SITA: Machine Learning Climb Optimization Saved 59,011 Tonnes of Fuel Across 1,137 A320 Aircraft in 2025
Why It MattersThe approach signals a shift toward per-aircraft, continuously retrained fuel-optimization models that could displace generic cost-index methods across airline fleets.
A single departure can be flown in up to two million different ways depending on four climb parameters: initial indicated airspeed, acceleration flight level, second indicated airspeed, and Climb Mach. SITA's climb optimization system, built on machine learning models trained on each aircraft's quick access recorder data, saved 59,011 tonnes of fuel across 1,137 Airbus A320 family aircraft in 2025, averaging 73 kg per departure.
The system addresses a limitation of the Cost Index method, which was designed for cruise speed economics and optimizes only two speed variables, whereas the climb problem's four interacting parameters create up to two million candidate profiles per flight. SITA's proprietary optimization engine evaluates fewer than 1,000 candidate profiles per flight to identify near-optimal results within minutes of departure.
Generic manufacturer performance models assume a standard aircraft in controlled conditions and cannot account for engine and airframe degradation or the variable behavior captured in over 40 quick access recorder parameters logged on every flight. SITA's models are trained on each aircraft's own quick access recorder history and retrained every three to four months to reflect ongoing performance drift, and the same model verifies fuel savings on a per-flight basis by predicting fuel burn for both the optimized and reference climb profiles over an identical ground distance.
