Abstract:This paper first analyzed the sensitivity of extreme weather extended range forecasting to the random errors, initial errors, and model parameter errors in the established single-variable Nonlinear Cross Prediction Error (NCPE) pattern, as well as the sensitivity of different meteorological element fields to extended range forecasting. And then established a Multi-chaotic variables Nonlinear Cross Prediction Error (MNCPE) model based on precipitation, temperature, and geopotential height data, and conducts an in-depth comparative analysis of these two models. Results show that, the ratio order difference between random errors and initial errorshas a critical value effect on the extended range forecasting characteristics for NCPE model. While for parameter errors, an appropriate phase space embedding dimension m can better represent the local details of the attractor.The sensitivity of different meteorological element fields to extended range forecasting of diverse extreme weather, such as heavy rain and typhoons also varies.The MNCPE model, compared to the single-variable NCPE, can more comprehensively represent the dynamic features of the local structure of the chaotic attractor and reduce the uncertainty of the 10-30-day extended range forecasting.