地理科学进展 ›› 2016, Vol. 35 ›› Issue (1): 78-88.doi: 10.18306/dlkxjz.2016.01.009

• • 上一篇    下一篇

切片采样算法在光释光年龄模型参数估计中的应用

彭俊1(), 董治宝1, 韩凤清2   

  1. 1. 中国科学院寒区旱区环境与工程研究所,兰州 730000
    2. 中国科学院青海盐湖研究所,西宁 810008
  • 出版日期:2016-01-31 发布日期:2016-01-31
  • 作者简介:

    作者简介:彭俊(1987-),男,湖北红安人,博士生,主要从事释光年代学与气候变化研究,E-mail: pengjun10@mails.ucas.ac.cn

  • 基金资助:
    基金项目:国家重点基础研究发展计划(973计划)项目(2013CB956000,2012CB426501)

Application of slice sampling method for optimizing OSL age models used for equivalent dose determination

Jun PENG1(), Zhibao DONG1, Fengqing HAN2   

  1. 1. Cold and Arid Regions Environmental and Engineering Research Institute, CAS, Lanzhou 730000, China
    2. Qinghai Institute of Salt Lakes, CAS, Xining 810008, China
  • Online:2016-01-31 Published:2016-01-31
  • Supported by:
    National Basic ResearchProgram of China (973 Program), No.2013CB956000, No.2012CB426501

摘要:

光释光(OSL)年代学模型是基于数理统计学的一类概率密度模型,它根据特定的假设条件对样品等效剂量(De)分布进行数学解释,由此估计具有不同沉积历史或者能够代表样品实际埋藏年龄的De组分。年龄模型参数估计常通过极大似然估计(MLE)算法实现,本文尝试了切片采样算法在年龄模型参数优化中的应用。切片采样属于一种马尔科夫链蒙特卡罗采样(MCMC)算法,能根据测量数据与模型的联合似然函数进行随机采样,由此获得参数的采样分布。本文编写了实现年龄模型切片采样算法的应用程序,并使用模拟及实测De数据验证了该算法估计的可靠性。相对于MLE算法,MCMC算法具有对参数初值依赖性低、误差估计更准确的特点,切片采样算法提供了实现释光年龄模型参数估计的一种新方法。

关键词: OSL测年, 等效剂量, 年龄模型, 参数优化, 随机采样

Abstract:

In Optically Stimulated Luminescence (OSL) dating, statistical age models used for equivalent dose (De) determination are probabilistic models constructed according to mathematical statistics. They are applied to distinguish De populations that are sedimentologically different or to determine a De value that represents the burial dose of a sample. Maximum likelihood estimation (MLE) method is routinely used to optimize parameters of an age model. In the present study, we used the Slice sampling algorithm to determine the parameters of age models. Slice sampling is a Markov chain Monte Carlo (MCMC) sampling method, which enables the sampling distributions of parameters to be obtained from the joint likelihood function that is determined by observations and the specified model. This study applied easily implemented and openly accessible numeric routines to performing the algorithm. We used artificial and measured datasets to check the reliability of the estimates. MCMC method is insensitive to the parameters’ initial states, and the standard errors (or confidence intervals) of parameters assessed using this method are more reliable compared to those based on the Fisher information matrix constructed through numerical differentiation. Our results indicate that the Slice sampling method provides an alternative for age model optimization. Slice sampling method generates an informative estimation for the results of MLE method in age model application.

Key words: OSL dating, equivalent dose, statistical age models, parameter optimization, stochastic sampling