Electronic Theses and Dissertation

Kehilangan data merupakan tantangan utama dalam analisis iklim yang dapat menyebabkan bias estimasi dan menurunkan akurasi model prediksi, terutama di wilayah tropis seperti Kota Banda Aceh yang memiliki variabilitas iklim. Ketiadaan data yang lengkap dapat menghambat pemodelan serta pengambilan keputusan berbasis data. Penelitian ini bertujuan untuk mengevaluasi dan membandingkan efektivitas tiga metode multiple imputation, yaitu Expectation Maximization (EM), Predictive Mean Matching (PMM), dan Fully Conditional Specification (FCS) dalam menangani data hilang pada data iklim di Kota Banda Aceh. Selain itu, penelitian ini juga menganalisis keterkaitan dinamis antarvariabel iklim menggunakan uji kausalitas Granger untuk mengidentifikasi hubungan sebab-akibat dalam sistem iklim. Data yang digunakan merupakan data iklim harian dari Badan Meteorologi, Klimatologi, dan Geofisika (BMKG) yang dikumpulkan oleh Stasiun Klimatologi Aceh untuk periode 1 Januari 2010 hingga 31 Desember 2024. Variabel yang dianalisis mencakup temperatur, kelembapan, curah hujan, dan lama penyinaran matahari. Analisis dilakukan melalui imputasi awal, penghilangan data sebesar 10%, 20%, dan 30%, serta imputasi ulang, dengan evaluasi performa berdasarkan Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Squared Error (MSE), dan Weighted Absolute Percentage Error (WAPE). Hasil penelitian menunjukkan bahwa metode PMM lebih unggul dalam mempertahankan struktur data asli dibandingkan EM dan FCS, sehingga PMM menjadi metode paling optimal untuk imputasi data iklim di Kota Banda Aceh dengan menghasilkan nilai kesalahan terendah di seluruh tingkat kehilangan data. Uji kausalitas Granger menghasilkan sebelas hubungan kausal signifikan pada variabel iklim (temperatur, kelembapan, curah hujan, lamanya penyinaran matahari) di Kota Banda Aceh dan satu yang tidak signifikan yaitu antara curah hujan dan lama penyinaran matahari.

Kata kunci: EM, PMM, FCS, uji kausalitas Granger, iklim, Banda Aceh.

Missing data is a major challenge in climate analysis that can cause estimation bias and reduce the accuracy of prediction models, especially in tropical regions such as Banda Aceh City where climate variability is high. The absence of complete data can hamper modeling and data-driven decision making. This study aims to evaluate and compare the effectiveness of three multiple imputation methods, namely Expectation Maximization (EM), Predictive Mean Matching (PMM), and Fully Conditional Specification (FCS) in handling missing data on climate data in Banda Aceh City. In addition, this study also analyzes the dynamic interrelationships between climate variables using the Granger causality test to identify cause-and-effect relationships in the climate system. The data used is daily climate data from the Meteorology, Climatology and Geophysics Agency (BMKG) collected by the Aceh Climatology Station for the period January 1, 2010 to December 31, 2024. The variables analyzed include temperature, humidity, rainfall, and sunshine duration. The analysis was conducted through initial imputation, data removal by 10%, 20%, and 30%, and re-imputation, with performance evaluation based on Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Squared Error (MSE), and Weighted Absolute Percentage Error (WAPE). The results show that the PMM method is superior in maintaining the original data structure compared to EM and FCS, so PMM is the most optimal method for imputing climate data in Banda Aceh City by producing the lowest error values at all levels of data loss. The Granger causality test resulted in eleven significant causal relationships on climate variables (temperature, humidity, rainfall, sunshine duration) in Banda Aceh City and one insignificant one between rainfall and sunshine duration. Keywords: EM, PMM, FCS, Granger causality test, climate, Banda Aceh.