Prediction of Hotspots in Riau Province, Indonesia Using the Autoregressive Integrated Moving Average (ARIMA) Model

Abstract

Various forms of disasters occur worldwide, one of which is fire. Indonesia has been suffering from frequent land and forest fires. These events are not a new phenomenon and seem to be an annual tradition, especially in the dry season. This nation was most affected by an excessively disastrous forest fire in 2015. The misfortunes suffered were massive and resulted in land and forest damage that may have great economic and environmental costs. One solution to reduce the impacts of such events is to predict the emergence of hotspots. Therefore, in this work, a modeling method using time series produced by the Box-Jenkins' Autoregressive Integrated Moving Average (ARIMA) model was used to predict the appearance of hotspots. Since the forecasting system does not expect any detailed form to be predicted in terms of the time series of historical data, the data demonstrated in the proposed model were different from data from other models used for prediction. The study was conducted based on monthly hotspot occurrence data from January 2014 through June 2019 in Riau Province, Indonesia. The data were downloaded from the collection of the "LAPAN-MODIS-Catalog". Based on the results shown, the Autoregressive Integrated Moving Average (ARIMA) model (2,1,2) produced good predictions based on its lowest value of Mean Squared Error (MSE), 9540.088. Moreover, the proposed model has produced highly accurate forecasts of hotspots for time periods of up to five months using the fitting model of ARIMA (2,1,2), and the values forecasted for 5 months ahead were 25, 31, 26, 30 and 27.

Autoregressive Forecasting of the Number of Forest Fires Using an Accumulated MODIS-Based Fuel Dryness Index

“…Our use of an autoregressive term of 1 lag, allowing us to forecast the number of fires based on the observed number from the previous 10 days, agrees with the observations of [65] who found a temporal correlation of up to 11 days to predict daily arson ignition counts in Florida. Nevertheless, our study, unlike the univariate autoregressive approach of Prestemon et al [66] or other authors [67,68], or unlike the majority of literature that have used fire danger indices only (e.g., [23,37,102]), included a combination of both fuel dryness and autoregressive terms of previous fire activity, to forecast fire activity of the next 10 days.…”

“…In this regard, in order to account for the important role of human-based temporal patterns on fire activity, some studies have suggested a good potential for autoregressive techniques (e.g., [65,66]). Compared to a vast majority of studies predicting fire activity from fuel dryness only, this promising autoregressive approach for temporal fire forecasting has nevertheless received relatively less attention in the literature (e.g., [67,68]) and demands further research. In particular, there is a relative knowledge gap in studies aiming at predicting fire activity from both autoregressive fire activity and fuel moisture (e.g., [69][70][71]).…”

Predictive Modeling of Forest Fires in Yunnan Province: An Integration of ARIMA and Stepwise Regression Analysis

“…The AR model represents future values as a linear combination of past observations and current disturbances. The AR model [30] is presented in Equation (2):…”