Imputation Method for Missing Values Considering Temperature Fluctuations with High Locality

Abstract

In recent years, the urban heat island phenomenon in the Tokyo ward area has become a significant problem. To determine the regional characteristics of the urban heat island, a high spatial density meteorological observation system is required. For this purpose, Tokyo Metropolitan Research Institute for Environmental Protection installed a high-density meteorological observation system called METROS in the Tokyo ward area in July 2002. However, missing values are often present in the observed data. Analyzing data without handling missing values can cause loss of precision and biased estimates. Thus, an imputation method for missing values that considers the temperature characteristics of the Tokyo ward area is required. In this study, we propose two imputation methods considering temperature fluctuations with high locality in central Tokyo. Both methods are modification of the inverse distance weight (IDW) method, which is a general imputation method for spatial data. Our proposed methods correct surrounding observed values before imputing unobserved value by their weighted average. Different approaches are considered for correcting surrounding observed values. A simulation study based on various missing data was conducted. Simulations revealed that our proposed methods exhibited higher performance in different settings of missing data than the IDW method.