Winds Are Fading Over the Tibetan Plateau and Climate Models Keep Missing It

High on the Tibetan Plateau, the wind is quietly dying. A new study published in the journal Climate Dynamics confirms that near-surface wind speeds across the region, often called the roof of the world, have declined sharply over recent decades, and that the world’s most advanced climate models are failing to capture just how fast the calm is setting in. The research, led by Qinglong You of Fudan University together with colleagues from institutions across China and Qatar, combines decades of ground observations with a large suite of high-resolution climate simulations to evaluate how well models reproduce the so-called stilling phenomenon and to project what the future may hold for winds above 4,000 meters.

The observational record is striking. Drawing on measurements from 150 meteorological stations operated by the National Meteorological Information Center of the China Meteorological Administration, the team found that annual mean near-surface wind speed over the Tibetan Plateau declined at a rate of 0.159 meters per second per decade between 1979 and 2014. The decline was not spread evenly through the year. Spring showed the steepest weakening, with wind speeds falling by 0.235 meters per second per decade, a seasonal signal that matters because spring winds play a central role in the plateau’s exchange of heat and moisture with the surrounding atmosphere and in shaping the Asian monsoon system downstream.

To test whether climate models could reproduce this stilling, the researchers turned to the Coupled Model Intercomparison Project Phase 6, specifically its High-Resolution Model Intercomparison Project, known as HighResMIP. This experiment was designed in part to answer a long-standing question in climate science: would simply cranking up the horizontal resolution of global models improve their simulations? The team assembled 22 historical simulations from 11 model pairs, each pair consisting of a higher-resolution and a lower-resolution version of the same model, and compared them against the station observations over the 1979 to 2014 period.

The verdict on resolution was sobering. Both the higher-resolution and lower-resolution ensemble means captured the broad spatial pattern of wind speed across the plateau, but both badly underestimated the magnitude of the decline. The higher-resolution ensemble produced an annual trend of only 0.024 meters per second per decade, while the lower-resolution ensemble managed just 0.020 meters per second per decade. In other words, the models reproduced only about one-seventh of the observed stilling. Perhaps more telling, the difference between the two resolution tiers was small, suggesting that increasing horizontal resolution alone offers limited improvement in capturing the historical wind decline over this complex, high-altitude terrain.

This widespread stilling bias is not merely an academic curiosity. Near-surface wind speed governs a host of processes with real-world consequences: the evaporation of water from land surfaces, the dispersal of air pollutants, the mixing of the atmospheric boundary layer, and the output of wind turbines. The Tibetan Plateau is also the source region of major Asian rivers and a critical driver of monsoon circulation, so systematic errors in simulating its winds can propagate into broader failures in regional climate prediction. Previous studies have linked wind declines across China to factors ranging from urbanization and increased surface roughness to weakening pressure gradients and changes in large-scale circulation, but representing these processes faithfully in global models has remained elusive.

To dig into why the models fall short, the team applied a random forest machine learning analysis, a technique that can rank the relative importance of many candidate variables in explaining an outcome. The analysis pointed to biases in key dynamical and thermodynamical processes, with snow variability and surface sensible heat flux emerging as particularly influential factors. This makes physical sense. Snow cover on the plateau alters surface albedo and the partitioning of energy between sensible and latent heat, which in turn modulates the temperature contrasts that drive near-surface winds. If a model misrepresents how snow accumulates and melts across the plateau’s rugged topography, the resulting errors in surface heating can distort the very circulation features that generate wind.

Having identified which simulations performed best against observations using multiple statistical metrics, the researchers selected seven optimal models and constructed an optimal ensemble mean for future projections. Under the high-emission Shared Socioeconomic Pathway 5-8.5, the projections indicate that the stilling of the Tibetan Plateau will continue through the coming decades. Between 2015 and 2049, the ensemble projects an annual wind speed decline of approximately 0.02 meters per second per decade, unfolding alongside continued regional warming. While the projected rate is gentler than the historical observed trend, the direction of change is consistent, lending robustness to the conclusion that winds over the plateau will keep weakening in a warming world.

The juxtaposition of a robust future decline with a poorly simulated past decline creates something of a paradox for climate scientists. If models underestimate how much wind has already slowed, should we trust their projections of future slowing? The authors argue that the consistency of the projected decline across the selected optimal models, combined with the physical plausibility of continued warming-driven circulation changes, supports confidence in the sign of the projection even if the magnitude remains uncertain. At the same time, the study makes clear that closing the gap between simulated and observed trends requires more than computational horsepower.

That message may be the study’s most consequential takeaway for the modeling community. HighResMIP was conceived on the premise that finer grids would better resolve steep orography, coastlines, and mesoscale processes, and for some variables that promise has held. But for near-surface wind over the Tibetan Plateau, the near-identical performance of high- and low-resolution model pairs indicates that the bottleneck lies elsewhere, in the representation of physical processes such as snow-atmosphere interactions, surface energy budgets, and boundary-layer dynamics. Improving these process representations, the study suggests, may matter more than simply increasing spatial resolution for reducing persistent model biases in wind simulation.

For the people and ecosystems of the plateau, the implications extend well beyond model evaluation. Weakening winds influence glacier mass balance through altered turbulent heat exchange, affect the timing and intensity of the monsoon onset that billions of people depend on, and reshape the viability of wind energy development in one of the world’s most promising but fragile high-altitude environments. As the Third Pole continues to warm faster than the global average, the slow, steady quieting of its winds serves as another reminder that climate change reshapes even the most fundamental features of the atmosphere, and that our models, for all their sophistication, are still learning to listen.

Subject of Research: Evaluation and projection of near-surface wind speed trends over the Tibetan Plateau using CMIP6 HighResMIP simulations and station observations

Article Title: Near-surface wind speed over the Tibetan Plateau: evaluation, trend and projection

Article References: You, Q., Wu, F., Wu, T., Cai, Z., Sun, G., Jin, Z., Jiang, Z., Ullah, S., & Li, M. (2026). Near-surface wind speed over the Tibetan Plateau: evaluation, trend and projection. Climate Dynamics, 64(10), Article 411. https://doi.org/10.1007/s00382-026-08342-6

Image Credits: AI Generated

DOI: 10.1007/s00382-026-08342-6

Keywords: Tibetan Plateau, near-surface wind speed, global stilling, CMIP6, HighResMIP, climate modeling, Climate Dynamics, SSP5-8.5, random forest, snow cover, sensible heat flux, monsoon