Hydroclimate-vegetation feedbacks drive Holocene temperature variability in Southeast Asia

The Holocene (last 11,700 years to pre-industrial) epoch offers a critical test bed for assessing natural temperature variability and benchmarking anthropogenic climate change. By leveraging TEX86 measurements (n=31) of speleothems from two caves ~400 km apart, we present the first 12,000-year quantitative regional temperature reconstruction for Mainland Southeast Asia (MSEA), a critical hydroclimatic region influenced by Indian and East Asian monsoon systems. Our speleothem TEX86 record reveals pronounced centennial- to millennial-scale temperature fluctuations throughout the Holocene, including an abrupt ~4 °C cooling during the mid-to-late Holocene transition. This contrasts with a suite of transient climate model simulations (CESM-1.2 3 Ma simulation, MPI-ESM andTRACE-21K-II), which show monotonic warming across the same interval, failing to reproduce internal variability. We then compared our proxy record with outputs from a suite of Global Climate Model equilibrium sensitivity simulations, in most of which dust–vegetation–albedo feedback in northern Africa is included, a feature missing in earlier models. Results show better agreement of thesemodels with the observed temperature changes in MSEA. We hypothesize that the Indian Ocean Dipole (IOD) might play an important role in driving the observed temperature variations, a notion supported by both sensitivity simulations and sea surface temperature proxy records. Cooling of the eastern Indian Ocean (i.e. positive IOD-state) may induce regional megadroughtconditions, potentially leading to widespread declines in C₃ vegetation over MSEA. These land surface changes likely enhanced surface albedo, and, in combination with reduced downwelling longwave radiation due to lower atmospheric water vapor, contributed to a net cooling effect. Our results underscore the need to improve the representation of hydroclimate–vegetation feedback in climate models to better capture regional climate dynamics and enhance model skill for both past climate reconstruction and future projections.