Abstract: Combining dynamical and statistical methods is one of the important ways to improve weather and climate prediction. A key issue is how to effectively combine numerical model results with historical data. Combining the above two methods with an analog method is an important direction for future improvement of weather and climate prediction, although the analog method is still limited to similarity assumptions and lacks solid physical basis. Based on the initial condition problem of a perfect model, this work proposes the concept of initial condition perturbation of the perfect model and develops a Dynamical-Statistical-Analog Ensemble Forecast (DSAEF) theory. The DSAEF theory shows not only why the analogue-based forecast can be conducted, but also how it can be conducted. That is, the perfect model is used to produce forecasts and an ensemble prediction scheme is used to achieve the forecast accuracy. Based on the DSAEF theory, the DSAEF_LTP (Landfalling Typhoon Precipitation, LTP) model has been developed. This model includes the following four steps: (Ⅰ) forecast typhoon track, (Ⅱ) Construct generalized Initial Value (GIV), (Ⅲ) identify analogs from historical observations, and (Ⅳ) produce an ensemble forecast of typhoon precipitation. The GIV is constructed by physical variables that affect LTP. The DSAEF_ LTP model has the characteristic of sustainable development, which can be improved by introducing new variables or refining the existing parameters of the model. At present, the model versions 1.0 and 1.1 have been released. In version 1.0, GIVs include three physical variables, i.e., typhoon track, landfall season and typhoon intensity. In the version 1.1, two extra improved parameters of "similarity region" and "ensemble scheme" are added. The most recent version 1.1 has shown significant improvements in the performance for forecasting LTP. Three large-sample forecast experiments using version 1.1 show that, compared ECMWF, CMA-GFS, NCEP-GFS and SMS-WARMS (Shanghai regional model), version 1.1 performs best in forecasting accumulated rainfall greater than 250 mm and 100 mm. Of course, there is a considerable room for improvement in the DSAEF_LTP model forecast performance, which can be achieved by incorporating more physical variables that affect LTP into the GIV of the model. This implies that the DSAEF_LTP model will have broad prospects for future development.