Integrated Modeling for Road Condition Prediction Phase 3 Project Report

Chapter 2. Project Description

The multidisciplinary nature of transportation systems management and operations (TSMO) can be exemplified in road weather-responsive management strategies (WRMS) that bring together meteorology, traffic management, law enforcement, maintenance, and traveler information to support agency decision-making and influence travel behavior. Similar needs are present in work zone management, traffic incident management, and active traffic management strategies. All of these strategies seek to provide actionable information to travelers, enabling them to make better choices for safe and reliable travel, and to agencies, enabling them to minimize and mitigate the impact of disruptions. Initiatives across all of these disciplines have been working toward developing similar frameworks and methodologies.

Travelers today have higher expectations for their travel experiences because of these management strategies and private-sector innovations in gathering, processing, and disseminating of information. Travelers have moved from being passive consumers of information with unknown accuracy to being vital players in generating and validating information. This trend will accelerate with deployment of connected vehicle systems, which will create a powerful new platform for collecting and sharing information. Within this context, the role of prediction and forecasting will become more important to choices made by travelers, as well as to agency decisions in the active management of surface transportation operations. Economic imperatives require freight carriers and logistics providers to factor in a variety of traffic conditions for planning routes, times, and delivery schedules.

While a wide variety of approaches have been proposed in the scientific literature for predicting traffic conditions, the development, application, and adoption of these approaches by operating agencies have been limited. Part of the reason has been limited data availability, a situation that is changing rapidly. Another reason is that available tools have been too narrowly focused and have not taken full advantage of developments in related disciplines and domains. As a result, the use of predictive strategies in TSMO remains nascent, and decisions continue to largely be reactive.

To support proactive operations, higher-quality predictions are needed for incorporating factors beyond the fundamental traffic models into the analysis. Previous efforts incorporating forecast weather conditions in traffic predictions¹ have shown considerable promise in improving the relevance and usefulness of traffic predictions for agency decision-making related to incoming weather. However, the richness and usefulness of traffic predictions can be enhanced by augmenting the forecast weather condition during the prediction window with known and likely capacity constraints (such as planned work zones or snow route restrictions or incidents). Incorporating reported conditions from environmental sensor stations (ESS), mobile observations from fleets, and citizen reports may further improve the predictions. Current and planned road treatment approaches, snowplow routing, parking restrictions, and maintenance decisions might be factored in as well.

Phase 3 of the Integrated Modeling for Road Condition Prediction (IMRCP) demonstration project has provided an opportunity to develop, deploy, and evaluate an integrated model for predicting road conditions that incorporates transportation and non-transportation data, deterministic and probabilistic data, and measured and reported data into a framework for agency decision-making and traveler information. The model provides a practical tool for transportation agencies to support traveler advisories and maintenance and operational decisions at both strategic and tactical levels.

Methodology and Approach

One of the main challenges in deploying an integrated prediction model is accounting for the different latencies, qualities, and levels of certainty in source data and methods to generate consistent, accurate, and useful results. For example, in the Utah Department of Transportation (UDOT), citizen and maintenance personnel reports of road conditions are subjectively weighted in decision-making based on the timing of the report and nature of the decision. The in-house meteorologists weigh all observed conditions along with ensemble weather forecasts to make a call on the road condition forecast.² Creating methods and a framework to accommodate those considerations is a significant undertaking.

The model also needs to translate road-segment-based information into meaningful and actionable information within a corridor and across the network. This is where it is critical to understand the impact of road conditions on traffic. Being able to translate road weather conditions into traffic impacts can inform the route, mode, and time choices of travelers. This also calls for integrating recent developments in using probabilistic information and forecasts in decision-making processes as new information becomes available. For example, optimal routing for both individuals and service vehicles (package delivery, repair vehicles, etc.) requires different algorithms than are typically used in deterministic conditions.

The recommendations from predictive models need to be conveyed in simple, understandable terms to the end user and enable querying the system for additional information. For example, using probabilistic models may entail communicating probabilities to the traveler, a concept traditionally avoided by departments of transportation but commonly employed in meteorology.

The products of the phase 3 project are the IMRCP system software and documentation, a demonstration, an evaluation of the demonstration, and an analysis of the experience. To facilitate use by State and local transportation agencies, including those involved in the demonstration, the model has had to be easy to use, rely on available data sources, integrate with existing legacy systems, generate timely predictions, and ultimately provide decision support to operators in a useful manner.

Project Tasks and Deliverables

Phase 1 identified user needs and set broad requirements for a demonstration of IMRCP capability. It surveyed the existing field of predictive models, engaged a broad stakeholder community, and developed a concept of operations and requirements for an integrated model for predicting road conditions that incorporates transportation and non-transportation data, deterministic and probabilistic data, and measured and reported data.

Phase 2 started with follow-on efforts to develop a system architecture and design with input from the IMRCP project stakeholders. A foundational system was then implemented and deployed in a suburban Kansas City study area in cooperation with the Kansas City Scout (KC Scout) transportation management center (TMC), which is operated cooperatively by the Missouri Department of Transportation (MODOT) and the Kansas Department of Transportation (KDOT). The effectiveness of the system's ability to incorporate real-time and archived data and results from an ensemble of forecast and probabilistic models to predict the current and future overall road/travel conditions was then evaluated.

