Abstract: Drought is a natural disaster caused by prolonged imbalance in water budgets, leading to a shortage of available water. It is crucial to accurately predict future droughts for food security. Previous studies have shown that Global Climate Models (GCMs) still exhibit uncertainties in simulating climate variables, which may further hinder accurate predictions of extreme events such as droughts. In order to accurately understand how the length of dry spell will change in the north China under climate warming, the potential application of emergent constraint (EC) methods for predicting the longest annual consecutive dry days (LAD) was investigated in this study. A physically-based constraint method, namely the hierarchical emergent constraint (HEC), was applied to establish a linear relationship between historical LAD and the future change magnitude of LAD under the two forcing scenarios of Coupled Model Intercomparison Project 5 (CMIP5) and 6 (CMIP6), respectively. The period from 1998 to 2018 is considered a historical period, and the period from 2080 to 2100 is considered a future period. Subsequently, the established relationships were constrained using seven high-performance observational LAD datasets the expected values and uncertainty ranges of future LAD changes. The results show that the observed LAD in the north China followed an aridity gradient, ranging from 32 days in the semi-humid northeast to 232 days in the arid northwest. Models in CMIP5 and CMIP6 underestimated the historical LAD, with greater uncertainty compared to observational data. Over 70% of the models showed a reduction in future LAD across the north China compared to the 1998-2018 period, with more severe emissions leading to larger reductions. In both the CMIP5 and CMIP6 model ensembles, a significant negative correlation was observed between the future change magnitude of LAD and the historical LAD. Under the medium forcing scenario, the correlations for CMIP5 and CMIP6 were -0.5 and -0.46, respectively, and under the high forcing scenario, the correlations are -0.46 and -0.54, respectively. Overall, the CMIP6 models overestimated the future LAD, after applying the HEC, the future LAD decreased by 4.97 days under the SSP2-4.5 scenario and 9.15 days under the SSP5-8.5 scenario. Besides, the uncertainty range of the future LAD is reduced by 8.7% under the SSP2-4.5 scenario and by 12.4% under the SSP5-8.5 scenario. At the grid scale, compared to unconstrained results, the LAD was underestimated by at least 9 days under the SSP5-8.5 scenario. Over time, the LAD in the north China gradually decreased, and the degree of reduction of future LAD and the uncertainty range after constraint is even greater. After applying the HEC, under the SSP5-8.5 scenario, the reduction in future LAD change will be 7.54 days by 2040 and 10.33 days by 2090. Besides, the uncertainty range decreased by 7% by 2040 and by 10.3% by 2090. In the northeastern region, the change in future LAD before and after applying constraints was relatively small. For example, in Jilin Province, the reduction in future LAD changed from 3.99 before constraints to 5.64 after applying HEC. In contrast, the most significant changes occurred in Ningxia Hui Autonomous Region and Qinghai Province, where the difference in future LAD change before and after constraints exceeded at least 10 days. Robustness tests were conducted on the scope of the chosen periods and the selected observational data products. The absolute values of correlation coefficients were always greater than 0.4, indicating that the HEC method can be effectively applied to future projections of the LAD in the north China. The results can provide a scientific reference for formulating agricultural policies to address drought, optimizing crop planting structures, and developing rational irrigation plans.