<![CDATA[The high rate of deforestation in Indonesia due to forest and land fires (karhutla) is still a problem that requires the government's attention because it has become a regional and global disaster. The worst forest fire incident in Indonesia occurred in 2019, where the area of the fire was 1,649,258 ha. Riau Province is one of the provinces in Indonesia that often experiences forest fires. Sipongi noted that an average of 52,986 ha of forest and land burned in Riau Province every year from 2016-2020. Thus, this study builds a predictive model for the emergence of hotspots as one of the forest fires that aims to reduce the rate of forest fires. Prediction model built using Long Short-Term Memory Recurrent Neural Network (LSTM-RNN). The modeling is carried out using 2 data scenarios, namely multivariate data and univariate data, where multivariate data uses weather variables as predictors of hotspot frequency, and univariate data is hotspot frequency data. The data used is daily data from 2013-2020. Multivariate scenario dataset that produces RMSE of 23,323 and the correlation between actual and predicted data is 0,675554. The RMSE generated by the multivariate dataset is smaller than the RMSE generated by the model with the univariate dataset scenario, which is 25,750. However, datasets with univariate scenarios produce a larger correlation between actual and predicted values when compared to multivariate dataset scenarios. The addition of weather factors as a predictor of hotspot occurrence can improve model performance, where this model is better at predicting values when compared to univariate dataset scenarios even though the running time is longer. Keywords: forest and land fire, hotspots, Long Short-Term Memory, Recurrent Neural Network, prediction, time series]]>
Copyrights © 2023
