Fuel Moisture Content Forecasting Using Long Short-Term Memory(LSTM) Model

Abstract

Fuel moisture content (FMC) of live vegetation is a crucial wildfire risk and spread rate driver. Current wildfire warnings generally use historical FMC data, which leads to inaccurate predictions. The accuracy of FMC forecasting can provide data support for assessing wildfire danger. Long short-term memory (LSTM) is a special recurrent neural network that learns long-term dependencies. It is suitable for predicting time series with both long-term and short-term dependencies. This study intends to use the LSTM model to forecast FMC. The experiment is divided into three parts: the conventional model; the LSTM model with FMC as the only feature variable; the LSTM model with seven feature variables, included FMC, maximum temperature (Tmax), minimum temperature (Tmin), mean temperature (Tmean), mean dew point temperature (Tdmean), maximum vapor pressure deficit (VPDmax), minimum vapor pressure deficit (VPDmin). The results of the LSTM model outperform that of the conventional model, and the effect of selecting multiple parameters as feature variables reached the best accuracy.