Abstract

Climate model simulations have improved their skills in reconstructing some of the major climatic and environmental transformations since the Last Glacial Maximum (LGM). However, their validation against proxy data is still a critical step for assessing potential biases and misrepresentations. Here we present a detailed analysis of the CHELSA-TraCE21k dataset (Karger et al., 2023), a statistical downscaling of the TraCE21k global transient simulation, focusing on the South American Altiplano (14-22 degrees S). We validated the model skill in reconstructing temperatures and precipitation in this high-Andean environment using both present-day and proxy data. We then discuss the simulation in the light of regional reconstructions to provide an updated view of the paleoclimate evolution in the Altiplano over the last 21,000 years. The simulation shows an average LGM temperature decrease of 3.4 degrees C with a slight precipitation increase, and reveals two humid phases between 17.5-15 and 12.8-11.7 ka BP. Holocene conditions were relatively warm and dry, with slow but persistent warming and drying between 11 and 5 ka BP. The CHELSA-TraCE21k dataset shows an overall good agreement with proxy records, indicating that the small-scale orography effects included in the downscaling algorithm improve the representation of past surface conditions relative to original global simulation. There are differences between the model and proxy data in the extent of paleo-glaciers and the amplitude of temperature and precipitation changes. The simulation reveals that the climate history of the Altiplano during the last deglaciation was forced by changes in austral summer insolation, superimposed by millennial-scale climate events in the Northern Hemisphere. The spatial distribution of hydroclimate anomalies suggests that the formation of the large paleolakes was aided by the overflow of Lago Titicaca. These findings contribute to a better understanding of the Altiplano's paleoclimate evolution and highlight the value of integrating both proxy and modelling data.