Ceiling and Visibility Products for Alaska

Background

Poor weather conditions, particularly restricted visibility and low cloud tops, were the leading cause of fatal general aviation (GA) accidents in Alaska from 2001-2012.  Traditional weather observations from Alaska’s widely dispersed airfields inadequately forewarn of weather likely to be encountered along routes between stations or, in particular, through hazardous mountain passes with localized conditions.  In 2014, the National Transportation Safety Board (NTSB) included “General Aviation: Identify and communicate hazardous weather” on its Most Wanted List to improve transportation safety.

Since 2016, the NCAR/RAL, the MIT/LL, Environmental Prediction Center (EMC), the Alaska Aviation Weather Unit (AAWU), and the FAA have been involved in a collaborative effort to produce a rapidly-updated, high resolution, gridded product of ceiling and visibility (C&V) conditions across Alaska.  This product, known as the Ceiling and Visibility Analysis – Alaska (CVA-AK) has supported AAWU forecasters in developing the Terminal Aerodrome Forecasts (TAFs) for C&V conditions across Alaska at or near instrumented and non-instrumented airfields and along data-sparse routes between airfields including treacherous and heavily-traveled mountain passes.

The CVA-AK product combines ceiling and visibility information from the latest NCEP RAP model with surface-based observations of C&V from the ASOS and AWOS stations across Alaska using data fusion techniques to produce grids of flight category for both ceiling and visibility. The NCEP model C&V forecast data are adjusted using a calibration algorithm that uses 2 hr model forecasts and corresponding observations from the last 30 days to reduce forecast bias. The final product is generated by applying a cloud mask based on GOES satellite data to remove modeled ceiling heights in regions deemed to be cloud free. The final gridded products are updated every 20 min and are viewable by AAWU forecasters on the IC4D display system that they use to produce their aviation weather forecasts.

Model Calibration

Frequency of ceiling heights of 1000 ft or less from (left) RAPv4 and (center) surface met stations along with (right) the frequency bias (M – O) at each surface met station site for (top) January and (bottom) June 2019.
Figure 1. Frequency of ceiling heights of 1000 ft or less from (left) RAPv4 and (center) surface met stations along with (right) the frequency bias (M – O) at each surface met station site for (top) January and (bottom) June 2019.

Understanding the level of consistency between surface-based observations and forecasts of C&V obtained from RAPv4 is critical for developing a useful analysis product. As seen in Figure 1, the model provides much greater detail than can be obtained from surface observations. The model resolves gradients in the frequency of low ceilings moving from peak values in coastal regions to much less frequent occurrence in the interior of Alaska. The model also resolves gradients in the frequency of low ceilings associated with terrain where the higher terrain is more often shrouded in clouds. Despite these attributes of the model, notable biases in the model predictions are also evident in the frequency difference (modeled minus observed frequency) plots (rightmost panels in Figure 1). These comparisons can be used to remove first order bias in the model and to better understand the representativeness of the surface observations.

Calibrated values of ceiling heights for a raw ceiling height of 1 kFT for June 2019. The calibrated values found in most coastal areas are generally greater than 1 kFT in order to correct the tendency of the model to over-predict the frequency of ceiling base heights below 1 kFT.
Figure 2. Calibrated values of ceiling heights for a raw ceiling height of 1 kFT for June 2019. The calibrated values found in most coastal areas are generally greater than 1 kFT in order to correct the tendency of the model to over-predict the frequency of ceiling base heights below 1 kFT.

Biases in the modeled ceiling and visibility forecasts are reduced using a quantile matching algorithm. Quantile matching simply maps the cumulative density function of the model values to that obtained from the observations. The matching is performed at a set of predetermined ceiling heights or visibilities to generate calibration functions for each quantity that are a function of space and time. An example of the calibrated values for a raw ceiling height of 1000 ft is shown in Figure 2. In this example, the largest adjustments to the raw ceiling height values are evident as the darker shades of blue or brown.

