Crustal thickness and elevation variations control mountain building and climate change at convergent margins. As an archetypal Andean-type convergent margin, eastern Asia preserves voluminous subduction-related magmas ideal for quantifying these processes and their impacts on climate. Here we use Sr/Y and Ce/Y proxies to show that the crust experienced alternating thickening and thinning episodes during the Late Mesozoic. We identify a noticeably thickened (50–55 km) crust associated with tectonic shortening at 120-105 Ma, corresponding to the emergence of a > 2500-m-high coastal mountain range. Using climate modeling, we demonstrate that the mountain uplift changed Asian atmospheric circulation and precipitation patterns, increased inland aridity (~ 15 %), and prompted the eastward desert expansion, contributing significantly to the arid zonal belt across mid- to low-latitude Asia. These findings, compatible with independent geological, geophysical, and climatic observations, have global implications for broadening our understanding of Earth-system interactions in the Cretaceous greenhouse world.

Sr/Y and Ce/Y proxies  Crustal thickness controls the Sr/Y variability of arc magmas by affecting the stabilization of mineral phases, which fractionates Sr (plagioclase) and Y (garnet) (14,21). Sr is compatible and partitions into plagioclase at low pressures (<10 kbar); at high pressures (>12 kbar), Sr becomes incompatible and enters the liquid phase, and plagioclase is less abundant (64). By contrast, Y is incompatible at low pressures and tends to partition into garnet and amphibole at high pressures (19). Consequently, Sr/Y is a qualitative indicator of the pressures (i.e., crustal depths) at which partial melting and crystal fractionation occur (22,23). Chapman et al. (21) suggest that the Sr/Y ratios of intermediate magmas vary with crustal thickness in the North American Cordillera, with larger Sr/Y ratios signifying greater pressures and, hence, crustal thicknesses. Based on large data sets of global arc lavas, Profeta et al. (19) propose Sr/Y and La/Yb empirical correlations that are feasible in monitoring the crustal thickness evolution of ancient convergent margins. In addition to intermediate magmas, crustal thickness controls the composition of arc basalts by modulating the degree of mantle melting (19). Mantle and Collins (20) quantitatively estimate the changes in the crustal thickness of the New Zealand orogen since 400 Ma using maximum Ce/Y ratios of arc basalts. The Sr/Y and Ce/Y empirical correlations have been corroborated and formulated by examining the calculated crustal thicknesses with the geophysically derived Moho depths (20,21,24).

Climate Modeling Here we use a fully coupled atmosphere-ocean general circulation model, i.e., the Community Earth System Model version 1.2.2 (CESM1.2.2) developed by the National Center for Atmospheric Research (NCAR). CESM1.2.2 includes dynamic atmosphere, ocean, land, sea ice, and runoff modules (Fig. S2a-b), which interact with each other through a coupler (65).

  To explore the influence of the East Asian coastal mountains on the Asian climate in the mid-Cretaceous, two experiments, i.e., noCM (Fig. S2c) and CM (Fig. S2d), are conducted using the global topography compiled by Sewall et al. (66). In experiment noCM, the East Asian coastal mountain range does not exist (Fig. S2c). In comparison, a north-south oriented mountain at ~22-38 °N along the East Asian margin is included in experiment CM (Fig 6d). Accordingly, the differences in the experiment directly reveal the effects of the coastal mountains on the Asian climate. Given the zonally limited East Asian coastal mountains, we use a relatively high resolution of 0.9° × 1.25° for the atmosphere (CAM4.0) and land (CLM4.0) modules in the experiments. Notably, fully coupled running with a dynamic ocean at such a high resolution is forbidden due to limited computing resources. The two experiments are driven by fixed sea surface temperatures (SSTs) and sea ice with annual cycling, which are the monthly climatology of the third experiment (CPL). This experiment is a fully coupled run of the CESM1.2.2, in which the CAM4.0 and CLM4.0 are modeled at a resolution of 3.75° × 3.75°. The ocean (POP2) and sea ice (CICE4) modules employ a gx3v7 grid possessing 116 and 100 grid points in the meridional and zonal directions, respectively. The CAM4.0 and POP2 have 26 and 60 vertical levels, respectively. The runoff module runs at a resolution of 0.5° × 0.5°. The abovementioned methods are commonly used to model the topographical impacts on the regional climate (67-69). In this study, the experiment CPL runs for 3000 model years, which is long enough for the Earth's surface to reach a quasi-equilibrium state. The monthly climatology of the last 100-year SSTs and sea ice are used to force the experiments noCM and CM. These two uncoupled experiments run for 40 model years, and the data of the last 30 model years are analyzed here.

The other settings in the three experiments are concordant. The used solar constant is 1350.87 W/m², 0.75% lower than the present value of ~1361 W/m² (70). We set the CFC, pCO₂, pCH_4, and pN₂O concentrations to 0, 1120 ppmv, 760 ppbv, and 270 ppbv, respectively. The influence of orbital forcing is not considered. We use the orbital parameters of 1990 CE in the modeling experiments. The default values of the CESM1.2.2 were used for all other atmospheric components (e.g., O₃ and aerosols), all at preindustrial levels.