Semiparametric Time Series Regression Using Exponential Complex Fourier Series for Temperature Prediction in the Tropical Rainforest of Samarinda

Abstract

Introduction. Forecasting temperature in tropical rainforest regions is challenging due to complex seasonality and nonlinear trends that conventional models often fail to capture. Addressing this, advanced modeling approaches are required for more accurate climate predictions. Objectives. This study aims to construct and apply the STSR-ECFS model for temperature forecasting in Samarinda, East Kalimantan, and to assess its predictive performance using standard accuracy metrics. Materials and Methods. The STSR-ECFS model combines an autoregressive parametric component with a nonparametric component constructed using the exponential complex Fourier series. The model was trained and validated on monthly temperature data from 2015 to 2024, with the optimal number of oscillations determined by Generalized Cross Validation (GCV). Performance was assessed using Mean Squared Error (MSE) and Mean Absolute Percentage Error (MAPE) for both in sample and outof- sample data. Results and Discussion. The model demonstrated high predictive accuracy, with MAPE consistently less than 10% for both in-sample and out-of sample. It effectively captured both seasonal fluctuations and long-term warming trends in Samarinda’s temperature data. Conclusions. The STSR-ECFS is a flexible and accurate model for temperature forecasting in tropical rainforest climates, with potential applications in climate adaptation strategies, early warning systems, and other climate variables with similar data characteristics. The model could inform policy decisions on climate adaptation, aiding local governments and environmental agencies in managing risks and formulating mitigation strategies. Its integration into national climate action plans can enhance decision-making for sustainable development and disaster risk reduction.