Answers are grounded in this paper and cite their sources.
Tropical peatland fires are complex environmental hazards with significant impacts. However, existing fire risk approaches, including empirical drought-fire indices and machine learning models that focus on either spatial or temporal aspects, remain limited in representing the dynamics of fire occurrence across heterogeneous peatland landscapes. Here, we develop an integrated spatio-temporal machine learning model with a probabilistic calibration method to analyze and predict tropical peatland fire risk. The framework employs a Convolutional Neural Network (CNN) to extract spatial features, Convolutional Long Short-Term Memory (ConvLSTM) to capture temporal dependencies, and isotonic regression to calibrate predicted fire probabilities. We employ meteorological, hydrological, land cover, anthropogenic, and historical fire data from tropical peatlands in Kalimantan, Indonesia. The proposed model achieves good performance and produces probability of fire maps that are consistent with observed fire hotspot distributions. Furthermore, the forecasting results show that the model can identify shifting fire risk patterns across peatland areas over different time horizons. These findings demonstrate that the proposed framework provides not only a predictive tool but also actionable information for early warning, risk assessment, and fire mitigation in tropical peatland ecosystems.
No takes yet. Share an insight, caveat, or question.
Mahdiyasa et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: