Weather Impacts on Unmanned Aerial Systems

Background

Unmanned aerial vehicles (UAVs) are rapidly becoming an important component of the national airspace, especially at low altitudes (less than about 1,500 ft or 450 m AGL).  To cope with the ever-increasing UAV traffic demands NASA is developing the Unmanned Aerial Vehicle Traffic Management System (UTM), which includes separation management, scheduling, demand capacity imbalance, contingency management, trajectory definition and prediction, and importantly, wind and weather.  Since many UAVs are very small in size they are quite susceptible to turbulence- and gust-induced loss-of-control and trajectory excursions.  Therefore, these weather-induced excursions are an important weather factor to consider in the management system is atmospheric turbulence.

Over the past three years RAL scientists have worked under a NASA-funded program aimed at (1) establishing  automated means for forecasting turbulence and gust levels in the atmosphere that affect UAVs, (2) developing and testing translation algorithms to provide a turbulence hazard metric given the state of atmospheric turbulence, and (3) coordinating weather-related  workshops for the UAV user community. 

Task 1 involves modifying NCAR’s Graphical Turbulence Guidance (GTG) (Sharman et al. 2006, 2017) which has been operationally available since 2003 on NOAA’s Aviation Digital Data Service (ADDS), hosted by the National Weather Service’s Aviation Weather Center (AWC) and is available 24 x7 at http://www.aviationweather.gov/adds/.   Uniquely, the GTG product provide forecasts and nowcasts of atmospheric turbulence intensity levels in terms of a well-known turbulence intensity metric known as the energy (or eddy) dissipation rate (EDR).  EDR is the International Civil Aviation Organization (ICAO) mandated metric for atmospheric turbulence, but computing EDR for operational use requires input from numerical weather prediction (NWP) models, which are too coarse to properly resolve atmospheric eddies that affect UAVs.  So the major emphasis has been on developing techniques to properly quantify EDR at low levels from relatively coarse NWP model output. 

Task 2 takes the output EDR from Task1 and maps this atmospheric metric to an aircraft (i.e., UAV) specific response.  For traditional winged systems, given information about the vehicle characteristics (e.g., stability derivatives, or size, weight, airspeed, and wing surface area), the turbulence response can be predicted using standard aerodynamic modeling methods.

FY2018 Accomplishments: Task 1   

An improved low-level turbulence (LLT) forecast algorithm was developed and implemented within the Graphical Turbulence Guidance (GTG) system. The new algorithm is based upon recent observational results from Muñoz-Esparza et al. (2018) and leverages the recently proposed statistical remapping technique introduced by Sharman and Pearson (2017). Within the model=-derived atmospheric boundary layer (ABL), the new LLT algorithm uses remapping to a lognormal distribution for nighttime (stable) conditions and to a log-Weibull distribution for daytime (convective) conditions to derive EDR from quantities such as the parameterized subgrid turbulence kinetic energy. The surface total heat flux is employed to determine the instances in which the ABL is convective or stable. Moreover, different suites of turbulence indices are used depending upon the atmospheric stability, to improve forecasting of turbulence under the different forcing mechanisms in stable and convective ABLs.

A 1-yr-long GTG calibration was performed using the WRF-based HRRR operational NWP product, which has a horizontal grid resolution of 3 km and runs over the CONUS, on the basis of 5-h forecasts with a latency of 1 h. Determination of the optimum CAT and MWT ensemble combinations was performed by minimizing the mean absolute percentage error (MAPE) to identify the suite of indices that provides the best agreement not only for high intensity (i.e. EDR) turbulence cases but for EDRs across the entire range of possible values, since smaller EDRs are relevant for applications such as unmanned aerial systems, takeoff and landing maneuvers, wake vortex decay, and wind energy. Validation using EDR observations from the BAO tower for the period of 9–30 March 2015 was carried out, coinciding with the XPIA campaign (Fig. 1a). The proposed LLT algorithm reduces the MAPE with respect to the GTG implementation described in Sharman and Pearson (2017) for CAT by a factor of ~2 (MAPE = 55%), as shown in the MAPE distributions from Fig. 1b. Moreover, the probability of detection within a 20% and 50% range for characteristically small and large ABL values is improved by a factor of ~1.3–2.3, respectively. In addition, EDR forecasts with GTG using the new LLT algorithm exhibit realistic features such as larger daytime values relative to nighttime over land and enhanced EDR in regions of complex terrain (Fig. 1c).

Figure 1: (a) Time evolution of EDR forecasts with 5-h lead time from the CAT GTG ensemble for the new LLT algorithm using the HRRR model compared to the current operational GTG v3.0 and in-situ tower observations at the BAO tower. (b) Mean absolute percentage error (MAPE) distributions. (c) Example of the new GTG LLT forecast that based on 5-h WRF-HRRR forecasts, showing contours of EDR (m2/3 s-1) at a height above the surface z = 168.5 m valid at 1800 UTC 15 Mar 2017.
Figure 1: (a) Time evolution of EDR forecasts with 5-h lead time from the CAT GTG ensemble for the new LLT algorithm using the HRRR model compared to the current operational GTG v3.0 and in-situ tower observations at the BAO tower. (b) Mean absolute percentage error (MAPE) distributions. (c) Example of the new GTG LLT forecast that based on 5-h WRF-HRRR forecasts, showing contours of EDR (m2/3 s-1) at a height above the surface z = 168.5 m valid at 1800 UTC 15 Mar 2017.

