A Data Driven Spatial Sampling Approach

• Main Authors: Cheng,Ching-Ru, 鄭敬儒
• Other Authors: Chen,Chun-Shu
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/xbhz49

碩士 === 國立彰化師範大學 === 統計資訊研究所 === 107 === To predict underlying surface based on noisy spatial dataset is one of the objectives in spatial statistics. Even in the same prediction, different sampling schemes will lead to different prediction results and favoring of different generating mechanisms. In this thesis, we focus on constructing a selection procedure for sampling scheme. From a prediction perspective, a data driven selection based on the mean squared prediction errors is proposed, and a generalized degrees of freedom is discovered to evaluate the complexity of prediction result corresponding to the utilized scheme. The proposed selection procedure is achieved by the data perturbation technique in practice. Validities of the proposed method are illustrated theoretically and numerically. Finally, we demonstrate the applicability of the proposed method by analyzing groundwater data set in Bangladesh.