A case study in the tropical region to evaluate univariate imputation methods for solar irradiance data with different weather types

Abstract

A complete dataset of ground-measured Global Horizontal Solar Irradiance (GHI) is vital for the design and performance assessment of a photovoltaic (PV) system. Hence, imputing the missing values of the dataset can be a cost-effective solution for a site that already has ground-based measurement devices installed when satellite-derived data cannot offer the required accuracy for the applications. Nevertheless, using one imputation method for all weather types may lead to inaccurate predictions, which may underestimate the investment risk, especially in the tropical region, where the solar irradiance often fluctuates unpredictably. This article investigated 11 univariate imputation methods for Clear Sunny, Sunny, Intermittent, and Cloudy/Overcast weathers, with missing ratios ranging from 10% to 50%. The performances of these imputation methods were evaluated using standard error metrics. Three significance test indicators were further used to rank the imputation methods. Finally, the imputation methods were categorized into three groups. The results found that methods in Spline and Linear families work well for all weather types with Mean Bias Error of Daily Solar Irradiation of less than 0.2%, except Intermittent weather where SMA (k = 2 or 4) often performs the best. Meanwhile, Bezier is only suitable for conservative estimation to minimize the investment risk.