FY2019 Plans

The initial calibration of the LLT GTG algorithm was based only on the XPIA data.  The use of other data sets at other locations and times of year will be assessed.  This may require the introduction of latitude-dependent mappings. Large eddy simulations will also be used to help further calibrate GTG performance.  In addition the LES output will be used to better characterize turbulence in the vicinity of UAV test sites\ by comparing the output to coarser resolution GTG output.

Task 2 Accomplishments

This activity is intended to implement a full non-linear, six degree of freedom fixed-wing UAS flight simulation capability. The key aspect of this work is to include the effects of arbitrary wind fields, via the induced aerodynamic forces and moments. In order to actuate these forces and moments, a simplified - though adequate for small UAS – method will be employed. This approach, termed the “four point method” computes the aerodynamic effect the wind has on the vehicle by assuming the local wind field can be approximated by linear functions, evaluated at the four points. These points are located at the vehicle center of gravity (COG), at a point along the left-hand wing, a symmetric one on the right-hand wing, and a point at the tail. The differential vertical wind component along the wing produces rolling moments, the differential vertical wind component from wing to tail produces pitching moments, and the differential lateral wind component from wing to tail produces yawing moments. Lift forces are computed via the vertical wind component at COG, and drag forces are computed from the longitudinal wind component at the COG.

Based on the need for the wind components at the four points on the vehicle described above, it can be seen that the spatial resolution of the wind field needs to be on the order of the vehicle size, e.g., in the meter scale for a small UAS. The goal of the simulation is to utilize small-scale LES model data as the input; however, to cover the extent of a UAS flight path, the typical spatial resolution of these models is in the tens of meters. Furthermore, even if the LES model is run at tens of meters scales, the inherent filtering of the model means that accurate resolution of the wind field occurs at, say, eight times those scales. In a previous project, a method was developed to merge a small-scale simulated turbulence wind field with a cloud model output. The idea is that, if the model can at least resolve some scales that are within the so-called universal inertial subranges (the classic “Kolmogorov -5/3rds law”), then a subgrid turbulence field can be generated and “matched” (in intensity) to the LES output. In order to merge the data, a low-pass filter is applied to the LES data and a high-pass filter is applied to the subgrid turbulence data – with the filter cutoff frequencies set appropriately. A similar approach, though with a different implementation, was used in this work. The initial results are presented below.

The vertical wind component of an LES wind field, at 25 meter resolution, was interpolated to one meter resolution. An averaged power spectrum of this field was computed along a single line through the field, and the energy dissipation rate (EDR) was computed. With an assumed length scale (chosen “by eye”), the mean square value of the wind field was then computed from the EDR. The subgrid turbulence was generated at one meter resolution, with the same length scale and at unit mean square value, and then scaled by the mean square value from the LES data. As described above, low- and high-pass filters were applied to the LES and subgrid winds, respectively, then arithmetically summed, point-by-point. Figure 2 shows the average power spectrum of the vertical component of the subgrid turbulence field (black) and that for the high-pass filtered field (red). Figure 3 shows a time series for a segment of the data used in Figure 2. Figure 4 shows the averaged spectrum of the interpolated LES data (black), the scaled subgrid turbulence (blue), and the merged wind field (green). The magenta points indicate those wavenumbers over which the EDR value was calculated from the LES data. From the spectrum of the merged wind field, it can be seen that the merging approach is working. That is, there are no apparent discontinuities in the power spectrum of the merged field. Figure 5 shows a segment of the time series used to produce Figure 4. The black curve is the LES data, the blue curve is the subgrid turbulence data, and the red curve is the merged field.

Figure 2. Averaged power spectrum of the subgrid turbulence field (black) and high-pass filtered data (red).
Figure 2. Averaged power spectrum of the subgrid turbulence field (black) and high-pass filtered data (red).
Figure 3. Time series of a section of the data shown shown in Figure 6.
Figure 3. Time series of a section of the data shown shown in Figure 6.

 

Figure 4. Averaged power spectrum of interpolated LES data (black), subgrid turbulence data (blue), and merged data (green). The magenta points indicated the wavenumbers over which the EDR was calculated from the LES data.
Figure 4. Averaged power spectrum of interpolated LES data (black), subgrid turbulence data (blue), and merged data (green). The magenta points indicated the wavenumbers over which the EDR was calculated from the LES data.
Figure 5. Example time series of LES data (black), subgrid turbulence data (blue), and merged data (red).
Figure 5. Example time series of LES data (black), subgrid turbulence data (blue), and merged data (red).

FY2019 Plans

Develop an automated wind, wind gust and turbulence forecast/nowcast system for UAVs.  We will continue to improve and validate turbulence forecast models with observational data to determine the balance between accuracy, precision, and execution time.

UAS impact metrics and translation algorithms Develop relevant UAV impact metrics based on steady and unsteady winds for various platforms and verify with UAV developers. A new focus will be on the development of a multi-rotor vehicle simulation capability. The intention of this task is to provide NASA with translation algorithms which take as input the HRRR fields, turbulence diagnostics (GTG), and small-scale modeling results from Task 1 and outputs vehicle-specific impact products. Verification using field test data will be performed if such data is available.

Develop an initial UTM Weather Concepts document. This is a user interaction task, which focuses on the development of an initial UTM Weather Concepts document.  This will be developed in collaboration with a UTM Weather Advisory group and the NASA UTM group. The advisory group will contain a cross-section of members from the UTM community, including public organizations, private companies, and research organizations.