Evaluation of Calibration Algorithm

Box and whisker plots showing interquartile range of model  (raw - red and calibrated - green) ceiling height values for an observed category of (top) IFR and (bottom) MVFR obtained for a subset of surface met stations across Alaska for the period 1-30 June 2019. The median value is denoted by a filled circle while the IQR is given by the error bars. The 75th percentile value occasionally lies outside the plotting range (i.e., 10 kFT). Station locations are indicated in the map on the right.
Figure 3. Box and whisker plots showing interquartile range of model  (raw - red and calibrated - green) ceiling height values when the observed category is (top) IFR and (bottom) MVFR. Evaluation is for a subset of surface met stations across Alaska for the period 1-30 June 2019. The median value is denoted by a filled circle while the IQR is given by the error bars. The 75th percentile value occasionally lies outside the plotting range (i.e., 10 kFT). Station locations are indicated in the map on the right.
Table. Heidke Skill Scores computed for all Alaska stations with observation vs model correlations greater than 0.35. Green fill indicates that the calibration improved the skill.
Table 1. Heidke Skill Scores computed for all Alaska stations with observation vs model correlations greater than 0.35. Green fill indicates that the calibration improved the skill. 

In 2019 the calibration algorithm was improved and evaluated using data collected in both January and June 2019. At each station location we determined whether or not the distribution of ceiling height values obtained for a given observed category was improved. Figure 3 (taken from June 2019) provides a general indication of how the calibration algorithm changes the distribution of ceiling height values. Ideally, the full range of model values should fall in the “target range.” It is seen that the calibration algorithm generally increases the value of ceiling height at the 75th quantile for both observed IFR and MVFR categories. This increase in the 75th quantile value is in response to the model’s tendency to predicted low ceiling too often. Despite this increase, the 75th quantile value for observed IFR conditions is generally maintained within the targeted range. At the same time, the calibration algorithm greatly improves the skill at predicting the MVFR category. Examples of this improvement are seen at several stations (e.g., PASC, PAPO, PAKH, PAAD, PAIW), where the median value of ceiling heights was correctly shifted from IFR to MVFR. The overall impact of calibration is to improve the skill in IFR, MVFR and VFR categories for both ceiling and visibility as indicated by the change in the Heidke Skill Score given in Table 1.

Slant Visual Range Study

Figure demonstrating the important distinction between RVR and SVR
Figure 4. The important distinction demonstrated between RVR and SVR

The slant visual range is a critical aspect of aviation especially for General Aviation (GA) and, more recently, remotely piloted vehicles (RPVs) and unmanned aerial system (UAS) operations. GA pilots often require surface-based visual landmarks like rivers and roads to navigate in VFR conditions and also need to be able to clearly see the airport/runway during approach and landing. Under FAA regulations, RPVs and most UAS operations require Visual Line of Site (VLOS). In this case, a surface-based observer looking upward must be able to visually detect the aircraft with the unaided eye. In each of these situations, the requirement is to be able to see on an angle away from the surface; however, current observations obtained by ASOS and AWOS are all reporting horizontal visibility or Runway Visual Range (RVR). Figure 4 demonstrates this important distinction between RVR and SVR. The goal of this study is to assess the impact of the disconnect between RVR measurement and SVR requirements  through literature survey and case study analyses. It is expected that the needs of GA for improved safety and emerging needs of UAS and Urban Air Mobility in terms of SVR will elevate the distinction between RVR and SVR to a critical issue in the years to come.

Plans for 2020

The calibration algorithm will be configured to utilize the Alaska High Resolution Rapid Refresh and run in realtime to supply a data feed of the calibrated products to EMC. This calibrated ceiling and visibility data will be assessed to determine whether or not it might be suitable for use in improving. first guess ceiling and visibility values used to produce the Realtime Mesoscale Analysis (RTMA). In addition, the literature review and case study exploring the importance of SVR to aviation operations including for emerging modes of transportation (e.g., air taxis, delivery drones) will be completed and recommendations will be made on next steps.