conference paper
research report
conference paper
California, where transportation accounts for over half of ozone precursors and particulate matter emissions, as well as nearly 40 percent of greenhouse gas emissions, has adopted the ambitious goal of reducing petroleum use in transportation by 50 percent by 2030. One of the proposed strategies to achieve this goal is to increase the share of alternative fuel vehicles (AFVs). Current incentives to foster the adoption of AFVs by households include single-occupant use of High Occupancy Vehicle (HOV) lanes and parking privileges. Although popular, the effectiveness of these incentives is still controversial. To shed some light on this question, this paper analyzes data from the 2012 California Household Travel Survey using a generalized structural equation model that accounts for residential self-selection, household demographic characteristics, and environmentalism. The results show that, households who live closer to freeways with HOV lanes, work closer to parking lots with AFV privileges, and have pro-environmental views are more likely to own AFVs. However, these households are also likely to drive slightly more than if they had conventional vehicles, which is not surprising since AFVs have a lower variable cost compared to conventional vehicles. On average, a household who lives 1 mile closer to HOV lanes will drive an additional 0.27 mile per month for work or school (0.88 mile if a parking facility with AFV privileges is 1 mile closer to work.) These unintended effects suggest that alternative measures such as pricing vehicle miles traveled (VMT) should be considered if policymakers decide to also reduce VMT.
book/book chapter
published journal article
Autonomous vehicles are cyber-physical systems that combine embedded computing and deep learning with physical systems to perceive the world, predict future states, and safely control the vehicle through changing environments. The ability of an autonomous vehicle to accurately predict the motion of other road users across a wide range of diverse scenarios is critical for both motion planning and safety. However, existing motion prediction methods do not explicitly model contextual information about the environment, which can cause significant variations in performance across diverse driving scenarios. To address this limitation, we propose CASTNet: a dynamic, context-aware approach for motion prediction that (i) identifies the current driving context using a spatio-temporal model, (ii) adapts an ensemble of motion prediction models to fit the current context, and (iii) applies novel trajectory fusion methods to combine predictions output by the ensemble. This approach enables CASTNet to improve robustness by minimizing motion prediction error across diverse driving scenarios. CASTNet is highly modular and can be used with various existing image processing backbones and motion predictors. We demonstrate how CASTNet can improve both CNN-based and graph-learning-based motion prediction approaches and conduct ablation studies on the performance, latency, and model size for various ensemble architecture choices. In addition, we propose and evaluate several attention-based spatio-temporal models for context identification and ensemble selection. We also propose a modular trajectory fusion algorithm that effectively filters, clusters, and fuses the predicted trajectories output by the ensemble. On the nuScenes dataset, our approach demonstrates more robust and consistent performance across diverse, real-world driving contexts than state-of-the-art techniques.
conference paper
Complete and accurate traffic information is becoming more and more available with the advance in transportation surveillance technology. Especially, vehicle classification information can contribute to many transportation related fields such as road pavement management, estimation of polluted emission etc. In previous sections and chapters, it was shown that vehicle signature is function of vehicle type and traffic conditions. By exploiting this concept, the algorithm development in vehicle classification is investigated and corresponding results are presented.
published journal article
Although increasing hidden layers can improve the ability of a neural network in modeling complex nonlinear relationships, deep layers may result in degradation of accuracy due to the problem of vanishing gradient. Accuracy degradation limits the applications of deep neural networks to predict continuous variables with a small sample size and/or weak or little invariance to translations. Inspired by residual convolutional neural network in computer vision, we developed an encoder–decoder full residual deep network to robustly regress and predict complex spatiotemporal variables. We embedded full shortcuts from each encoding layer to its corresponding decoding layer in a systematic encoder–decoder architecture for efficient residual mapping and error signal propagation. We demonstrated, theoretically and experimentally, that the proposed network structure with full residual connections can successfully boost the backpropagation of signals and improve learning outcomes. This novel method has been extensively evaluated and compared with four commonly used methods (i.e., plain neural network, cascaded residual autoencoder, generalized additive model, and XGBoost) across different testing cases for continuous variable predictions. For model evaluation, we focused on spatiotemporal imputation of satellite aerosol optical depth with massive nonrandomness missingness and spatiotemporal estimation of atmospheric fine particulate matter łeq 2.5 mu textm (PM2.5). Compared with the other approaches, our method achieved the state-of-the-art accuracy, had less bias in predicting extreme values, and generated more realistic spatial surfaces. This encoder–decoder full residual deep network can be an efficient and powerful tool in a variety of applications that involve complex nonlinear relationships of continuous variables, varying sample sizes, and spatiotemporal data with weak or little invariance to translation.
