ARSTAN程序和R语言dplR扩展包进行树轮年表分析的比较研究

Abstract

PDF HTML阅读 XML下载 导出引用 引用提醒 ARSTAN程序和R语言dplR扩展包进行树轮年表分析的比较研究 DOI: 10.5846/stxb201403300597 作者: 作者单位: 北京师范大学资源学院;北京师范大学地表过程与资源生态国家重点实验室,北京师范大学资源学院;北京师范大学地表过程与资源生态国家重点实验室 作者简介: 通讯作者: 中图分类号: 基金项目: 国家自然科学基金资助项目(41171067) A comparative analysis of ARSTAN and the dplr package of R language in analyses of tree-ring chronologies Author: Affiliation: College of Resources Science and Technology, Beijing Normal University; State Key Laboratory of Earth Surface Process and Re-source Ecology, Beijing Normal University,College of Resources Science and Technology, Beijing Normal University; State Key Laboratory of Earth Surface Process and Re-source Ecology, Beijing Normal University Fund Project: 摘要 | 图/表 | 访问统计 | 参考文献 | 相似文献 | 引证文献 | 资源附件 | 文章评论 摘要:在树轮年代学领域,ARSTAN是去趋势处理和建立年表方面应用最为广泛的程序,而新兴的R语言dplR扩展包实现了ARSTAN的主要功能,且具有源代码公开、扩展性强等优点,是传统程序的良好补充.使用贺兰山青海云杉(Picea Crassifolia)树轮宽度数据,分析了ARSTAN和dplR进行树轮年代学分析所得结果的差异.结果显示,两种程序计算平均敏感度和一阶自相关系数的平均误差为0.005-0.008,但具有确定的转换关系;两种程序如果使用同种方法去趋势,拟合曲线的参数相近,建立标准年表的平均误差为0.002;拟合自回归模型时差异较大,其中时域上表现为差值年表起始30a内差异显著,在频域上表现为dplR的差值年表保留的低频信息较少;年表统计量计算和公共区间分析中,不同程序计算样本总体代表性和信噪比的差异较大.分析表明,两程序在拟合生长趋势和自回归模型时存在算法上的较大差异,同时年表统计量和公共区间各指标的算法也不尽一致,但存在较为确定的转换关系.对开展不同来源数据的整合分析提出了建议,应明确不同研究中树轮数据的处理过程,在条件允许时使用同一程序或算法重新处理数据,确保结果的可比性. Abstract:Dendrochronology plays an important role in estimating past climatic conditions and predicting future climate change. Detrending and chronology development are the fundamental steps of the study of dendrochronology. ARSTAN is the most popular program used to accomplish this step, and it has played an important role in the development of dendrochronology. However, ARSTAN uses the Fortran programming language, so users find it difficult to understand and revise the algorithm of the source program to meet their needs. An emerging package of the R language named dplR provides similar functions to ARSTAN. R and dplR's source code is fully open to the public; thus, it has numerous users. When scholars from different domains communicate and share the methods and results of dendrochronology, it can help them improve those chronologies. In addition, R and dplR have become a good supplement to traditional analysis software. This paper compares the different dendrochronological analysis algorithms and results provided by ARSTAN and dplR with tree ring width data from Picea crassifolia on Helan Mountain, Ningxia Hui Autonomous Region, China. The results show that the two programs calculated exactly the same means and standard deviations. The mean error of the mean sensitivities (MS) and first-order autocorrelations (AC) were 0.005 and 0.008, respectively, but they had a clear conversion relationship. When using the same method for detrending with both types of software, the parameters of fitting curves were generally equal, and the corresponding standard chronologies developed by the two programs had a mean error of only 0.002. However, the residual chronologies were very different. In the time domain, a significant difference was observed in the residual chronologies in the first 20-30 years. In the frequency domain, the residual chronologies created using ARSTAN showed more low frequency information than that created using dplR. For example, the former showed periods of 32 years with higher power than those of dplR. In the common interval analysis, ARSTAN gave a higher expressed population signal (EPS) and signal-to-noise ratio (SNR) of chronologies than dplR. EPS error was 0.4% and SNR error was 30%-40%. By comparing the algorithms of the two programs, we found that ARSTAN and dplR have different initial value setting rules and nonlinear fitting methods to choose the best fitting model during detrending. When fitting an autoregression model, ARSTAN used a pooled algorithm to find the integral growing pattern and used the same fitting order for different sequences. However, dplR directly used different optimal fitting models for different sequences. In addition, the two programs used different, but similar, formulas for calculating MS, AC, EPS, and SNR. Although the absolute value of the results was different, calculation results of the same program using different data were comparable. In conclusion, this paper offers two suggestions for the meta-analysis of tree ring data from different sources. First, if the source data are available, researchers should choose a single program for statistical calculation, detrending, and common interval analysis based on their needs. Second, if the source data are not available, information related to the chronologies is sufficient; researchers should use only a single program to calculate EPS and SNR chronological statistics to ensure that the results will be comparable. 参考文献 相似文献 引证文献