This white paper describes the observational data challenges that impact modern numerical weather forecasts and how these could be overcome with unmanned aerial vehicles (UAV). Small rotary wings UAV have the potential to provide a unique observing system capable of measuring detailed vertical profiles of temperature, humidity, air pressure and wind. These meteorological data – captured within the planetary boundary layer (PBL) – help to determine the potential for severe weather formation and enhance the forecasting ability for atmospheric conditions such as hail, icing and fog formation among others.
Computation of a Numerical Weather Forecast
Mathematically modern weather forecasts are formulated as an initial value problem. Or in other words, knowing the current weather conditions allows mathematicians to abstract a future weather state. Typically, a future state is derived by applying physical laws, like coupled thermo-dynamic and NavierStokes equations. Unfortunately, the above practice suffers from two downsides:
(i) There are no analytical solutions known to the NavierStokes equations. A solution up to a certain degree of resolution can only be found by using numerical approximation.
(ii) The description of the initial (current) state of the atmosphere lacks sufficiently accurate information.
In the last 10 years the computational capacity for running global weather models in a resolution of 10 km or locally even less (1 km) has become more easily available and more affordable. Consequently, facing problem
(iii) is still challenging but manageable off the-shelf. Regarding (ii) major advances have been made due to satellite data that has been assimilated into global models. However, a closer look at the different data sources reveals a data gap in the planetary boundary layer (PBL),
ie. the first 1 to 2 km above ground level (AGL). Even though weather events forming in this layer directly affect us, actual measurements of meteorological parameters are scarce. A common solution for the last 100 years were balloon soundings that collect and deliver readings of different weather parameters. Unfortunately, these balloons are usually lost after deployment as they are carried away with the wind. Moreover, their landing spot is often not easily accessible; thus, they cannot be retrieved after descending. Due to highly-sensitive accurate measurement devices attached to the balloons their operational use is costly. Therefore, balloon soundings are carried out only twice a day at selected locations.
As a consequence, weather phenomena like fog, low stratus and storms cannot properly be predicted as they have their main trigger in the PBL. A number of different attempts have been made to overcome this issue collecting more reliable data.
Remote sensing techniques – either satellite-borne, airborne or ground based – have been designed over the last 20 to 30 years. These remote sensing techniques usually make use of laser (e.g. LIDAR), active or passive microwave (e.g. radiometer) or different types of radar devices. All of these measurement methods share the following downsides: • relatively expensive • limited mobility • designed for one specific use case/ physical parameter • no data in adverse conditions.
Small unmanned aerial vehicles (UAV) do not suffer these problems. Therefore, they can improve on the information gathered in the PBL by directly and accurately measuring prognostic variables. As UAVs are not lost during soundings, several atmospheric profiles can be flown in one session or even for a longer time period. Hence, a temporal evolution of the measured parameters can be observed.
Following three systems are available.
A Close-Up of the Meteodrone Classic
Each UAV is equipped with a ground station that enables a redundant telemetry link to the drone. All flight-relevant parameters are shown on a display attached to the ground station. These parameters are provided in real-time. Hence, they enable the pilot to keep track of the UAV: • Position (moving map) • Altitude and heading • Power consumption • Current weather • Wind conditions.
The primary radio link uses 2.4 GHz and the secondary uses either 433 MHz, 900 MHz (USA) or a frequency according to local regulations. The gathered data is stored on an SD card on the UAV. It further transmits the most important variables to the ground station where they are stored on another SD card. The Classic and XL are equipped with an emergency rescue system (ERS) which is a necessity for flying under BVLOS. The ERS can be triggered from the ground station.
All Meteodrones are pre-programmed to perform vertical ascents/descents with constant climb rates around 3-10 m/s up to an altitude of 3000 m AGL. Custom flight profiles are possible, however. It is strongly recommended to ensure with your aviation safety authority that the aircrafts are considered to be airworthy. Moreover, the air traffic control and the operator must inform other air traffic participants of planned operations, e.g. by issuing NOTAMs, blind calls, restricted airspaces.
The raw meteorological data is sampled with 4 Hz and transmitted to the ground station. In addition, the data is stored on an on-board SD-card. After landing the data can be accessed via Wi-Fi. The raw data is post-processed online. This includes also the transformation into a RAOB format to display soundings. Sequences of flights can be visualized online with charts such as:
Using UAV data in Numerical Weather Forecasting
Assimilation into WRF
Once the data from the Meteodrone have been acquired, they can be straightforwardly ingested into mesoscale models such as MM5 and WRF without implementing any additional forward observational operators. Therefore, we use a data format recognized by WRF which allows for the “recycling” of all existing data assimilation routines for balloon soundings. Depending on the topography and the height of the mission, the radius of influence of the Meteodrone gathered data can be 15 to 45 km. Existing 4d-nudging or 4d-VAR routines can also be used to assimilate the drone data into the initial state of the weather model. In order to estimate meteorological parameters at higher altitudes the atmospheric lapse rate is traditionally calculated using weather station data. However, these calculations are prone to errors. The Meteodrone data enables the measurement of the actual lapse rate which can then be applied to other regions as well resulting in a better estimate.
The following is an example comparing the occurrence of early morning fog and low stratus, one with and one without Meteodrone data assimilated into the WRF model. Column a) shows the satellite cloud cover for 7 AM (upper image) and 8 AM (lower image). The columns b) and c) represent the Swiss1k model forecasts for these two dates. Of special interest here is the area of Lake Constance (North-Eastern Switzerland) which is covered by fog and low stratus. In this particular case, the data of three Meteodrone systems – flown at three locations at the southern westeast axis of the lake (Schaffhausen, Amlikon and Marbach) – were assimilated. In column b) the WRF run is shown without any drone data. Column c) shows how the moisture recorded by the drone flights has been picked up by the model: Fog and low stratus were detected and resolved in the early morning. An additional observation is the model’s inability to pick up the shallow fog on the western Swiss plateau and in northern Italy. In these areas, no supporting Meteodrone data was gathered to help to correct moisture profiles.
Depending on the stability of the atmosphere, topographically induced storms can be observed within Alpine regions. Such a case happened on the 29th May 2017. The image on the upper lefthand side shows the precipitation rendered from the model with Meteodrone data. The red dots indicate flight profiles with their effective radius of impact (red dashed line) taken in Schaffhausen, Amlikon and Marbach – all cities are close to Lake Constance. In comparison to the right, the operational models, without any added drone data, such as ECMWF, NCEP and Met Office, were not able to capture the severe storms that formed in the late evening. The lower left picture shows the radar image data measured at that time.
Icing does not only pose a problem to passenger aviation and helicopters but to UAVs as well. The ice accumulations cause a loss of controllability. Since more UAVs are used in everyday life the risk to the public is increasing. Two conditions have to prevail in order to cause ice accumulations on propellers:
• Air temperature < 0°C • Relative humidity > 95%
Drone enhanced forecasts of temperature, relative humidity and wind speed allow for identifying icing conditions (red area in figure 6).
In this project test-flights in real icing-conditions were conducted. Moreover, icing in different environments was tested: outdoors during winter, in an indoor ski slope and in the Vienna Climatic Wind Tunnel (VCWT). Based on these tests the effect of icing on the Meteodrones was examined and different anti-icing methods were analyzed. A reliable heating method for the propellers is shown in the following image.