10th International Congress on Information and Communication Technology

Authors - John Khoo, Rayner Alfred, Khalifa Chekima, Rayner Pailus, Chin Kim On, Ervin Gubin Moung, Raymond Alfred, Oliver Valentine Eboy, Normah Awang Besar Raffie, Ashraf Osman Ibrahim Elsayed, Nosius Luaran
Abstract - Climate change remains one of the most pressing global challenges, with effective management of carbon stocks in forests playing a vital role in mitigating its impact. Carbon stock estimation, which quantifies the amount of carbon stored in various biomass forms such as trees, soil, and deceased organic matter, is essential for understanding the role of forests in carbon sequestration and developing strategies to reduce carbon emissions. Traditional methods for carbon stock estimation are often labour-intensive, time-consuming, and lack the precision required for large-scale analysis. Advances in machine learning and remote sensing technologies offer a significant opportunity to improve the accuracy and efficiency of carbon stock estimation. This paper investigates the use of machine learning models, specifically Multilayer Perceptron (MLP), Recurrent Neural Network (RNN), and Autoencoder, for estimating carbon stocks using datasets from sources such as Landsat 7, NDVI, and SAR. Comprehensive analyses were conducted with different train-test splits (70/30, 60/40, 50/50), sample sizes (1,000; 10,000; 100,000), learning rates (0.01, 0.05, 0.1), and epochs (1,000, 10,000, 20,000). The results indicate that the AE model consistently outperforms MLP and RNN models, demonstrating superior predictive accuracy (lower RMSE and MAE) and reliability (higher IOA). The AE model's robustness was evident across all settings, making it the most effective model for carbon stock estimation. In contrast, the RNN model showed higher error rates and longer training times, particularly with smaller sample sizes and higher learning rates. The MLP model exhibited moderate performance. These findings underscore the importance of model selection and hyperparameter tuning in enhancing the accuracy of carbon stock estimation. The study highlights the potential of the AE model as a valuable tool for environmental monitoring and management, providing insights into improving machine learning applications for sustainable ecosystem assessment.