Dani Al Mahkya, Anik Djuraidah, Aji Hamim Wigena, Bagus Sartono
In the modern era, the possibility for large-scale data collection becomes better and easier through various methodologies. When data are collected with respect to both spatial and temporal dimensions, it is referred to as spatiotemporal data. Such datasets often exhibit autocorrelation arising from spatial proximity, temporal continuity, or the interaction of both dimensions, which is commonly called space-time autocorrelation. Rainfall measurements recorded at different locations and at different times are important due to their relevance in various fields. Due to its spatiotemporal nature, rainfall data often exhibit space-time autocorrelation. Empirical studies further show that rainfall patterns exhibit spatial and temporal variations, which contribute to heterogeneity across regions and times. Motivated by these challenges, this study aims to develop a modeling framework that integrates Eigenvector Space-Time Filtering (ESTF) based on the Varying Coefficient Model to address complex space-time autocorrelations in rainfall data, while integrating CMIP6 global climate projections with local characteristics to enhance the model's relevance to regional conditions. This study uses General Circulation Model (GCM) output from CMIP6-DCPP and incorporates local geographical and environmental features in the modeling process. The GCM output data is represented by 22 principal components to overcome multicollinearity. Furthermore, the varying coefficient components are used to explore the effect of spatial and temporal varying variables on rainfall. The modeling results show that the use of space-time dependency structure in the eigenvector filtering approach effectively reduces space-time autocorrelation. In addition, the performance of the model also improved compared to the OLS, ESF, and ESF-VC models. The final ESTF-VC model shows the best performance based on the RMSE (58.33), R2 (0.64), and AIC (22128.94) values. Based on these results, it can be concluded that the ESTF-VC model is able to handle space-time autocorrelation in rainfall modeling using GCM outputs and local characteristic information and improve estimation performance. © (2025), (Pandawa Institute). All rights reserved.
Statistics and Data Science Study Program, IPB University, Bogor, 16680, Indonesia; Statistics and Data Science Study Program, IPB University, Bogor, 16680, Indonesia; Actuarial Science Study Program, Institut Teknologi Sumatera, Lampung Selatan, 35365, Indonesia; Statistics and Data Science Study Program, IPB University, Bogor, 16680, Indonesia; Statistics and Data Science Study Program, IPB University, Bogor, 16680, Indonesia