Policy adjustment in crisis situations involves multidimensional changes in policy system elements.Previous studies have tended to focus on the overall relationship between crises and policy trends, neglecting to observe the fluctuations of policy system elements triggered by crises.This study draws on policy paradigm theory to extract observational elements, selecting environmental emergencies occurring between 2011 and 2020 for empirical research. Utilizing web crawlers to assemble a database of provincial ecological and environmental department policies, and employing supervised machine learning algorithms for automated coding, the study comparing the adjustments of policy system elements between normal and crisis periods based on statistical results. The results demonstrate that crisis events propel more substantial modifications in policy scale and instruments compared to normal conditions, with the magnitude of adjustments escalating in accordance with the severity of the crisis events. Nevertheless, policy objective adaptations predominantly manifest as dynamic fine-tuning of focal tasks, while crisis events do not incite policy objective adjustments surpassing conventional levels. The non-congruent shifts within the policy system can be encapsulated as "differentiated adaptation," underscoring the persistent tension and sequential metamorphosis between "change" and "stability" in policy systems under crisis circumstances. This offers a theoretical foundation for decoding the practicality of policies in effective risk management, and aids policymakers in focusing on balancing endogenous stability with exogenous conflicts, thereby promoting the maintenance of flexible adaptation and robust operation in policy implementation.