Windblown dust emissions are subject to large uncertainties in Earth system models (ESMs), yet model discrepancies in dust variability and its physical drivers remain poorly understood. This study evaluates the consistency of 21 ESMs in simulating the climatological distribution and interannual variability of global dust emissions and applies dominance analysis to quantify the relative influence of near-surface wind speed and five hydroclimate variables (precipitation, soil moisture, specific humidity, air temperature, leaf area index) across different climate zones. In hyperarid regions, the models exhibit poor agreement in dust variability, with only 10 % of pairwise comparisons showing significant positive correlations. Most models capture the dominant wind control except GFDL-ESM4 which display dominant hydroclimate influence (wind contributing 42 %) and high spatial variability. In arid and semiarid regions, dust variability is shaped by a dual effect of land surface memory: models with consistent hydroclimate variability converge in dust responses, while those with divergent hydroclimate representations show increased disagreement. While all models capture the expected increase of hydroclimate influence with decreasing aridity, the extent of this transition varies by model, resulting in greater model disagreement regarding the relative importance of wind and hydroclimate drivers in arid/semiarid regions. Implementing the Kok et al. (2014) scheme in CESM reduces the wind contribution from 86 % to 64 % in hyperarid regions and from 56 % to 46 % in arid regions, indicating enhanced hydroclimate influence compared to the Zender et al. (2003) scheme. These findings underscore the importance of improving hydroclimate and land surface representations for reducing uncertainties in dust emission responses to climate variability and change.