Home > Research > Publications & Outputs > Imputing censored data with desirable spatial c...
View graph of relations

Imputing censored data with desirable spatial covariance function properties using simulated annealing

Research output: Contribution to Journal/MagazineJournal articlepeer-review

Published
<mark>Journal publication date</mark>07/2012
<mark>Journal</mark>Journal of Geographical Systems
Issue number3
Volume14
Number of pages18
Pages (from-to)265-282
Publication StatusPublished
Early online date12/12/10
<mark>Original language</mark>English

Abstract

When measurements of values that are less than the limit of detection are reported as not detected, the data are referred to as censored. The non-recording of values below the limit of detection is common in soil science research although modelling data affected by censoring can be problematic. This paper develops and tests a modified version of Spatial Simulated Annealing, called Simulated Annealing by Variogram and Histogram form, for drawing values for censored points given a mixed set of observed and censored data. The algorithm aims to maximise the goodness of fitting between the experimental and theoretical variograms (by allowing variation in its parameters) while the imputed values are constrained to a target histogram form. In practice, the experimental histogram is estimated by transforming the available data (interval and exact observations) to quantiles and fitting a plausible distribution. The theoretical distribution of the data is used to constrain the variogram fitting. The proposed simulated annealing method is designed to find the optimal spatial arrangement of values, given by the lowest errors in variogram and histogram fitting and kriging prediction. The accuracy of the method proposed is assessed on a simulated data set in which the censored point values are known and compared with the Spatial Simulated Annealing algorithm. According to the results obtained, the Simulated Annealing by Variogram and Histogram form (SAVH) approach can be recommended as a useful tool for the analysis of spatially distributed data with censoring.