published journal article
Background: There are now 3.1 million breast cancer (BC) survivors with 4.5 million survivors projected by 2030. This high and growing burden speaks to the need of identifying modifiable factors that influence BC survival. Particulate matter (PM) air pollutants have been associated with increased mortality for cardiovascular disease (CVD) and several cancers. However, few studies have examined the impact of air pollution on mortality among BC cases. Within the Multiethnic Cohort (MEC), we examined the association between outdoor air pollution and mortality among African American (AA), Latino (LA), Japanese American (JA), and White (WH) women diagnosed with BC. Methods: Kriging interpolation of air pollution data from air monitoring stations was used to estimate PM2.5, PM10, and nitrogen oxides (NOx, NO2) exposures for 3,089 BC cases in the MEC, residing largely in Los Angeles County, linked to residential histories from date of diagnosis to date of death or 12/31/2013. Cox proportional hazard models were used to examine the association between time-varying air pollutants and mortality, accounting for age at diagnosis, race/ethnicity, marital status, chemotherapy, hormone therapy, radiation, surgery, stage, grade, hormone receptor status, tumor size, body mass index, smoking, alcohol, age at first birth, diabetes, CVD, and neighborhood socioeconomic status. All-cause, BC, CVD, non-BC/non-CVD mortality outcomes were evaluated. Stratified analyses were conducted by race/ethnicity, hormone receptor status, and stage. Results: Among the 3,089 BC cases, there were 1,125 all-cause deaths (474 BC, 272 CVD, and 379 non-BC/non-CVD) with an average of 8.1 years of follow-up. PM2.5 (per 10μg/m3), PM10 (per 10μg/m3), NOx (per 50 ppb), and NO2 (per 20 ppb) were associated with risk of all-cause (hazard ratios (HRs) range=1.25-1.72; p’s<0.004) and CVD mortality (HR range=1.62-3.93; p’s<0.0005). For AA cases, PM2.5, PM10, and NOx were associated with risk of CVD mortality (HR range=1.69-3.81; p’s<0.04). For WH cases, PM2.5, PM10, and NOx were associated with risk of all-cause mortality (HR range=1.48-3.06; p’s<0.02). Risk patterns were similar for LA and JA cases although associations did not reach statistical significance. For hormone receptor positive BC, PM2.5 and NOx were associated with risk of all-cause and CVD mortality (HR range=1.59-2.92, p’s<0.006). For hormone receptor negative BC, NOx was associated with risk of CVD mortality (p<0.01). For localized BC, PM2.5 and NOx were associated with risk of all-cause, CVD, and non-BC/non-CVD mortality (p’s<0.04). For advanced BC, PM2.5 and NOx were associated with CVD mortality (p’s<0.005). Conclusion: These findings provide strong evidence that air pollution is associated with overall and CVD mortality among BC cases, although no effect was found on risk of death due to BC. Moreover, it is essential to maintain stringent clean air laws. Future confirmatory studies and better understanding of the mechanisms of action are needed.Citation Format: Iona Cheng, Juan Yang, Chiuchen Tseng, Jun Wu, Shannon M Conroy, Salma Shariff-Marco, Scarlett Lin Gomez, Alice Whittemore, Daniel O Stram, Loic Le Marchand, Lynne R Wilkens, Beate Ritz, Anna H Wu. Association between outdoor air pollution and breast cancer survival: The Multiethnic Cohort Study [abstract]. In: Proceedings of the Twelfth AACR Conference on the Science of Cancer Health Disparities in Racial/Ethnic Minorities and the Medically Underserved; 2019 Sep 20-23; San Francisco, CA. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2020;29(6 Suppl_2):Abstract nr C050.
