本研究利用 HYDRO_GEN 繁衍法模擬隨變域，針對繁衍隨機變域之理論基礎作探討 ，並驗證該方法的正確性。對於應用方向，我們依據區域化變數理論提出以隨機變域繁衍評 估前人研究中以現存之觀測站與觀測值所計算之半變異元的正確性。針對過去淡水河流域雨 量站網設計中所計算出的半變異元，本研究中模擬了淡水河流域時雨量的隨機變域，結果發 現若是以均勻分佈的 26 站， 或是 32 個雨量站的位置則以 180 次的表現值均可正確的推 估時雨量的空間變異特性，因此前人所計算的半變異元應具有一定的可信度。再針對雨量站 網設計之結果研究， 則若是將前人研究中建議可優先捨棄的 8 個站去除，再計算半變異之 結果極為相似，因此該 8 站的資料可替代性確實很高。而若是以研究中建議的 7 個基本站 則無法正確的計算時雨量之半變異元。 若是模擬影響範圍與年雨量相當之隨機變域， 以 7 個基本站配合 150 次表現值便可正確地計算半變異元， 同時以 32 站或 24 站所計算之結 果亦是十分類似，因此顯示該 8 個可優先捨棄之測站的雨量資料確有相當高的可替代性。
Spatial estimation using method of geostatistics, also known as kriging, has spreaded into many diversified research areas in the last two decades. Variogram modeling is often based on samples of one or only a few realizations. Thus, it is difficult to verify whether the estimated variogram adequately characterize the underlying natural processes. In this study, a random field generation model, HYDRO_GEN, is utilized to generate many realizations of a random filed with given statistical distribution and variogram. Using technique of random field generation, the effect of number of evenly distributed raingauges and number of realizations on variogram modeling can be evaluated. A previous study suggested that in a network of 32 raingauge stations in Tan-Sui River Basin, only twenty-four of them are needed to account for almost the same amount of information. Among the twenty-four stations, seven are basic stations that account for the trendof spatial rainfall variation. We also verify that rainfall data of the twenty-four stations which are not evenly distributed are capable of characterizing the variogram of hourly rainfall, whereas rainfall data of the seven basic stations represent the spatial variation of annual rainfall, as was suggested in the previous study.
