The global carbon-cycle is crucial for climate change. Desert, which has long been neglected in the global carbon-cycle, may sequester enormous volumes of CO₂ and play the role of a carbon-sink. As the world's second-largest shifting desert, the Taklimakan Desert (TD) contributes substantially to desert carbon-sinks. However, the contributions of the internal processes of the TD to its carbon-sink and the long-term trend of the carbon-sink under climate change are still unclear. This study will address this important knowledge gap. Through field observations, we found that the expansion/contraction of soil air containing CO₂ caused by heat fluctuation in shifting sand, in combination with salts/alkali chemistry dominates the release/absorption processes of CO₂ in shifting sand. The mutual counteraction of these processes means that the TD shifting sand acts as a stable carbon-sink that had a CO₂ annual uptake of 1.60×10⁶ t·a^-1 during 2004–2017. It suggests that global shifting deserts maybe uptake of ~2.125×10⁸ t of CO₂ per year. However, an increasing soil temperature-difference will stimulate soil air expansion of desert and release more CO₂ into the atmosphere under climate change, causing the shifting sand carbon-sink decrease in the TD gradually in the future. These processes will be accelerated by positive feedback effect under climate change and enhance regional warming. These conclusions are very important for re-recognizing the status of deserts in the carbon-cycle, narrowing the gap in the missing carbon-sink and assessing the global carbon-cycle.

Through experiments of dismantling and temperature-controlled, we found that the expansion/contraction of soil air containing CO₂ caused by heat fluctuation, in combination with salts/alkali chemistry dominates the release/absorption processes of CO₂ in shifting sand. Finally, combined with historical soil temperature data and the Fifth Coupled Model Intercomparison Project (CMIP5) simulations, we reveal a stable carbon-sink property of the TD shifting sand.

The data file contains all the data of the corresponding manuscript, and there is no missing value.