Prediction of Convective Storm Hazards for Aviation

Background

The Next Generation Air Transportation System (NextGen) is a national priority designed to meet the air transportation needs of the United States in the 21st century—in particular, a significant growth in demand for air traffic services, possibly on the order of two to three times today's demand levels. In addition, the number of commercial applications of Unpiloted Aerial Systems (UAS) has been growing extremely rapidly over the past few years with primary operating space being the lowest 400 feet of the atmosphere. The expected increase in congestion of the NAS requires improved detection and prediction of weather hazards and their translation into air traffic flow impacts in order to maintain aviation safety and improve the efficiency of Air Traffic Flow Management. For the past several years, the Aviation Applications Program within NCAR’s Research Applications Laboratory (RAL) has been engaged in multiple FAA-funded research and development efforts geared toward improved convective weather support on topics ranging from lightning impacts on airport operations and subsequent ripple-effects throughout the National Airspace System (NAS), to developing state-of-the-art CONUS-scale short term predictions of convective storms for tactical-to-strategic time scale planning of the NAS, to optimization of global-scale probabilistic predictions of convective storms in support of the ICAO-led harmonization of global weather hazard products.

FY2018 Accomplishments

Convective Weather Forecast System Bridging Tactical-to-Strategic Planning Time Horizons.

Figure 1. Image examples from NWP V5.4.4 demonstrating the performance of the (left) 2 hour forecast of precipitation when compared with (right) corresponding observed precipitation intensity valid as obtained from the new NWP display.
Figure 1. Image examples from NWP V5.4.4 demonstrating the performance of the (left) 2 hour forecast of precipitation when compared with (right) corresponding observed precipitation intensity valid as obtained from the new NWP display.

CoSPA is a forecast system that produces 0-8 hour forecasts of convective storm intensity and convective cloud top heights by merging extrapolation and model based forecasts using image processing techniques and forecast heuristics.  It was developed for the FAA for improved convective weather forecasts ranging from tactical to strategic time scales. CoSPA was developed through a collaboration between MIT Lincoln Laboratory, NOAA Earth System Research Laboratory, and NCAR RAL. The inputs to CoSPA blending include MIT-LL multi-scale advected VIL and Echo Tops and longer range model forecasts from the High Resolution Rapid Refresh (HRRR) Model. As the blending algorithms developed in RAL are reaching maturity, a final set of enhancements was funded by the FAA in order to meet the requirements of end users. Key upgrades over the past year include a combination of algorithmic improvements, bug fixes and most importantly an extraordinary increase in processing speed (see Table 1). The system latency has been reduced by an order of magnitude compared to the legacy version of CoSPA. This increased processing speed was critical for meeting end user requirements of a 5 min update rate. The faster update rate also necessitated the development of a new regime to handle nowcasted Convective Initiation for lead times between 1 and 2.5 hours.  Finally, new code was developed to extend the processing domain further north into Canada and south to encompass Puerto Rico.

Table 1. Current version of blending algorithm. The NWP-latest version is currently undergoing technology transfer to the FAA.
Table 1. Current version of blending algorithm. The NWP-latest version is currently undergoing technology transfer to the FAA.

The latest version of the blending system (V5.4.4) was delivered to the FAA in Oct 2018 and is being implemented as part of the NextGen Weather Processor (NWP). An example of the latest version’s performance is shown in Figure 1. We continue to support a legacy version of the blending called CoSPA which will continue to be made available year-round to aviation planners (i.e., ARTCCs, the FAA Command Center, the Aviation Weather Center and airline industry partners) via a web-based display until the NWP comes online.

Lightnings Impact on Airport Terminal Operations and Safety.

Figure 2. Total traffic delay and safety risk cost accumulated over 3 months for an airport. The solid line represents total costs based on the 3 mile / 6 minute rule, the dotted line represents the 5 mile / 10 minute rule and the dashed line the 5 mile / 15 minute rule. The different colored icons represent the variety in cost based on lightning safety rules derived from different lightning sources.
Figure 2. Total traffic delay and safety risk cost accumulated over 3 months for an airport. The solid line represents total costs based on the 3 mile / 6 minute rule, the dotted line represents the 5 mile / 10 minute rule and the dashed line the 5 mile / 15 minute rule. The different colored icons represent the variety in cost based on lightning safety rules derived from different lightning sources.

