained from one-dimensional stratospheric model calculations, clearly has boundary layer contributions; this was confirmed by simultaneous long-path DOAS measurements. However, even the NO3 columns that did follow the modeled time evolution were often much larger than modeled stratospheric partial columns constrained by realistic temperatures and ozone concentrations. This discrepancy is attributed to substantial tropospheric NO3 in the free troposphere, which may have the same time dependence as stratospheric NO3.

Concurrent tropospheric O3 and CO vertical profiles from the Tropospheric Emission Spectrometer (TES) during the MILAGRO/INTEX-B aircraft campaigns over the Mexico City Metropolitan Area (MCMA) and its surrounding regions were used to examine Mexico City pollution outflow on a regional scale. The pollution outflow from the MCMA occurred predominantly at 600–800 hPa as evident in O3, CO, and NOx enhancements in the in situ aircraft observations. TES O3 and CO are sensitive to the MCMA pollution outflow due to their relatively high sensitivities at 600–800 hPa. We examined O3, CO, and their correlation at 600–800 hPa from TES retrievals, aircraft measurements, and GEOS-Chem model results. TES captures much of the spatial and day-to-day variability of O3 seen in the in situ data. TES CO, however, shows much less spatial and day-to-day variability compared with the in situ observations. The DO3/DCO slope is significantly higher in the TES data (0.43) than the in situ data (0.28) due partly to the lack of variability in TES CO. Extraordinarily high DO3/DCO slope (0.81) from TES observations at 618 hPa over the Eastern U.S. was previously reported by Zhang et al. [Zhang, L., Jacob, D.J., Bowman, K.W., et al., 2006. Ozone–CO correlations determined by the TES satellite instrument in continental outflow regions. Geophys. Res. Lett. 33, L18804. 10.1029/2006GL026399.]. Thus the application of TES CO–O3 correlation to map continental pollution outflow needs further examination. Published by Elsevier Ltd.

An adjoint model for the internationally used Community Multiscale Air Quality (CMAQ) modeling platform of the U.S. EPA is developed. The adjoint version for CMAQ (CMAQ-ADJ) provides the user community with forward (decoupled direct method or DDM) and backward (adjoint) sensitivity analysis capabilities. Current implementation is for gas-phase processes. Discrete adjoints are implemented for all processes with the exception of horizontal advection, for which, because of inherent discontinuities in the advection scheme, the continuous approach is superior. The adjoint of chemistry is constructed by interfacing CMAQ with the kinetic pre-processor, which provides for increased flexibility in the choice of chemical solver and facilitates the implementation of new chemical mechanisms. The adjoint implementation is evaluated both on a process-by-process basis and for the full model. In general, adjoint results show good agreement with brute-force and DDM sensitivities. As expected for a continuous adjoint implementation in a nonlinear scheme, the agreement is not perfect for horizontal transport. Sensitivities of various air quality, public health, and environmental metrics with respect to emissions are calculated using the adjoint method. In order to show applicability to regional climate studies, as an example, the sensitivities of these metrics with respect to local temperatures are calculated.

Satellites provide unique perspectives on aerosol global and regional spatial and temporal distributions, and offer compelling evidence that visibility and air quality are affected by particulate matter transported over long distances. The heights at which emissions are injected into the atmosphere are major factors governing downwind dispersal. In order to better understand the environmental factors determining injection heights of smoke plumes from wildfires, we have developed a prototype system for automatically searching through several years of MISR and MODIS data to locate fires and the associated smoke plumes and to retrieve injection heights and other relevant measurements from them. We are refining this system and assembling a statistical database, aimed at understanding how injection height relates to the fire severity and local weather conditions. In this paper we focus on our working proof-of-concept system that demonstrates how machine-learning and data mining methods aid in processing of massive volumes of satellite data. Automated algorithms for distinguishing smoke from clouds and other aerosols, identifying plumes, and extracting height data are described. Preliminary results are presented from application to MISR and MODIS data collected over North America during the summer of 2004.