published journal article
The toll design problem (TDP) provides a quantitative approach to the design of road pricing schemes. Its practical use, however, can be computationally challenging if the formulated TDP requires time-consuming computer models to evaluate candidate designs, especially if such designs must account for multiple objectives. For TDPs to be of practical relevance to the real-world planning of sustainable transportation networks, efficient TDP solution heuristics must be developed. To this end, two surrogate-based solution heuristics for multi-objective TDPs are proposed in this paper. Surrogate-based optimization uses simple approximations to computationally expensive models in order to accelerate the discovery of good solutions. The general search strategy of the proposed heuristics is as follows. In each iteration of the heuristics, a pool of candidate pricing schemes with unique sets of tolling locations and associated tolling levels is generated. From this pool of designs, the heuristics use the surrogate models to screen for solutions that are expected to be nondominated and that meet a specified selection criterion. Then, these promising designs are evaluated by the computationally expensive models, and the outputs obtained from these evaluations are used to update the surrogate models. Both heuristics repeat this general process until a maximum number of iterations are completed, at which point the best TDP solutions are returned. In addition to the solution heuristics, this paper presents a transportation network paradox that highlights how transportation network interventions intended to reduce traffic emissions could have unintended effects on a population’s exposure to pollutants. The paradox also is used to illustrate the practical complexity of accounting for environmental inequality objectives, as well as the relevance of multi-objective analysis approaches to transportation planning. Formulations of multi-objective TDPs that consider both travel and pollutant exposure-related objectives are also presented, including the objectives of reducing human intake of vehicle-generated air pollutants and of minimizing environmental inequality. The Sioux Falls and Chicago Sketch networks were used in tests that examined the relative performance of the heuristics, as well as the characteristics of pricing configurations obtained under different budget constraints. Among other results, the tests show that a pricing configuration could decrease total pollutant intake and environmental inequality, while at the same time producing an increase in pollutant concentrations in a significant number of pollutant receptor points.
published journal article
Abstract Flood and debris hazards are heightened following wildfires, but are challenging to quantify due to interdependence between fire frequency and severity, runoff and sediment fluxes during storms, and sedimentation that reduces infrastructure capacity. Herein we present a stochastic simulation framework to estimate compound flood and debris hazards from sequences of wildfires and rainstorms and the accumulation of sediment within flood infrastructure. Application of the framework to a hypothetical watershed representative of southern California shows that the present‐day compound hazard may be up to 6 times greater than the marginal hazard posed by peak flows in the absence of wildfire, and that future compound hazards could be up to 11 times greater than the marginal hazard based on future increases in wildfire frequency. Numerous sensitivities are investigated, including infrastructure design and maintenance, which are shown to be crucial for moderating future increases in post‐fire flood hazards. , Plain Language Summary More frequent and intense wildfires lead to increased risks of flooding and debris hazards following rainstorms. However, existing models for estimating hazards to communities do not account for the filling of protective flood infrastructure with sediment, which reduces capacity. We present an original model for estimating post‐fire flood and debris hazards that captures the interconnected influences of wildfire, rainstorms, and reduced infrastructure capacity from sedimentation. The modeling approach can simulate present and future hazards to aid both short‐term and long‐term risk management efforts. Simulations show that present‐day hazards are up to six times greater based on the interaction of processes compared to flood hazards in the absence of fires. Based on future increases in wildfire frequency, future flood hazards could be up to 11 times greater. Simulations also show that hazards can be reduced with more intensive cleaning and maintenance. , Key Points Post‐fire flooding represents a compound hazard based on interdependence between fire, runoff, and infrastructure sedimentation A new modeling framework is formulated and applied to simulate compound post‐fire flood hazards using stochastic inputs Compound post‐fire flood hazards can be up to an order of magnitude greater than flood hazards in the absence of fire