Peatfr: An R package to forecast tropical peatland fire risk with stochastic, machine learning, and optimisation methods

Early detection of tropical peatland fire is crucial to anticipate and mitigate fire risks effectively. However, existing software packages for predicting the occurrence of fires often lack comprehensive integration of methods and techniques, which potentially limits their application. Here, we propose peatfr , a novel R package to forecast tropical peat fire risk with stochastic, machine learning, and optimisation methods. The peatfr is designed with three main functions, which comprise of data imputation, time series forecasting, and fire risk prediction. The data imputation process addresses missing or incomplete time series data, ensuring that the dataset remains reliable for the subsequent analysis and forecasting. This package provides stochastic and machine learning methods, including ARIMA with Box-Cox transformation, Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU), to forecast time series data that significantly influence tropical peatland fire. The fire risk is calculated through Peat Fire Vulnerability Index, which employs Nelder-Mead optimisation method to obtain the optimal parameters. This approach allows peatfr to update the parameters automatically based on the input data, which results in a reliable package. Furthermore, because the peatfr operates independently without relying on external software to forecast the fire event, this package becomes straightforward and user-friendly. The capability to update parameters automatically and self-contained framework enhance its potential as an early warning tool for tropical peatland fire. We demonstrate the application of peatfr to predict the occurrence of tropical peatland fire in Sabangau, Central Kalimantan, Indonesia. • We developed a novel R package peatfr to forecast tropical peat fire risk. • peatfr employs stochastics, machine learning, and optimisation methods. • The main process includes data imputation, forecasting, and fire risk calculation. • peatfr parameters were updated automatically based on the input data. • The package operates independently without relying on external software.