We develop and examine a microphysics-based black carbon (BC) aerosol aging scheme that accounts for condensation, coagulation, and heterogeneous chemical oxidation processes in a global 3-D chemical transport model (GEOS-Chem) by interpreting the BC measurements from the HIAPER Pole-to-Pole Observations (HIPPO, 2009–2011) using the model. We convert aerosol mass in the model to number concentration by assuming lognormal aerosol size distributions and compute the microphysical BC aging rate (excluding chemical oxidation aging) explicitly from the condensation of soluble materials onto hydrophobic BC and the coagulation between hydrophobic BC and preexisting soluble particles. The chemical oxidation aging is tested in the sensitivity simulation. The microphysical aging rate is ∼ 4 times higher in the lower troposphere over source regions than that from a fixed aging scheme with an e-folding time of 1.2 days. The higher aging rate reflects the large emissions of sulfate–nitrate and secondary organic aerosol precursors hence faster BC aging through condensation and coagulation. In contrast, the microphysical aging is more than 5-fold slower than the fixed aging in remote regions, where condensation and coagulation are weak. Globally, BC microphysical aging is dominated by condensation, while coagulation contribution is largest over eastern China, India, and central Africa. The fixed aging scheme results in an overestimate of HIPPO BC throughout the troposphere by a factor of 6 on average. The microphysical scheme reduces this discrepancy by a factor of ∼ 3, particularly in the middle and upper troposphere. It also leads to a 3-fold reduction in model bias in the latitudinal BC column burden averaged along the HIPPO flight tracks, with largest improvements in the tropics. The resulting global annual mean BC lifetime is 4.2 days and BC burden is 0.25 mg m−2, with 7.3 % of the burden at high altitudes (above 5 km). Wet scavenging accounts for 80.3 % of global BC deposition. We find that, in source regions, the microphysical aging rate is insensitive to aerosol size distribution, condensation threshold, and chemical oxidation aging, while it is the opposite in remote regions, where the aging rate is orders of magnitude smaller. As a result, global BC burden and lifetime show little sensitivity (< 5 % change) to these three factors.

We present a study on MOPITT (Measurements Of Pollution In The Troposphere) detection of CO emission from large forest fires in the year 2000 in the northwest United States. Fire data used are from the space-borne Advanced Very High Resolution Radiometer (AVHRR) at 1-km resolution. The study shows that MOPITT can reliably detect CO plumes from forest fires whenever there are N30 AVHRR hotspots in a 0.258 � 0.258 grid, which is comparable to the pixel area of MOPITT in the region. The spatial CO pattern during the fire events is found to be consistent with the location and density of AVHRR hotspots and wind direction. While the increase of CO abundance inside the study area is closely correlated to the AVHRR-derived hotspot number in general (RN0.75), the non-linearity of fire emission with fuel consumption is also observed. MOPITT can also capture the temporal variation in CO emission from forest fires through 3-day composites so it may offer an opportunity to enhance our knowledge of temporal fire emission over large areas. The CO emission is quantitatively estimated with a one-box model. The result is compared with a bottom-up approach using surface data including burnt area, biomass density, and fire emission factors. If mean emission factors for the region are used, the bottom-up approach results in total emission estimates which are 10%–50% lower than the MOPITT-based estimate. In spite of the limitations and uncertainties addressed in this study, MOPITT data may provide a useful constraint on uncertain ground-based fire emission estimates.

We examine the driving effects of various factors on technological innovation to renewables (TIRES), focusing on a set of 29 provinces in China, and apply a dynamic panel approach. China has become a leading player in research and development spending in renewables, and the dynamic panel estimators we use prove themselves to be suitable in handling the persistent effect on TIRES. The level of TIRES in the previous periods is positively and highly correlated with that in the current period and confirmed the need for a stable and consistent policy support for renewables. Electricity consumption is the most important driver for all renewables and wind energy, but the driving effect was weaker for solar energy and biomass. Research and development intensity is the most important driver for biomass, but is only the second most important driver for all renewables, solar energy and wind energy. Unexpectedly, electricity price has had significant negative impacts on TIRES, which reveals that lowering electricity prices will lead to higher innovation in renewables. The driving effect of renewable energy tariff surcharge subsidy is not significant, which means that Chinese subsidy policy has not played the desired role. The driving effect of installed renewable energy capacity is also minimal, which may be due to the fact that overcapacity will hinder TIRES in China. This paper may help policy-makers and the industry understand how to promote TIRES in China effectively based on these above influential factors.

We estimate black carbon (BC) emissions in the Western United States (WUS) for July–September 2006 by inverting surface BC concentrations from the Interagency Monitoring of PROtected Visual Environment (IMPROVE) network using a global chemical transport model (GEOS-Chem) and its adjoint. Our best estimate of the BC emissions is 49.9 Gg at 2° × 2.5° (a factor of 2.1 increase) and 47.3 Gg at 0.5° × 0.667° (1.9 times increase). Model results now capture the observed major fire episodes with substantial bias reductions (∼35% at 2° × 2.5° and ∼15% at 0.5° × 0.667°). The emissions are ∼20–50% larger than those from our earlier analytical inversions (Mao et al., 2014). The discrepancy is especially drastic in the partitioning of anthropogenic vs. biomass burning emissions. The August biomass burning BC emissions are 4.6–6.5 Gg and anthropogenic BC emissions 8.6–12.8 Gg, varying with the model resolution, error specifications, and subsets of observations used. On average both increase twofold relative to the respective a priori emissions, in distinct contrast to the halving of the anthropogenic and tripling of the biomass burning emissions in the analytical inversions. We attribute these discrepancies to the inability of the inversion system, with limited spatiotemporal coverage of the IMPROVE observations, to effectively distinguish collocated anthropogenic and biomass burning emissions on model grid scales. This calls for concurrent measurements of other tracers of biomass burning and fossil fuel combustion (e.g., carbon monoxide and carbon isotopes). We find that the inversion system as is has sufficient information content to constrain the total emissions of BC on the model grid scales.