Handling Missing Values and Unusual Observations in Statistical Downscaling Using Kalman Filter

• Published in: Journal of physics. Conference series 2021-03, Vol.1863 (1), p.12035
• Main Authors: Saputra, M D, Hadi, A F, Riski, A, Anggraeni, D
• Format: Article
• Language: English
• Subjects: Data analysis; Forecasting; Kalman filters; Model testing; Physics; Rainfall; Root-mean-square errors; State space models
• ISSN: 1742-6588; 1742-6596
• DOI: 10.1088/1742-6596/1863/1/012035

Rainfall forecasting model using data Global Circular Model (GCM) with Statistical Downscaling technique has a fairly high accuracy. However, missing local climate information poses a constraint in data analysis and forecasting. Missing value imputation is one solution that can be used. Kalman Filter Imputation and State Space Model Arima are imputation methods that operate recursively where there is an update of prediction values when data updates occur. This study aimed to find the best model to use for missing value imputation with small imputation errors. The results of the missing value imputation were used to obtain the best statistical downscaling model on a 3 × 3 to 12 × 12 grid. The research was conducted on the daily rainfall data of Kupang City with 17% missing values and 8% unusual data at the Eltari Meteorological observation station, Kupang city. The average daily rainfall data in East Nusa Tenggara Province were utilized as a reference for the characteristics of rainfall data at the Kupang City observation station. The best missing value imputation was obtained by using the Arima State-Space Model (2,1,1) with a Root Mean Square Error (RMSE) of 0.930 and the model was statistical downscaling best obtained on a grid 6 × 6 with a Mean Absolute Percentage Error (MAPE) of 1.3 % and the number of PCs 11.