科研成果

International trade separates consumption of goods from related environmental impacts, including greenhouse gas emissions from agriculture and land-use change (together referred to as “land-use emissions”). Through use of new emissions estimates and a multiregional input-output model, we evaluated land-use emissions embodied in global trade from 2004 to 2017. Annually, 27% of land-use emissions and 22% of agricultural land are related to agricultural products ultimately consumed in a different region from where they were produced. Roughly three-quarters of embodied emissions are from land-use change, with the largest transfers from lower-income countries such as Brazil, Indonesia, and Argentina to more industrialized regions such as Europe, the United States, and China. Mitigation of global land-use emissions and sustainable development may thus depend on improving the transparency of supply chains.

This paper presents a CFD modeling of deNOx process in a coal-fired power plant selective catalytic reduction (SCR) system, with focus on the transient hydrodynamics of multi-species flow and the influence of vortex on the deNOx process. For this purpose, a comprehensive CFD model is established, parameter study and model validation are performed, and the hydrodynamics, vortex evolution and species concentration distribution are numerically investigated. Simulation results indicate that many vortices with various scale/intensity/shape exist in the SCR system, causing apparent pressure pulsations and velocity fluctuations. High-intensity eddies are mainly distributed in the deflector group Ι, the NH3 nozzles, the static mixer, and the right part of the rectifying grille. The number of eddies decreases significantly with reducing the unit loads. Affected by vortex evolution, the NH3 concentration fluctuates in the SCR system, especially in the vertical flue. The deNOx process completes within 6 s, and the ammonia slip is less than 1.0 ppm, which well meets the requirement of industrial standards. In addition, the static mixer severely destroys the velocity uniformity but favors the mixing of NH3 and NOx. The rectifying grille improves the uniformity of flow field and species concentration field significantly.

Phosphate addition is commonly applied as an effective method to remediate lead contaminated sites via formation of low solubility lead phosphate solids. However, subsequent transport of the lead phosphate particles may impact the effectiveness of this remediation strategy. Hence, this study investigates the mechanisms involved in the aggregation of lead phosphate particles and their deposition in sand columns as a function of typical water chemistry parameters. Clean bed filtration theory was evaluated to predict the particle deposition behavior, using Derjaguin–Landau–Verwey–Overbeek (DLVO) theory to estimate particle-substrate interactions. The observed particle deposition was not predictable from the primary energy barrier in clean bed filtration models, even in simple monovalent background electrolyte (NaNO3), because weak deposition in a secondary energy minimum prevailed even at low ionic strength, and ripening occurred at ionic strengths of 12.5 mM or higher. For aged (aggregated) suspensions, straining also occurred at 12.5 mM or higher. Aggregation and deposition were further enhanced at low total P/Pb ratios (i.e., P/Pb = 1) and in the presence of divalent cations, such as Ca2+ (≥ 0.2 mM), which resulted in less negative particle surface potentials and weaker electrostatic repulsion forces. However, the presence of 5 mg C/L of humic acid induced strong steric or electrosteric repulsion, which hindered particle aggregation and deposition even in the presence of Ca2+. This study demonstrates the importance of myriad mechanisms in lead phosphate deposition and provides useful information for controlling water chemistry in phosphate applications for lead remediation.

This paper identifies and estimates the causal effect of an intervention on repeatedly measured units that co-exist and interact with one another in a social network, when the dichotomous intervention is not randomly assigned and the network evolution may be driven by choices of social agents. We adopt the potential outcome framework and develop identification assumptions to define and identify three estimands, namely, the direct treatment effect, the spillover effect, and the general treatment effect.  Our framework incorporates  social network ties as part of the joint treatment and treats longitudinal networks as variables rather than constants. It also considers complicated causal paths generated by interdependent outcomes. We propose a model-based estimation strategy and use a factor analysis to correct for biases caused by latent homophily. By imputing potential outcomes based on simultaneous equations, we disentangle spillover effects from direct treatment effects and explicitly estimate first-order and higher-order causal effects. The proposed method is easy to implement and flexible to accommodate a wide variety of networks. 

Video temporal grounding is a challenging task in computer vision that involves localizing a video segment semantically related to a given query from a set of videos and queries. In this paper, we propose a novel weakly-supervised model called the Multi-level Attentional Reconstruction Networks (MARN), which is trained on video-sentence pairs. During the training phase, we leverage the idea of attentional reconstruction to train an attention map that can reconstruct the given query. At inference time, proposals are ranked based on attention scores to localize the most suitable segment. In contrast to previous methods, MARN effectively aligns video-level supervision and proposal scoring, thereby reducing the training-inference discrepancy. In addition, we incorporate a multi-level framework that encompasses both proposal-level and clip-level processes. The proposal-level process generates and scores variable-length time sequences, while the clip-level process generates and scores fix-length time sequences to refine the predicted scores of the proposal in both training and testing. To improve the feature representation of the video, we propose a novel representation mechanism that utilizes intra-proposal information and adopts 2D convolution to extract inter-proposal clues for learning reliable attention maps. By accurately representing these proposals, we can better align them with the textual modalities, and thus facilitate the learning of the model. Our proposed MARN is evaluated on two benchmark datasets, and extensive experiments demonstrate its superiority over existing methods.