Diagnosis of seasonally varying regression slope coefficients and application to the MJO
Diagnosis of seasonally varying regression slope coefficients and application to the MJO
复制标题
季节变化回归斜率系数的诊断及其在 MJO 中的应用
DOI：
--
复制
发表时间：
2017
期刊：
影响因子：
--
通讯作者：
P. Roundy
中科院分区：
文献类型：
--
作者：
P. Roundy
简单线性回归经常用于诊断天气和气候时间序列之间的关系。然而，当根据一年中不同时间的数据进行类似计算时，系数会有所不同。本文介绍了一种新的算法来估计季节变化的回归斜率系数。该算法生成一个季节性的时间序列的预测变量之间的预测变量和预测变量的协方差。它通过将季节周期及其选择谐波拟合到预测值和被预测值之间的乘积的时间序列以及预测值的平方的时间序列来实现。在一年中的某一天，得到的季节协方差除以季节方差构成了该天回归斜率系数的最佳拟合。该算法的人工数据进行了测试，与以前的算法具有类似的意图相比，然后应用到一个Madden-Julian振荡指数和全球大气模式之间的关系。统计显著性通过蒙特卡罗技术进行评估，该技术从训练集中包含的年份样本中随机选择。
Simple linear regression is frequently applied to diagnose relationships between time series in weather and climate. Yet, when similar calculations are applied based on data from different times of the year, the coefficients vary. This article introduces a new algorithm to estimate seasonally varying regression slope coefficients. The algorithm generates a seasonal time series of the variance of the predictor and of covariance between the predictor and the predictand. It does so by fitting the seasonal cycle and select harmonics thereof to the time series of products between the values of the predictor and the predictand and to the time series of the squares of the values of the predictor. On a given day of the year, the resulting seasonal covariance divided by the seasonal variance constitutes the best fit to the regression slope coefficient on that day. The algorithm is tested on artificial data, compared with a previous algorithm with similar intent, and then applied to the relationship between a Madden–Julian Oscillation index and global atmospheric patterns. Statistical significance is assessed through a Monte Carlo technique that randomly selects from the sample of years included in a training set.