Will Indonesia's Forests Survive Development Pressure? Machine Learning Predictions for Energy-Critical Tropical Watersheds

Land Use and Land Cover (LULC) changes play an important role in influencing the hydrological conditions of a watershed. The conversion of land such as forests, shrubs and grasslands into agricultural land can disrupt the hydrological balance of the watershed. The availability of information related to LULC dynamics in the future is needed to assist sustainable watershed management planning. Machine learning technology, such as Cellular Automata, can provide accurate predicting. The objective of this research is to simulate LULC based on machine learning in the Mamasa Sub-watershed. Two model combinations were employed to simulate LULC: Artificial Neural Network-Cellular Automata (ANN-CA) and Logistic Regression-Cellular Automata (LR-CA). The research results found that the ANN-CA model achieved percent of correctness and overall kappa of 83.6745 and 0.75412, respectively, which were higher than those of the LR-CA model (82.3498 and 0.73361). The prediction results of both model combinations still fall below the actual LULC values, especially in the case of large LULC classes such as forests, range-shrub, rice, and pasture. Conversely, higher accuracy is observed for smaller classes such as wetlands-forested, orchard, residential, and oak. However, it should be noted that this research did not include several socio-economic variables, such as population and income level, which are considered to influence changes in LULC. Future research is expected to analyse the influence of each variable and include some socio-economic variables that may have a significant influence on LULC change.