Enhanced Carbon Flux Forecasting via STL Decomposition and Hybrid ARIMA-ES-LSTM Model in Amazon Forest

This study presents a hybrid model, STL-ARIMA-ES-LSTM, developed to improve the accuracy of Gross Primary Productivity (GPP) forecasts in the Amazon region. The model combines seasonal and trend decomposition using Loess (STL) with statistical methods (ARIMA and Exponential Smoothing - ES) and a machine learning technique (Long Short-Term Memory - LSTM). Applied to data from the PE-QFR site, the hybrid model outperformed LSTM alone, with lower prediction error and higher correlation with real data, ensuring a more accurate representation of GPP variability. In addition, the results show the effectiveness of STL decomposition to isolate trend, seasonality and residual components, allowing for more accurate and robust modeling. The study reinforces the potential of hybrid approaches to capture complex environmental dynamics and contributes to a deeper understanding of the carbon cycle, offering promising tools for environmental monitoring and climate change mitigation strategies in the Amazon.