學術會議
國際學術報告:Statistical Framework for Simultaneous Handling of Input, Parameter, Model and Observational Error in Spatial Uncertainty Analysis
文章來源： | 發布時間：2016-10-12 | 【打印】 【關閉】
報告人：Gerard B.M. Heuvelink,
內容簡介：Spatial uncertainty propagation analysis is usually concerned with analysis of how uncertainties in the inputs to spatial environmental models propagate to the model output. However, input uncertainty is only one of several sources of uncertainty that affect the accuracy of the model output. Often there are also uncertainties about the parameters and structure of the model. In many cases model uncertainty may even be the dominant source of uncertainty, which implies that it cannot be ignored if a realistic assessment of the overall output uncertainty is to be made. In this work we present an uncertainty analysis framework that integrates all sources of uncertainty. The framework assumes that the uncertainties about the model inputs have been derived externally, while uncertainties about model parameters and model structure are obtained using Bayesian calibration. In a first stage, model structural uncertainty is represented by an additive stochastic residual. Prior distributions for the parameter and structural uncertainty are defined, and next observations of the model output are used to calculate posterior distributions. The Bayesian calibration procedure takes into account that part of the discrepancies between observed and predicted model output is caused by input uncertainty and by observational error. Once the calibration is completed and all error sources have been characterized by probability distributions, the uncertainty propagation analysis is done using a straightforward Monte Carlo approach. The framework is illustrated with a simple example in which a multiple regression model is used to predict the moisture content at wilting point from soil porosity and moisture content at field capacity for a study area in the floodplain of the Allier river, France.
報告人簡介：Gerard Heuvelink博士目前是世界土壤信息中心（ISRIC World Soil Information）計量土壤學與數字土壤制圖領域的高級研究員，同時也是荷蘭Wageningen大學土壤地理和景觀研究組的副教授和中國科學院地理科學與資源研究所的客座教授。他從博士期間開始從事空間數據不確定性及誤差傳遞模型研究，經過多年工作現已發表250余篇關于地統計、空間不確定性分析和計量土壤學的相關論文和著作。論文具有較高引用率，其中多篇論文被評為期刊年度最佳論文。Gerard Heuvelink博士曾在六個歐洲研究項目（如，UNCERSDSS, HarmoniRiB, Intamap, UncertWeb）中擔任過主席或成員，同時還是 European Journal of Soil Science 和 Spatial Statistics 的副主編，Geoderma, Environmental and Ecological Statistics, International Journal of Applied Earth Observation and Geoinformation 和 Geographical Analysis等雜志的編委，并在多個與GIS和spatial accuracy相關的國際會議中出任指導委員會成員。
主 持 人：葛詠 研究員
報告時間：2014年5月23日（星期五）上午9:00
報告地點：中科院地理資源所2209會議室
主辦單位：資源與環境信息系統國家重點實驗室
附件下載：