Geochemical evidence indicates that deserts are not the dominant source of loess in Central Asia

To what extent are deserts in Central Asia (CA) the dominant source of loess? In this paper, we compare geochemical data for loess deposits and desert sands across CA to address this question. Sr–Nd isotopic data indicate that the sources of loess deposits in northern and southern CA are different. While caution is advised, isotopic data do not strongly support a direct genetic link between deserts and loess for this region. Trace element geochemistry further elucidates this relationship. The source rocks of both loess and desert sands in CA are dominated by felsic components; however, significant differences in the spatial heterogeneity of their trace element compositions were observed using clustering analysis based on the Calinski–Harabasz and Bayesian information criteria, with four clusters identified for loess deposits, and only two clusters determined for desert sands. The difference in optimal cluster numbers challenge the validity of aeolian comminution and abrasion in deserts as sources of silt-sized particles. Principal Component Analysis shows that the <75/63 μm fractions of desert sands are geochemically distinct from the adjacent loess, questioning the deserts as transition zones for the CA loess formation. Therefore, combined with the relative spatial distributions of loess and deserts and the scarcity of silt-sized particles within the deserts, we propose that deserts are unlikely to be the main contributors to loess sedimentation in CA. Instead, alluvial plains and proximal debris fans along mountain foothills are more plausible provenances. Specifically, the northern Central Asian loess is sourced mainly from glacial-fluvial alluvium and frost-shattered debris of the Tianshan Mountains; the southern Central Asian loess derives predominantly from the Afghan foothills and the Panj–Amu Darya floodplains. Loess in the southern Tarim Basin appears to derive primarily from the Kunlun Mountains, a provenance assignment that would benefit from further testing. Although CA deserts are excluded as a dominant loess source based on multi-proxy evidence, uncertainties persist. Specifically, we emphasise the need for improved grain-size-controlled sampling and analysis across a wider spectrum of source regions. Integrating these expanded datasets within Bayesian unmixing models and identifying optimal grain-size windows will constitute a critical frontier for quantitative and nuanced reconstruction of loess provenance in CA.