Abstract: [Purposes] Monitoring land resource change patches based on satellite imagery is an effective measure for land resource protection.However,when predicting the area and quantity of rapidly fluctuating land resource change patches,the accuracy and stability of the existing time series prediction methods are not satisfactory. [Methods] In this paper,the prediction method for land resource change patches driven by large language models and cycle time prompts (LLM-CTP)is proposed.The LLM-CTP is the first method to apply large language models to predict land resource change patches.Meanwhile,by introducing cycle time prompts to each value in the time series,the LLM-CTP method can obtain higher prediction accuracy. [Findings] In predicting the area and quantity of land resource change patches,the comparative experiment results show that the average absolute percentage errors of the LLM-CTP method are 19.46% and 8.69% respectively,which are lower than LLMTime method (30.54%and 17.91%)and LSTM method (22.61% and 26.25%). [Conclusions] The relevant researches in this paper laid the foundation for timely warning and preventing the land resource damage from illegal activities,as well as ensuring the responsibilities of land resource protection.

Key words: land resource protection, land resource change patches prediction, large language models, cycle time prompts