That matters because icing is not simply a question of whether ice appears on a propeller. The aircraft has to keep producing enough thrust while the rotor geometry changes. Motors and ESCs have to supply the extra torque. A battery that is already cold has to provide both propulsion energy and, if fitted, de-icing or anti-icing heat. Sensors need to keep reporting useful data. The controller has to cope with vibration, imbalance and shrinking control margin. Then the test program has to show that all of those effects remain inside a defined operating envelope.
UAF's tower makes one part of that chain repeatable. It can create calibrated icing and leave enough vertical room to observe hover dynamics. It also exposes the next test-method problem: the current site does not have a horizontal wind component, so it cannot reproduce every airflow and droplet-impingement condition that a forward-flying aircraft will see.
That limitation is not a criticism of the facility. It is the reason controlled icing work is valuable. A useful test program needs repeatability first, then a deliberate path from component tests to representative flight.
Small rotors can lose performance before the ice looks dramatic
Ice changes a propeller in several ways at once. It changes the leading-edge shape and surface roughness, adds mass, can add that mass unevenly, and changes the aerodynamic load on the motor. A rotor can therefore remain physically intact while the propulsion system moves rapidly away from its clean-blade operating point.
A 2026 Aerospace Science and Technology study followed the three-dimensional development of ice on a rotating UAV propeller and measured thrust and power through the icing process. In the authors' test condition, the early rough-ice stage already produced a 10-20% reduction in thrust coefficient with about a 20% increase in power consumption. Later, after runback icing and larger feather-like and horn-like structures developed, the reported thrust-coefficient loss reached 86% and power required to hold the intended rotational speed rose by more than 300% after 90 seconds of icing.
Those numbers belong to one experimental setup and should not be transferred directly to a different rotor, droplet field or aircraft. The useful result is the shape of the failure. Performance degradation was not linear with visible ice mass. Once the accretion changed the leading-edge geometry enough, thrust, power and rotational stability deteriorated together.
That has a practical consequence for small multirotors. A system cannot define its icing limit only by "minutes in icing" or by an arbitrary visible-ice thickness. It needs a propulsion margin tied to the specific rotor, RPM range, forward speed and atmospheric condition.
A separate 2026 forward-flight experiment reinforces the point. Researchers at Iowa State University used phase-locked imaging and three-dimensional scanning to characterize ice on a multirotor propeller in an icing research tunnel while the propeller experienced forward-flight conditions. That work is important because a hovering rotor and a translating rotor do not see the same relative airflow or droplet trajectories.
The propulsion penalty is also a battery-range penalty
A drone can survive the aerodynamic effect of icing and still lose the mission to energy consumption.
A 2023 natural flight-testing campaign on a fixed-wing UAV documented ice on the wings, fuselage, stabilizers, Pitot probe and propeller. The iced propeller required more than 80% additional power to complete the same mission compared with the non-icing case. The study also observed structural vibration and a blocked Pitot probe that produced false airspeed readings.
The figure is not a universal multiplier for iced drones. It is a warning about where the system margin disappears. A mission planner may start with a comfortable battery reserve based on clean-air current draw, then encounter a condition that raises propulsive power while cold temperature is reducing available battery performance. If the aircraft also uses electrical heating, the protection system draws from the same finite energy store.
That coupling makes a simple "heater on or off" design insufficient. The useful quantity is remaining mission energy after propulsion degradation, thermal protection and cold-battery effects have all been accounted for.
For an electric multirotor, the test log should therefore connect rotor condition to current draw, bus voltage, motor speed and available thrust. A controller that automatically increases motor command can hide the aerodynamic loss for a while. The aircraft still appears to hold altitude, but the electrical system is spending reserve to do it.
The same problem appears in a different form on a fixed-wing aircraft. Ice raises drag and can reduce propeller efficiency, so the vehicle may need more shaft power to hold speed and altitude while its glide margin is also getting worse. Energy reserve and aerodynamic reserve are not independent.
Heating competes with propulsion for the same limited power
UAF doctoral researcher Morakinyo Labiran described the first intuition behind anti-icing neatly: heat the surface. His follow-up is the important part. The problem becomes harder because a drone has limited power capacity, real material constraints and complicated ice adhesion.
