Fire Hotspots Mapping and Forecasting in Indonesia Using Deep Learning Algorithm

Abstract

Indonesia is one of the countries in South East Asia has significant forest fire with dangerous impact to neighboring countries of the emission of haze and carbon. In this research aims to do plotting and mapping location with high number fire hotspot then forecasting potential number of hotspots in future time based on previous of history data collected. The forecasting data achieve is very important and beneficial for the authorities as one of references for preventive action and avoid scattering of forest fire. Long Short-Term Memory (LSTM) algorithm implemented in this research for analysis and forecasting of fire hotspot number while Python programming used to plot hotspot point. The source of fire hotspot dataset is referred to The National Aeronautics and Space Administration (NASA) Moderate Resolution Imaging Spectroradiometer (MODIS) recorded from year 2021 with total number is about 100,000 hotspots in Indonesia region. Results show the distribution of fire hotspot concentration most in Sumatra and Kalimantan Island because the typical of land which peat that potential for getting fire. Forecasting of number hotspot for the year 2022 has achieve with good results with error less than 5% which only 4.56%.