A spatial-temporal projection method (STPM) is developed to predict the spring (March-May, MAM) rainfall in northern Taiwan. Seven large-scale atmospheric and oceanic fields (925-hPa zonal wind, meridional wind, and moisture, 850-hPa, 500-hPa, and 200-hPa geopotential height, and sea surface temperature) with their temporal evolutions during the preceding 11 months are used as predictors. An optimal ensemble (OE) strategy is developed based on the best cross-validation results from each predictor over the training period. Some predictors adopted in the OE show the longest lead time of 10-month. The deterministic forecast result based on the OE approach indicates that the STPM predictions are skillful with an averaged temporal correlation coefficient of 0.6. However, the amplitude of the forecasted rainfall is underestimated, which is treated by introducing an amplifier coefficient. The STPM is also skillful for the probabilistic prediction of spring rainfall in northern Taiwan. The averaged Brier skill score reaches 0.37 for the below-normal categorical case.