国設大気測定綱(NASN)浮遊粒子状物質観測データにおける欠損値の補完

概要

論文の詳細を見る
Incomplete data set, which has missing values in the matrix, is a problem that is repeatedly encountered in environmental researches. To cope with the problem, this study aims to provide useful methods to complement missing values in air quality data sets. As a case study, authors have tested some methods to fill in the missing values in compiled SPM and multi-elemental concentration data by National Air Surveillance Network, Japan, using simulated missing data patterns. Missing rows in the matrices were filled with mean of surrounding values (MSV), mean of values before and after 12 months of the missing data (MV12) and linear interpolation (LIN). Performance of those methods were evaluated by statistical indices such as Index of Agreement developed by Willmott, root mean square squared error (RMSE) and mean absolute error (MAE). Whereas MSV and LIN were strong tools to fill in the short length of gap, MV12 could be an alternative way to interpolate the longer length of gap. Liner regression analysis was applied to complement the missing values in column. This method was found useful when a correlation coefficient was greater than 0.7 between concentrations of element with missing values and predictor element. Based on the test results, full complement was demonstrated on the incomplete data matrix of Kawasaki site from April 1974 to March 1997.
東海大学の論文
2006-03-30