Abstract: In order to better improve dynamical seasonal prediction by utilizing statistical experiences, the problem of capturing statistical experiences associated with climate model is discussed from the point of view of studying the impacts of physical predictors in climate system on model prediction errors. As model errors objectively exist in climate model and vary with the sate of climate system, the prediction errors of model change with state. Thus, deeply examining the characteristics that how the prediction errors of model are influenced by climate system state has important values to capture the statistical experiences associated with model and improve dynamical prediction in terms of the combination of dynamical and statistical methods. Problems and their meanings related with above characteristics are primarily proposed.Based on hindcast data of air-ocean coupled general circulation model in National Climate Center, China Meteorological Administration, the relationship between the prediction errors of summer mean circulation and total precipitation in dynamical seasonal prediction, and some primary physical predictors are comprehensively examined. These physical predictors selected include the sea surface temperature index in Niño3 region, the Pacific decadal oscillation index, the southern hemispheric annular mode index, the northern hemispheric annular mode index, and the North Atlantic oscillation index, which reflects the prevailing modes of interannual variability in climate system over tropics and extratropics.Results of correlation analyses show that there are some significant early or simultaneous relationships between the above-mentioned physical predictors and the model prediction errors of summer circulation and precipitation. In the five selected physical predictors, the sea surface temperature index in Niño3 region mainly correlates with the prediction errors of low-latitude circulation and precipitation. The correlationship between the Pacific decadal oscillation index and prediction errors is mostly characterized by simultaneous correlations and is significant not only over some low-latitude but also mid-and high-latitude areas. The southern and northern hemispheric annular mode indices standing for the leading modes over two extratropics are not corresponding to the evident early correlation with only some significant correlations over partial extratropics, whereas the simultaneous correlations are relatively evident and characterized by the distribution patterns similar to spatial modes of the two extratropical indices. The correlation situations between the North Atlantic oscillation index and circulation errors are very analogical to those corresponding to the northern hemispheric annular mode index, whereas the early and simultaneous correlations between it and precipitation errors are much better than those corresponding to circulation errors.By examining the physical processes that physical predictor influences the distribution of model prediction errors, investigating the relationship between predictor and prediction errors will not only help to evaluate the performance of model and provide reference for improving model, but also benefit the development of the new prediction strategy and methodology for correcting prediction errors. Thereby, physical basis and practical reference are provided to develop new method of predictor-based error correction, which will be studied and verified in further work.