A 2023 study of an electro-thermal ice-protection system for a 53 cm UAV propeller shows that gap between concept and implementation. The researchers integrated carbon-fiber heating elements into the propeller and tested it at 4200 rpm in an icing wind tunnel at -5 °C, -10 °C and -15 °C with a liquid water content of 0.44 g/m³.
The system prevented the measured performance penalty at -5 °C. At the colder test points it limited accretion and retained useful thrust, but the experiments also required significantly higher heat flux than the preceding CFD analysis predicted.
That last result is as important as the successful anti-icing case. A heater sized only from a thermal model can consume too little power to protect the rotor when the real droplet field, convective cooling, material stack and rotating boundary layer arrive together.
There is also a design conflict between where heat is most useful and where electrical hardware is easiest to place. The blade leading edge is the critical collection surface, but power delivery to a rotating component, temperature uniformity, conductor durability and rotor balance all become engineering constraints. A heater that prevents ice but introduces unacceptable imbalance or structural weakness has not solved the aircraft-level problem.
For small aircraft, the protection system should therefore be budgeted as part of propulsion from the beginning. The relevant question is not simply how many watts the heater needs. It is how much thrust and endurance remain when that heat is being generated continuously under the coldest approved condition.
Liquid water content can matter more than a simple temperature threshold
"Icing weather" is not defined by air temperature alone.
Supercooled liquid water has to be present, and droplet size changes how water strikes, spreads and freezes on the aircraft. UAF's overview notes the aviation distinction between smaller cloud droplets and larger supercooled droplets such as those associated with freezing drizzle and freezing rain. Larger drops can spread or shatter on impact and freeze farther aft on a surface.
Natural-cloud data makes the operational consequence clearer. A 2024 study flew a multirotor with rotor heating 59 times into supercooled stratus clouds between -3 °C and -10 °C. The researchers quantified icing severity using the rate of increase in battery current and compared it with co-located liquid-water-content measurements. Across the tested liquid water content range of 0.02 to 0.5 g/m³, icing severity was strongly positively correlated with liquid water content.
In those specific conditions, the relationship was stronger than the relationship with temperature or mean droplet size. That does not make droplet size unimportant in general. Laboratory propeller work shows that droplet diameter and icing regime can materially change accretion shape and torque. It means a weather decision based only on "below freezing" leaves out a variable directly connected to how much water is available to freeze.
The 2024 work also compared remote-sensing estimates of liquid water content with co-located in-situ measurements and found useful agreement for assessing the tested conditions. That points toward a better operational model: an icing envelope informed by the actual cloud-water field, not only a surface-weather temperature and a generic icing flag.
For drone operators, this is still an emerging capability rather than a universal preflight product. The engineering implication is simpler. If a test campaign never records liquid water content and droplet characteristics, its icing result is difficult to reproduce and difficult to compare with another facility.
Protecting the propeller does not protect every sensor
The propulsion system gets most of the attention because thrust loss is immediate and measurable. The 2023 flight campaign shows why aircraft-level icing validation cannot end there.
Ice blocked the test aircraft's Pitot probe and produced false airspeed readings. At the same time, ice was accumulating on the airframe and propeller and the aircraft experienced increased structural vibration.
That creates several coupled failure paths. A flight controller may increase power because propeller efficiency has fallen. An iced airspeed sensor may report a value that no longer represents the actual flow. Rotor imbalance can increase vibration seen by inertial sensors and the camera. Ice on an antenna enclosure, camera window or other exposed surface can create another degradation that has little to do with propeller thrust.
The exact vulnerable sensors depend on the aircraft. A hovering multirotor may not use a Pitot system at all, while a fixed-wing cargo UAV may depend heavily on airspeed for energy management and stall margin. A mapping platform may remain controllable but lose the image quality or navigation accuracy required for its actual mission.
This is why an icing test needs mission-success criteria as well as continued-flight criteria. "The aircraft stayed in the air" is too low a bar for an inspection drone if vibration has made its imagery unusable, or for a survey aircraft if navigation and camera timing no longer meet the accuracy requirement.
A useful test also distinguishes transient and residual effects. Does a sensor recover as soon as the aircraft leaves the icing cloud? Does ice remain on the airframe after the protected rotor has shed it? Does a de-icing event temporarily create vibration or a control transient? These are system questions, not blade questions.