Phase 3 builds on the prior work to more deeply investigate operations applications. The Kansas City study area was expanded from the congested commuting corridor used in phase 2 to include the entire metropolitan area monitored by the KC Scout TMC. The Traffic Estimation and Prediction System (TrEPS) dynamic traffic assignment (DTA) model deployed in phase 2 was enhanced with additional scenario management and calibrations. An alternative machine learning-based prediction (MLP) for traffic forecasting was added, based on several years of detailed traffic data from the TMC and other sources. System documentation was updated to reflect the enhancements and respond to prior user feedback, then reposted with the system code to the U.S. Department of Transportation's (USDOT) open-source application development portal on GitHub.³ The system was operated for 18 months through two winter seasons, during which the system was enhanced to address operational challenges as they occurred. A formal evaluation was undertaken to identify achievements and challenges in system operations and in the operations response at KC Scout.

Taken together, phase 3 activities have formalized many of the modeling techniques, built operational experience with the IMRCP, and provided a more thorough understanding of the opportunities and limitations associated with the integrated model.

Phase 3 began with a review of documentation from the prior phases to uncover any gaps between the documentation and the deployed system capabilities. The review also identified opportunities for system and documentation improvements in the phase 3 development and deployment.

The IMRCP stakeholder engagement plan was updated in phase 3 to reflect the expansion plan and focus on operational rather than developmental activities. The engagement plan identified the stakeholders, schedule for engagement, webinar formats and approach, and main outcomes expected from each engagement with the group. The plan was used to guide stakeholder interactions throughout the project life cycle.

The system architecture and design description were updated in phase 3 to reflect the enhancement of the TrEPS model, addition of the MLP traffic forecast, and accommodations for expansion of the roadway network model. The map interface was significantly upgraded to improve performance and reliability when looking at broader views of the metro, regional, and national weather conditions. Descriptions of each of these efforts are provided later in the document.

Deployment of the upgraded phase 3 IMRCP system required significant expansion of the roadway network configuration and data collectors. The study area expanded from a significant high-congestion corridor to the entire metro area monitored by the KC Scout TMC. Traffic and weather data for the entire metro area over several years were compiled and used in the training and calibration of the MLP traffic model. The system acceptance test plan was updated and executed against the expanded and completed system.

An evaluation of the IMRCP system and its use were conducted as part of the phase 3 deployment. The purpose of the evaluation was to explore what impact IMRCP had on KC Scout operations and assess whether or not the information was useful to the KC Scout operators and supervisors. The findings from the evaluation could be used to inform others who may be considering similar deployments, and provide the Federal Highway Administration with information to help determine next steps for IMRCP.

Stakeholder Engagement

Stakeholder involvement in the IMRCP project was needed to understand user needs and potential modeling and data constraints. Meteorologists helped in understanding forecast models and ensemble methods. Traffic researchers and modelers assisted in understanding current and emerging models and predictive capabilities. DOT maintenance workers and DOT traffic operations personnel were needed for gathering user needs, including use of modeling and prediction in operational decision-making. Third-party information provider and traveler needs were collected from DOT traveler information managers.

Staff at KC Scout TMC, Missouri Department of Transportation (MODOT), and Kansas Department of Transportation (KDOT) have been particularly supportive of the demonstration study area deployment in the Kansas City metropolitan area. They have participated in stakeholder engagement and activities with other IMRCP stakeholders, supported the development team with data and access to their transportation management systems, used the demonstration IMRCP alongside their TMC systems, and provided input to the evaluation.

Specific stakeholder meetings and webinars included:

• Presentation to ITS Heartland Annual Meeting — April 24, 2018.
  • ITS Heartland is the regional ITS America association including MODOT, KDOT, and three other States. The presentation provided an opportunity to describe the IMRCP efforts with KC Scout to a broader set of WRMS stakeholders.
• Stakeholder core working group kickoff — July 12, 2018.
  • Stakeholders were introduced to the phase 3 IMRCP project, scope, and objectives. The deployed and operating phase 2 system was demonstrated as a basis for soliciting input on opportunities for enhancement in phase 3. Stakeholder participation opportunities were also described.
• IMRCP phase 3 MODOT update — August 17, 2018.
  • An update and demonstration were provided to KC Scout and MODOT personnel to solicit renewed operations interest for the phase 3 enhancements.
• National Operations Center of Excellence (NOCoE) webinar — August 30, 2018.
  • An introductory presentation and demonstration were provided through a NOCoE webinar to a national audience interested in TSMO.
• Road Weather Management Program (RWMP) Stakeholder Meeting — September 18, 2018.
  • The project objectives, phase 2 system deployment and accomplishment, and phase 3 plans were presented with KC Scout to attendees at the RWMP meeting. Demonstrations were available to attendees in the technology vendor area.
• Stakeholder update — January 31, 2019.
  • The phase 3 objectives and progress were described with a system demonstration for a broad audience of interested stakeholders.
• Presentation to ITS Heartland Annual Meeting — April 30, 2019.
  • An update on the Kansas City model deployment was provided to the ITS Heartland Annual Meeting. Results from the 2018–2019 winter season of operations were summarized.
• Stakeholder core working group update — June 27, 2019.
  • An update on the Kansas City model deployment was provided to the IMRCP core working group. Results from the 2018–2019 winter season of operations were summarized and plans for the evaluation and next season of operations were provided.
• Stakeholder core working group update — August 21, 2019.
  • The project team met with the core working group to discuss the KC Scout operating experience, evaluation, broader potential applications for WRMS, and the upcoming season.
• RWMP Stakeholder Meeting — August 29, 2019.
  • Phase 3 objectives and results were presented to RWMP meeting attendees by the KC Scout representative. Demonstrations were available to attendees in the technology vendor area.