How Birds Soar to Extreme Heights Using Thermal Air Currents

Researchers applied reinforcement learning to teach a simulated glider to soar using thermals (rising columns of warm air). The model used Rayleigh-Bernard convection equations to simulate turbulent atmospheric conditions. After extensive training, the AI learned to search for thermals, enter them, and execute tight loops to maximize altitude gain. Remarkably, the most important features for success were vertical acceleration and vertical velocity gradient, not temperature as many glider pilots assumed. This demonstrated that machine learning could discover optimal strategies that even expert humans might not consciously recognize.

Research groups like NASA and Dan Edwards at North Carolina State University have developed autonomous soaring systems for UAVs. These systems use simple mathematical models (like Gaussian distributions) to represent lift patterns and continuously update these models based on sensor observations. The UAVs fly straight lines, detect lift when encountered, attempt to center themselves in thermal cores, and then move to find new lift sources. In October 2006, Dan Edwards achieved a 97 km autonomous cross-country flight, demonstrating that machines can learn to exploit atmospheric lift patterns. The challenge lies in developing algorithms that can recognize and respond to complex atmospheric phenomena as effectively as human pilots.

Autonomous soaring enables unmanned aerial vehicles to extend flight duration by capturing energy from wind structures such as thermals and dynamic wind gradients, using exploration-exploitation algorithms that balance wind mapping with energy collection; this involves Gaussian Process-based wind field estimation combined with reinforcement learning approaches to learn optimal control policies for energy-efficient flight, addressing the fundamental challenge of how aerial platforms can gather wind data while simultaneously using that data to sustain prolonged flight operations.

This video presents research on enabling electric fixed-wing UAVs to achieve energy-efficient cross-country soaring by using end-to-end reinforcement learning, where an agent with an LSTM-based policy network learns to balance path tracking against exploiting atmospheric updrafts for height gain, validated through simulation-to-real-world transfer experiments.

Once thermals are detected, Gaussian distribution models approximate lift patterns with central cores of stronger lift. An EKF-like algorithm estimates four thermal parameters (position, strength, width) using variometer measurements. The estimated center becomes the L1 controller target. This approach handles observability challenges through covariance matrices. The system manages trade-offs between circling tighter for stronger lift versus efficiency losses from tighter turns, dynamically adapting to thermal characteristics.

Dynamic soaring is the flight technique that allows albatrosses to cover nearly 1,000 km in a day without flapping their wings. It works by exploiting vertical gradients in wind speed: wind travels faster at higher altitudes above the ocean surface. The albatross dives with the wind to gain speed, then turns into the wind to exchange kinetic energy for altitude, repeating this S-shaped pattern continuously.

For dynamic ridge lift soaring, pilots maximize glider performance by using control lines in the minimum sink rate position. Constant ridge distance is maintained by observing terrain and using body weight to correct distance when lift is felt. When the ridge is crossed, wind becomes more horizontal, creating risk of being blown behind the ridge. Pilots must remain windward of the ridge. Wind speed affects heading angle selection, with stronger winds requiring more orientation into the main wind direction. Foot trim and accelerator are used to compensate for wind changes and maintain optimal flight path.

This extensive segment teaches two primary methods for extending paragliding flight duration. Ridge lift occurs when wind deflects off terrain obstacles, creating upward currents that pilots follow by maintaining close proximity to ridgelines. Thermals form when solar heating creates rising columns of warm air, identifiable by puffy clouds and circling birds. Pilots gain maximum altitude by performing 'coring'—circling inside thermals in large, gentle patterns, as lift is strongest toward the center. Both techniques enable hours of flight under favorable conditions.

Pelagornis likely used dynamic soaring, a technique similar to modern albatrosses. Birds perform this maneuver by exploiting wind speed differences at different altitudes above ocean waves. They swoop down toward the ocean surface, then pull up at the last moment to gain momentum and catch faster-moving air currents higher up. This looping flight pattern allowed them to cover vast ocean distances while burning minimal energy, enabling travel across thousands of miles without much effort.

Dynamic soaring is a paragliding technique where pilots use rising air currents along terrain features to gain altitude and extend flight distance, requiring specific skills including finding the minimum sink rate hand position, maintaining proper harness posture by loading the wing on the terrain side, and understanding the dynamic bubble—a wave-like air mass that forms along the terrain where pilots can efficiently ride the airflow; pilots must also follow priority rules (terrain on the right has priority) and be aware of dangers such as sudden acceleration from dynamic thermals, always having a speed bar ready and knowing escape procedures like climbing high and turning to face the wind.

Thermal updrafts form when uneven heating of Earth's surface by the sun causes warm air to rise from sun-warmed areas (like bare earth, rocks, or mountain slopes), creating vertical air currents that birds of prey such as turkey vultures, eagles, and hawks use to soar and migrate with minimal wing-flapping; larger birds with greater wing surface area can ride these thermals for longer distances, while smaller raptors can utilize weaker thermal updrafts that larger birds cannot, meaning different species may encounter usable thermals at different times depending on weather conditions.

Raptors are diurnal migrants that use thermals (rising columns of warm air) for free lift during migration. Birds rise in altitude using thermals until activity dies out, then stream out in a southward or northward direction depending on season. Bodies of water like the Great Lakes act as barriers, forcing raptors to follow shorelines and creating funnel points where birds concentrate. The Detroit River area serves as an excellent example of such a funnel point, concentrating migrating raptors for observation.

Many birds of prey migrate south for winter, traveling along traditional routes. When flying long distances, they use thermals—columns of rising warm air—to gain height and rest, conserving energy. Thermals are scarce over sea, so migrants make for narrow crossing points. Autumn migrations include black storks from Poland, white storks from Germany, buzzards from Czechoslovakia, and sparrowhawks from Russia.

Migrating raptors concentrate at specific geographic features because these locations provide efficient migration corridors. In the intermountain West, birds avoid flying across the Great Salt Lake due to energy costs and instead follow north-south mountain ranges like the Wasatch and Ghost Shoot Mountains. When wind hits a north-south running ridge, it rises upward due to deflection, allowing birds to 'surf' along ridges efficiently. Raptors also use thermal air currents—hot air rises, creating upward currents that birds circle in (called 'kettling') before soaring out for miles. This combination of ridge line winds and thermals allows birds to travel long distances while conserving energy, making these geographic features ideal locations for monitoring.

Migration corridors are narrow pathways birds follow between major geographic barriers. European raptors migrating to Africa are forced to follow a corridor along the eastern Black Sea coast because the sea and Caucasus Mountains create barriers. The Black Sea is particularly challenging because it lacks strong thermal currents above the water surface. Thermals are upward air currents created when the sun warms the Earth's surface, causing warm air to rise. Birds like buzzards, hawks, and eagles use these thermals for energy-efficient soaring flight, spreading their wings to gain altitude without flapping. Birds navigate using visual cues from cumulus clouds and learn from experience which landscape types generate the best thermals. Migration efficiency is subject to natural selection—birds that learn more efficient routes complete migration faster and produce more offspring. Juvenile raptors must learn migration routes independently on their first migration, often flying directly across barriers initially.