Intelligent LSTM-based deforestation prediction model and dashboard for Kalimantan using global forest change and night-time light data

Abstract

Forest assessment based on predictions and evidence is an appropriate step toward environmental conservation. This study developed a Long Short-Term Memory (LSTM) based prediction model to monitor and predict forest cover and light intensity in Kalimantan. Using Global Forest Change (GFC) and Night-Time Light data, the prediction results are applied to a web-based dashboard. The data were available for the period from 2001 to 2023 and has been processed and aggregated to the provincial level. The LSTM model was trained using five optimizers, namely Adam, RMSprop, Stochastic Gradient Descent (SGD), Adadelta, and Adamax. Model performance was evaluated using Root Mean Squared Error (RMSE). The results showed that the Adam and RMSprop optimizers produced lower RMSE values compared to the other optimizers on the GFC dataset, making them more effective at learning patterns and forest cover change. On the NTL dataset, Adam, RMSprop, and Adamax demonstrated relatively similar performance due to the data’s more stable characteristics. Overall, Adam provided the most accurate and consistent predictions. The prediction results are presented via a web-based dashboard that allows for the visualization of historical trends as well as predictions of change in forest cover and nighttime light intensity. These findings suggest that the combination of the LSTM model and an interactive dashboard can support more effective deforestation monitoring.