Projections of the Asian–Australian, African, and American monsoons are currently challenged by considerable levels
of uncertainty, which influences the effectiveness of climate change adaptation strategies. Clarifying the uncertainty
sources is essential to reduce this uncertainty. Most previous studies have addressed this issue based on limited members in
individual models, which cannot strictly isolate the forced model response from the internal variability. Here, we first
employ the latest multi-model large ensemble (MMLE), with a total of 550 members from eight models, under very-high
emission scenarios. The results show that model uncertainty (internal variability) increases (decreases) with time for all
monsoon regions, but with notably regional disparities in their relative contributions. On the grid scale, internal variability
dominates the total uncertainty of summer precipitation changes during the near-term (2020–39) and mid-term (2040–59)
periods in most monsoon regions. For monsoon circulation, internal variability exerts an even greater influence over the
Asian–Australian monsoon region. Compared with the MMLE results, a conventional approach to isolate the forced signal
based on polynomial fitting tends to underestimate the fraction of internal variability, particularly when and where that
fraction is large. Consequently, the conventional approach overestimates the forced signal of monsoon precipitation relative
to internal noise, leading to an earlier time of emergence by about 10 years compared with that derived from the MMLE,
which is before 2050 for most monsoon regions. The results highlight the necessity of using MMLEs to quantify sources of
uncertainty in climate projections, providing important implications for improving the robustness of future climate
assessments.

global monsoon precipitation, projection uncertainty, internal variability, model uncertainty, time of emergence, large ensemble