1.5
Using stochastic, dynamic weather-impact models in strategic traffic flow management

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Tuesday, 25 January 2011: 9:30 AM
Using stochastic, dynamic weather-impact models in strategic traffic flow management
310 (Washington State Convention Center)
Mengran Xue, University of Michigan, Ann Arbor, MI; and S. M. Zobell, S. Roy, C. P. Taylor, Y. Wan, and C. Wanke
Manuscript (1.6 MB)

In order to effectively and efficiently manage the airspace, traffic flow managers need to make strategic plans beyond two hours into the future. When planning for a response to a potential weather event, strategic planning becomes a challenging problem because of the significant amount of uncertainty in weather evolution at these long look-ahead times. As such, there is not often a single plan that can manage the disruptions from all possible weather outcomes well.

Flow Contingency Management (FCM) is proposed as decision support tool and operational concept for generating and managing sets of contingency plans that anticipate several future weather impact outcomes. In order to generate these weather impact outcomes we must first capture the set of weather scenarios. For this purpose we consider the use of ensemble weather forecast members, where each member or cluster of similar members represents a potential weather future. Currently-available probabilistic forecasts may not be suitable for this purpose because it is difficult to determine multiple potential weather and weather-impact futures from these forecasts.

In conjunction with this process we are developing a stochastic weather impact simulation model as a tool for studying and evaluating FCM concepts. This simulator is able to generate multiple simulated weather impact outcomes for each set of initial conditions, and allows simple analysis of weather-impact statistics at critical airspace locations. At its essence, the simulator aims to capture the propagation of weather-impact among airspace regions using simple probabilistic selection and adaptation rules. This abstract description permits scenario-generation and statistical analysis with little computational effort, while capturing the rich temporal and spatial propagations/correlations that are observed in real weather. A key need is to parameterize the model to match a particular forecast period's weather forecasts so that the model generates weather futures that are likely for that particular forecast period. We envision using ensemble-forecast data for this parameterization, and so the link between ensemble forecasting and weather-impact modeling is critical to our development.

This paper describes the FCM concepts and the approaches being developed to manage the risks of uncertain weather outcomes, and the development and evaluation of the weather impact simulation. This paper also discusses the weather forecast requirements to support these concepts, particularly the use of ensemble members.

To give the reader a more detailed understanding of our research goals, let us briefly outline the key ideas that we expect to develop in the full paper.

Topic 1: Flow Contingency Management and Weather-Impact Modeling. At a decision horizon of over two hours, weather and other uncertainties significantly modulate National Airspace System (NAS) dynamics, and interdependencies among traffic throughout the NAS impact performance. At this time horizon, it is thus critically important to identify contingencies for allocating NAS-wide resources to traffic flows for potential weather scenarios. To design flow contingencies, we require tools for representing the evolving impact of uncertain weather propagation on NAS parameters (capacities, traffic-flow characteristics, etc). These tools must be able to 1) simulate forecast weather-impact scenarios and 2) allow computation of weather-impact statistics (including spatial correlations), in a way that captures the intrinsic uncertainty in weather-impact propagation at the time-horizon of interest yet is simple enough for significant tractability (including in cohesion with traffic flow models). Our belief is that existing weather forecasting capabilities do not directly yield weather-impact models that meet our criteria, yet these forecasting capabilities must be leveraged in developing impact models.

Topic 2: Weather-Impact Simulator—Idea and Data Requirements. Our group is pursuing development of a spatio-temporal weather-impact simulator. The simulator tracks weather-impact statuses at contiguous geographical regions in the airspace. The impact statuses are viewed as propagating through the airspace due to simple probabilistic influences among the regions; such an influence-modeling framework can capture the spatial correlation, generation, and decline of weather impact, while also tractably representing the significant uncertainty that is present in weather-impact propagation at a day-long time horizon.

To leverage the weather-impact simulator in FCM, we will need to parameterize (construct) the simulator on each use, to produce likely weather scenarios for the time horizon of interest. We expect that parameterizing the simulator will require a variety of data, including 1) ensemble weather forecasts, 2) current weather conditions, and (possibly) 3) historical data, 4) terrain maps, etc. Of these, ensemble forecast data is perhaps most likely to be a limiting factor in parameterization, and so an important aspect of our work is to identify what ensemble forecast data is needed for parameterization at a sufficient resolution and accuracy. As first steps in this direction, we are studying conversion of weather forecasts to weather-impact forecasts, and parameterization of stochastic network dynamics from forecast snapshots. More broadly, we are also studying the question of how much data is needed for network model inference, as needed for the weather-impact simulator. Yet, we are far from fully understanding what data (and how much) is needed for constructing the weather-impact simulator. By adapting the promising parameterization approaches listed above to various possible data paradigms, we will aim to better understand what data is needed for model parameterization.

Topic 3: Determining Whether Ensemble Forecasts Meet the Data Requirements. In tandem with understanding what forecast data is needed for model parameterization, we will aim to understand forecast capabilities of ensemble forecasts, to see whether these capabilities meet the requirements. As a first step in this direction, we will complete a literature review of ensemble forecasting, with a focus on understanding 1) prediction outputs, 2) methods for ensemble-member selection, and 3) computational complexity of these models. Through such a survey, we will hope to determine whether the ensemble forecasts can provide snapshots for parameterizing the weather-impact simulator in real time. If there is a gap between required data and available ensemble-forecast outputs, we will aim to clearly identify what further data is needed, with the hope of fostering discussion with ensemble-forecast specialists on how these additional data needs can be met.