柳杉88個種源與杉木42個種源造林木之樹高及胸徑經量測及分析後，取其標準得分點數構成一個生長合成指數。空間變異係藉巾二次迴歸模式及變異曲線譜模式而予數量化。經巾迴歸分析顯示柳杉生長變異可藉種源所在地之經緯度而加以預測，然杉木則否。經由變異曲線譜之統計分析亦看出兩樹種問之顯著差異，杉木比柳杉有較高的塊聚值，但其較低之限界及變域值。就模式精密度而言，杉木比柳杉有較大的機差變方，但具較小之R乘方。通常一個1°或0.5°的留滯寬度對模式介量無任何效應，但能影響變異曲線譜模式之精密度。距離幅度（留滯寬之個數）能影響限界及變域值之推估，但對塊聚值則否。模式公式的選擇能影響全部3個介量及模式的R乘方，但無法影響殘餘平方和，就高氏模式而言，如有一距離幅度含9個留滯寬以及有1°之間隔，則可成爲樹種之最好模式。杉木之各向同性變域不大(<2°)，而且其不同變異曲線譜也無法顯示任何明確之趨向；相反地，柳杉之各向同性變域較大(>18°)，而且其不同變異曲線譜也顯示出較明確之由東而西，以及由東北而西南的趨向。在此情況下，我們可描繪出柳杉種源之生長型式是屬連續漸變型變異，而杉木種源之生長型式屬區域局部犁的逢機變異。至於其他樹種之半變異模式建立方面，我們建議一個留滯寬度大小應含70對以上之數據，並且距離幅度或留滯寬個數應火於20。
Tree height and diameter were measured from 88 provenances of Cryptomeria and from 42 provenances of China fir. A composite index of growth was constructed by taking the average of the standard scores. Spatial variation was quantified using a quadratic regression model and a variogram model. Regression analysis shows that growth of Cryptomeria, but not of China fir, can be predicted by the latitude and longitude of the provenance. Statistical analysis using the variogram also shows significant differences between the two species. China fir has a higher value for the nugget, but smaller values for the sill and range than does Cryptomeria. In terms of model precision, China fir has a greater error variance, but a smaller R^2 than does Cryptomeria. Using a lag of 1.0° or 0.5 has no effects on the model parameters, but it affects the precision of the variogram model. The distance span (number of lags) affects estimates of the sill and range but not the nugget. The choice of model formula affects all 3 parameters and the R^2 of the model but not the residual sum of squares. The Gaussian model, with a span of 9 lags and a lag width of 1, seems to be the best model for the 2 species. For China fir, the isotropic range is short (<2°) and the aniso-variogram shows no significant trends in any direction. On the contrary, the isotropic range is long (>18°) for Cryptomeria, and the aniso-variogram shows significant E-W and NE-SW trends. Therefore, we may describe the growth pattern in Cryptomeria as clinal variation, while localized random variation may be more suitable to define the China fir provenance growth. In modeling the semivariance of other species, we suggest that the size of the lag should include more than 70 data pairs, and the number of lags should exceed 20.
