Quantitative Assessment of Interpolation Methods for Handling Missing Data in Long-Term Tropical Meteorological Time Series

Abstract

The integrity of long-term tropical meteorological time series is crucial for climate modeling, yet missing data compromises dataset reliability. This study aimed to identify the most accurate interpolation method for handling missing values in key meteorological variables to support robust climate research and action. This study conducted a simulation study using a 14-year daily time series (2010–2023) from North Aceh, Indonesia, comprising 25,565 observations across five variables: temperature, humidity, rainfall, sunshine duration, and wind speed. The baseline dataset was subjected to simple random missingness at 10%, 20%, and 30%. Four interpolation methods: linear, spline, stineman, and moving average were applied. The performance of these methods was rigorously evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Scaled Error (MASE). Linear interpolation consistently demonstrated superior performance, yielding the lowest error rates across all variables and missing data percentages. In contrast, spline and moving average methods exhibited higher error metrics and greater sensitivity to outliers, particularly for volatile variables like rainfall. Stineman interpolation showed moderate, intermediate performance. For long-term tropical meteorological data, linear interpolation is the most effective and stable method for imputing missing values. This finding provides a validated, practical solution for enhancing the quality of climate records, a fundamental requirement for achieving the monitoring and analysis goals of Sustainable Development Goal 13 on Climate Action. The study recommends adopting this method for creating reliable datasets for climate trend analysis and modeling.