設計暴雨為某集水區過去之雨量紀錄利用降雨強度一延時一頻率分析所推求未來於特定重現期距與延時之暴雨量。而受限於雨量紀錄的長短，經過分析所推求之設計暴雨量仍然存在不確定性。本研究利用臺大實驗林轄區鄰近59測站之資料，利用地理統計法與HYDRO_GEN程式，進行不同重現期距延時設計暴雨及年月雨量之序率模擬，並推求部份重現期距延時設計暴雨超越豪大雨門檻值之空間機率分布。研究結果顯示隨延時及重現期距之增大，其降雨量增加，其變方與均值亦增加。半變異元參數亦有此趨勢，但距離參數與延時及重現期距之變化不明顯；以24 hr 5年重現期距測試超過超大豪雨門檻值之機率分佈，隨模擬次數之增加，可發現其分佈之等值線趨於平滑與穩定；年月雨量模擬1,000次結果年雨量與原年均值等值線差異不大，而月雨量較小之月份呈現較明顯之差異性。本研究說明即使以既有長期之雨量紀錄為基礎之前題下，以序率模擬進行雨量之空間推估仍可能存在不同之差異性，顯示雨量在不同時間尺度下之空間分佈以長期而言仍有相當之不確定性，本研究之分析結果未來可作為林地經營之重要之參考。

Design storm is the rainfall under certain return period and duration estimated by Intensity-Duration-Frequency analysis using the past rainfall records. There still exists uncertainty for design storm due to the lengths of record of rainfall. This study aims at stochastic simulation of design storm at the National Taiwan University Experimental Forest with different combinations of return periods and durations, annual and monthly rainfall by geostatistics and HYDRO_GEN program based on 59 rain stations. Part of the return periods and durations are also applied for calculation of exceeding probability the extremely torrential rainfall. The results indicated that the variance and average increased as the return period and duration and rainfall increased. The contour of exceeding probability for extremely torrential rainfall became more stable and smooth as the number of simulation increased. On the other hand, there was no large difference in contour line between the average annual rainfall and the average of 1,000 simulations, but significant difference was found for monthly rainfall of small amount. It can be referred that by stochastic simulation for spatial estimation, there still exists uncertainty for the rainfall at different temporal scales in the study area, even based on long-term rainfall records. The results can provide as key reference for future forestland management.