Controlled icing gives repeatability, but the test still needs a representativeness ladder
SWEAT solves an important experimental problem: repeatability.
UAF says the tower can create multiple calibrated icing conditions while a drone flies inside the enclosure. Its approximately 16-foot height gives researchers enough room to hover and observe flight dynamics in real time. That is far stronger evidence than waiting for a convenient freezing-fog day and hoping two test runs see comparable cloud water.
But UAF also states the current limitation plainly. The site lacks a horizontal component for wind scenarios, and the team hopes to expand the facility to support fixed-wing work.
That makes the tower a good example of how a qualification program should be layered.
At the component level, a stationary or rotating rig can isolate a propeller, coating or heater and measure ice shape, thrust, torque, RPM and heat flux. A controlled hover test adds the flight controller, battery, vibration environment and real vehicle airflow. A wind-tunnel or forward-flow test adds translation and changes droplet impingement. Natural-cloud flight finally introduces the variability, spatial gradients and combined weather effects that a lab deliberately suppresses.
No single layer is sufficient on its own. Natural-cloud flight is representative but hard to reproduce. A controlled tower is repeatable but cannot reproduce every combination of wind, forward speed, droplet distribution and cloud structure. The strongest evidence comes from using one layer to calibrate the next.
That sequence also prevents a common development mistake: changing three things at once. If a new coating, new heater control law and new propeller are all introduced before the baseline is measured, the team may know that a flight improved without knowing why.
An icing test program should measure margin, not just survival
A practical small-UAS icing program can be organized around the margins that disappear as ice accumulates.
For the atmosphere, record temperature, liquid water content, droplet-size information where available, exposure time and airflow. For propulsion, record RPM, thrust or thrust proxy, torque where the rig permits it, current, voltage and motor or ESC temperature. For the airframe, record mass change, ice location and post-test condition. For navigation and control, log attitude error, control output, vibration and sensor-health flags. For the mission, record whether the payload still produces usable data.
Battery state needs special attention because average current alone can hide the part that matters near the end of a mission. The important result is how much energy and voltage margin remain while propulsion and protection loads are both elevated.
The test should also have predefined stop criteria. A research program learns more from a repeatable threshold than from flying until the aircraft can no longer hold altitude. The threshold could be tied to control saturation, minimum thrust reserve, a battery-voltage limit, a sensor-integrity failure or a mission-quality limit. The right choice depends on the aircraft and intended operation.
Repeated exposures matter as well. Heating cycles, water ingress, blade coatings, connectors and composite interfaces may survive one event and degrade after many. If an aircraft is meant to fly regular winter cargo routes, the relevant reliability question is not whether a fresh prototype survives one icing encounter.
Finally, the baseline should include clean cold-weather flight. Cold-soaked batteries, stiff lubricants, condensation and altered material properties can change performance without any ice accretion. Subtracting that baseline is necessary before attributing every current increase or control change to ice.
"All weather" should be replaced with a defined icing envelope
The engineering target is not a drone that is vaguely "icing capable." It is an aircraft with a measured operating envelope and known exit criteria.
That envelope should state the atmospheric conditions that were tested, the duration of exposure, the airspeed or hover condition, the payload and battery state, the anti-icing configuration, and the minimum residual performance. If the aircraft depends on heating, its endurance calculation has to include that load. If the protection works only down to a particular temperature or within a liquid-water-content range, that boundary belongs in the operating procedure.
The UAF tower is useful precisely because it can turn some of those variables into controlled inputs. Current propeller research adds the measurements needed to connect ice geometry with power and thrust. Natural-cloud work shows that liquid water content can be strongly connected to the electrical signature of icing. Flight testing shows that the problem can spread from propulsion into vibration and sensor integrity.
Together, those results make the design problem more concrete. Small-drone icing is an energy-management problem, an aerodynamic problem, a sensing problem and a validation problem at the same time.
A heater can be part of the solution. A coating can be part of the solution. Better weather information can be part of the solution. None of them, by itself, establishes that the aircraft has enough margin to finish the mission safely.
The useful evidence is a test chain that shows when ice begins to change the aircraft, how quickly the electrical and aerodynamic reserves shrink, which sensors remain trustworthy, and where the approved envelope ends.