- 期刊
鄰點數量與分布型態對一般克利金模式雨量估值誤差的影響之研究
Effects of Neighboring Points and Distribution Shape of the Ordinary Kriging Model in Estimating the Rainfall Errors
摘要
森林對降水的淨涵養量是規劃水源涵養林與林地分級以供推動森林生態系經營的重要因子，而森林區域內的降水淨涵養量空間資訊，可以藉由有限的林區降雨量觀測資料利用空間資料內插的技術推導之。為落實嘉義林區所轄森林生態系的林地分級，本研究特利用涵蓋全林區的52個雨量觀測站的雨量資料為材料，探討利用一般克利金方法繪製林區降雨量圖時，有關空間自相關與方向相依性等特性對建構半變異圖試驗模型的效應，並探討鄰點數量與鄰點的空間分布型態對試驗模型推估雨量準確度之影響。研究結果顯示，本試驗區的雨量分布同時存在著空間自相關與方向相依性兩種特性，所建構的一般克利金Circular、Spherical、Exponential以及Gaussian等半變異圖試驗模型的主軸方向均為「北北東-南南西」的結構，在各種試驗模型中以Anisotropic circular model的推估準確度最高，其次為anisotropic spherical model，二者皆能在以預測點為中心所區劃的八個方向中各選取1個最接近的已知點為鄰點，利用該8個鄰點資料值推估所得的雨量值最準，全區整體的雨量估測誤差最低。
並列摘要
Net holdback water, NHW, is an important factor used for delineating water conservation area and forestland classification, which is an elementary task for sustainable management of forest ecosystem. To achieve this application, it is required to incorporate the rainfall of the rainfall stations to derive the spatial quantities for accounting the NHW in watersheds. Recently, a geostatistics method, ordinary kriging technique was proven valuable in mapping those geo-spatially random variables, especially for the estimation of rainfall and soil pollution. Whereas prediction accuracy of the ordinary kriging model depends on the interaction between the types of semivariogram model and directional dependence of spatial autocorrelation; and also relies on the number of neighbors (search size) and the direction for collecting those neighbors (neighborhood shape). This paper used rainfall data from 52 rainfall stations located around the district of Chiayi National Forest to explore the interaction effect of those factors mentioned above and find out the suitable parameters for constructing a suitable ordinary kriging model. This model was then used for mapping the rainfall of the forest land in Chiayi area. Rainfall of the site of Chiayi National Forest was distributed with spatially autocorrelation and also showed a “NNE-SSW” directional dependence. The combination of 4 types of semivariogram model (circular, spherical, gaussian and exponential) and 2 types of directional dependence (isotropic and anisotropic) reaches to 8 types of ordinary kriging models. Among those models, the anisotropic circular model with 8 neighbors distributed in 8-direction of the prediction point was proven the most suitable one with a relatively smallest prediction error. The next suitable model was the anisotropic spherical model with the same neighboring conditions.