Multivariate analysis on Indonesian forest fire using combined empirical orthogonal function and covariance matrices

Abstract

Forest and land fires are annual events that occur very frequently during the dry season in several regions in Indonesia. One of the main triggers is low rainfall and extreme anomalies in those regions. The objective of the research was to find the correlation between rainfall and Forest fire based on Tropical Rainfall Measuring Mission (TRMM) rainfall data and Global Fire Emission Database (GFED) forest fire data. Both of these data are big data, this research utilized singular value decomposition to generate information on the correlation of fire patterns and rainfall spatially and temporally. Empirical Orthogonal Function (EOF) based on Singular value decomposition (SVD) was used to reduce large data, without removing main information from the data. SVD approach in this research was conducted using the covariance matrix and combined EOF. EOF mode contribution is able to explain the joint variability of the data set to the phenomenon of fire and its relation to rainfall in Indonesia. Spatial patterns in this analysis show that Kalimantan, Sumatra, and Papua emerged as endemic areas for fires in 1998, 2002, 2006, 2009 and 2015. The year of 2004 and 2014 was also recorded as years of severe forest and land fires but was relatively not as severe as mentioned earlier. The temporal pattern shows that the characteristics of Forest fire in Indonesia are dominant in the final months of the year, from June to October.