Chlorophyll and leaf nitrogen content (LNC) are important indicators of crop growth. Real-time and accurate monitoring of Chlorophyll and LNC is helpful for field management and improvement of crop quality and yield. Currently, hyperspectral technology and empirical regression methods are widely used to construct crop biochemical parameter prediction models. However, there are still gaps in the research on predicting leaf biochemical parameters across periods. In this study, 120 leaf hyperspectral data, chlorophyll and LNC were obtained at five stages of early rice. Three different algorithms, namely, Partial Least Squares Regression (PLSR), Random Forest (RF), and Support Vector Regression (SVR) were used to establish inter-temporal prediction models for Chlorophyll and LNC in different growth stages. In addition, the intertemporal prediction model and the traditional non-intertemporal prediction model were evaluated. The results showed that the early rice chlorophyll inter-temporal prediction SVR model had the best prediction effect, with the inter-tillering prediction model having R² of 0.54, the inter-booting prediction model has R² of 0.36, the inter-heading prediction model has R² of 0.30, inter-grouting prediction model has R² of 0.55, and inter-mature prediction model has R² of 0.74. The intertemporal prediction model of early rice leaf LNC was poor, and the LNC cannot be intertemporally predicted. Compared with the non-intertemporal prediction model, although the accuracy of the intertemporal prediction model has decreased, it can effectively overcome the shortcomings of poor universality of the empirical model, which is beneficial to realize the physiological characteristics of different growth stages of crops in the same life cycle. At the same time, we found that crop physiological indicators have inter-temporal predictability. This concept provides new ideas for the prediction of crop phenotype, internal crop quality and yield.