Work has also been performed to quantify the trade-offs between airport efficiency and safety when lightning impacts an airports through a collaboration with AvMet. The safety risk experienced by personnel working outdoors at airports when lightning is in the vicinity has been quantified using detailed models that determine the likelihood of being struck by lightning and the associated economic impact of a person being struck.  The impacts of airport closures on the NAS have been simulated using AvMet’s Dynamic Airspace Routing Tool (DART). Analyses of air traffic data in conjunction with lightning data and DART simulations suggest that lightning-induced ramp closures have substantial impacts on traffic in and out of an airport and cause ripple effects through the NAS that result in significant delays.  Furthermore, analyses of traffic simulations for two specific airports suggests optimal rulesets for lightning safety at these airports to be roughly 5mile/10minutes (example is shown in Figure 2).  In addition, results suggest that the mitigation of human factors would often increase safety and decrease inefficiencies. This points to opportunities for better lightning guidance containing improved information about lightning hazards and associated risk.

Ensemble Prediction of Oceanic Convective Hazards (EPOCH).

Table 2.  Brier Skill Scores for 2-model and 4-model combinations using the logistic regression (LR) approach and bias-corrected relative frequency (BCRF).  GC = GEFS + CMCE; GCEU adds ECMWF and UK Met Office. Data are from 24 hour forecasts for the Caribbean region in JJA 2013.
Table 2.  Brier Skill Scores for 2-model and 4-model combinations using the logistic regression (LR) approach and bias-corrected relative frequency (BCRF).  GC = GEFS + CMCE; GCEU adds ECMWF and UK Met Office. Data are from 24 hour forecasts for the Caribbean region in JJA 2013.

Work on improving probabilistic forecast computation techniques continued this past year, with the overarching goal of achieving a practical, transparent approach that could be easily implemented for operational use.  We examined in more detail two computation techniques: The logistic regression (LR) approach and the bias-corrected relative frequency (BCRF) approach. The latter approach is currently being used in EPOCH and requires the availability of all the ensemble members (typically on the order or 20 model runs per center).  The former approach uses only a single representation of predictors (e.g., the ensemble mean) so significantly reduces the bandwidth requirement for producing forecasts.

Figure 3. Comparison of Bias-Corrected Relative Frequency (RF) with Logistic Regression (LR) as a function of probability category in the Brier score decomposition (reliability – resolution), with negative values indicative of adding skill. Scoring is as compared to bias-corrected GEFS. Data same as that used for Table 2.
Figure 3. Comparison of Bias-Corrected Relative Frequency (RF) with Logistic Regression (LR) as a function of probability category in the Brier score decomposition (reliability – resolution), with negative values indicative of adding skill. Scoring is as compared to bias-corrected GEFS. Data same as that used for Table 2.

The LR approach is attractive in comparison to BCRF because of increased reliability in forecast probability levels > 70%, particularly for models with a consistent pattern of over-confidence.  However, the increased reliability comes at the cost of sharpness.  For example, in the Caribbean region as shown in Figure 3, the comparison of the 2-model (North American combination) and 4-model (North American and European combination) forecasts for both the LR and BCRF approaches show distinctly different distributions of skill as a function of probability.  While the Brier Skill Scores shown in Table 2 are similar for the BCRF vs. LR, the skill is achieved in quite different ways.  For the LR the skill is primarily derived from probabilities in the 40 to 60% range, while for the BCRF (RF) the skill is derived from probabilities near 0 and above 70%, indicating a much sharper forecast in both the two and four model combinations.  This pattern is consistent with season and most regions studied (South America is a outlier due to overall very low performance). These findings indicate that for risk-adverse users who might prefer a sharper forecast,  the BCRF approach could provide advantages over that obtained with LR.

FY2019 PLANS

RAL will continue to support the technology transfer of the latest version of the blending to the FAA for inclusion in the NextGEN Weather Processor (NWP). Additional work this year will focus on evaluating the performance of various lightning nowcasting techniques for a range of atmospheric conditions in order to improve safety and efficiency of airport operations when thunderstorms are nearby and developing a circular pamphlet describing improved guidelines for airport operations under direct impact of thunderstorms. Finally, work will continue on the EPOCH system. Enhancements will include updates to the calibration algorithm to support higher resolution model inputs and increased latitude coverage.  Other updates will seek to optimize probabilistic forecast skill, bandwidth requirements, and sharpness through examination of net expected cost and other metrics.   These products are aimed at aiding aviation weather forecasters in their development of guidance products as well as aid airline dispatchers in their strategic planning for transoceanic flights.