STATISTICAL DOWNSCALING DENGAN PERGESERAN WAKTU BERDASARKAN KORELASI SILANG

Abstract

Pergeseran waktu (time lag) dalam analisis data deret waktu diperlukan terutama untuk analisis hubungan dua peubah (variable), seperti dalam statistical downscaling. Pergeseran waktu ini ditentukan berdasarkan korelasi silang tinggi yang setara dengan hubungan yang kuat antar kedua peubah tersebut sehingga dapat digunakan dalam pemodelan untuk prakiraan yang lebih akurat. Makalah ini mengenai statistical downscaling dengan memperhatikan korelasi silang antara data curah hujan dengan data presipitasi Global Circulation Model (GCM) dari Climate Model Inter Comparison Project (CMIP5). Salah satu syarat dalam statistical downscaling adalah peubah  skala lokal dan global berkorelasi tinggi. Kedua tipe peubah tersebut berupa data deret waktu sehingga fungsi korelasi silang diterapkan untuk memperoleh pergeseran waktu. Korelasi silang yang tinggi menentukan pergeseran waktu pada luaran GCM yang menghasilkan hubungan fungsional lebih kuat antara kedua tipe peubah. Model regresi komponen utama dan regresi kuadrat terkecil parsial digunakan dalam makalah ini. Model-model dengan pergeseran waktu menduga curah hujan lebih baik daripada model-model tanpa pergeseran waktu.

Time lag in time series data analysis is required especially to analyze the relationship of two variables, such as in statistical downscaling. Time lag is determined based on high cross correlation which is equivalent to strong relationship between the two variables and can be used in modeling for a more accurate forecast. This paper is about  statistical downscaling by considering the cross correlation between rainfall data and precipitation data from Global Circulation Model (GCM) of Climate Model Inter Comparison Project (CMIP5). One of the conditions in statistical downscaling is that local scale and global scale variables are highly correlated. Both types of variables are time series data, thus cross correlation function is applied to find time lags. High cross correlation determines time lags in GCM output which was resulted in higher functional relation between both types of variables. Principal Component Regression and Partial Least Square Regression model were used in this paper. Models with time lags had forecasted rainfall better than those without time lags.

Time lag in time series data analysis is required especially to analyze the relationship of two variables, such as in statistical downscaling. The time lag is determined based on high cross-correlation which is equivalent to a strong relationship between the two variables and can be used in modeling for a more accurate forecast. This paper is about statistical downscaling by considering the cross-correlation between rainfall data and precipitation data from the Global Circulation Model (GCM) of Climate Model Inter Comparison Project (CMIP5). One of the conditions in statistical downscaling is that local scale and global scale variables are highly correlated. Both types of variables are time series data, thus cross-correlation function is applied to find time lags. High cross-correlation determines time lags in GCM output which was resulted in higher functional relation between both types of variables. Principal Component Regression and Partial Least Square Regression model were used in this paper. Models with time lags had forecasted rainfall better than those without time lags.