PEATLAND FIRE DETECTION USING SPATIO-TEMPORAL DATA MINING ANALYSIS IN KALIMANTAN, INDONESIA

Abstract

Peatland fire has been an important environmental issue in Indonesia as well as in ASEAN region as it is strongly related with transboundary haze pollution. Hotspot has been used as an indicator of forest and land fires detection. Source of hotspot in Indonesia varies from a low to a relatively high degree of accuracy. Not all hotspots are strong indicators of forest and land fires. The incidence of hotspots in a sequence of at least three days in an adjacent location can be a strong indicator of land and forest fire occurrences. The use of spatio-temporal data mining approach to obtain sequence patterns of hotspots is still in question. This study aims to obtain sequence patterns of hotspot occurrence in Kalimantan peatland in 2015 by applying a spatio-temporal data mining approach. The algorithm used is Sequential Pattern Discovery Equivalent Classes (SPADE) that enables to discover all possible sequence patterns on hotspot datasets. Verification of hotspot sequences was conducted through ground truth and fire supression data. The study proved that real fire detection is indicated by three consecutive days of hotspot occurrences. The fire detection in peatland area is strongly affected by fire characteristics dominated by ground fire.