Correlation of Climate Variability and Burned Area in Borneo using Clustering Methods

Abstract

The island of Borneo has faced seasonal forest fires for decades. This phenomenon is worsening during dry seasons, especially when droughts are concurrent with the El Niño-Southern Oscillation (ENSO) phenomenon. Climate is therefore one of the drivers of the fire phenomenon. This paper studies the relationship between climate variables, namely temperature, precipitation, relative humidity, and wind speed, and the occurrence of forest fire using two clustering methods, K-means and Fuzzy C-means (FCM) clustering methods. Borneo is clustered into four areas based on burned area data obtained from Global Fire Emission Data (GFED). It is also clustered according to the combinations of climate variables. Both methods reach the highest correlation between the climate variable and the burned area clusters in September. The K-means method gives a correlation of -0.54 while the FCM gives -0.55. In August until October, relative humidity provides the dominant correlation affecting burned area, even though an additional precipitation or wind variable slightly increases the correlation in the FCM method. In November, temperature largely contributed to the burned area by a positive correlation of 0.31 in K-means and 0.33 in FCM. The evaluation performance of the methods is conducted by an internal validation called the Silhouette index. Both methods have positive index values ranging from 0.39 to 0.69 and the maximum value is influenced by the wind cluster. This indicates that the clustering methods applied in this paper can identify one or a combination of climate variables into dense and well-separated clusters.

Time-Series Model for Climatological Forest Fire Prediction over Borneo

“…The same incident happened again in 2014/2015 [2]. The study results in [4] also showed that the climate variable strongly correlates with fires in Borneo.…”

“…There are many studies about forest fires that implement machine learning, as described in [4], [5], [6]. The problem domains in those studies also vary, from fire detection to planning and policy.…”

“…The problem domains in those studies also vary, from fire detection to planning and policy. Some of the methods used for fire prediction are random forest [7], clustering [4], neural networks [6], and deep learning [8]. Both [6] and [8] use time-series data in their studies.…”

Scholarly Interest in Forest Fires in Indonesia: A Bibliographical Review

“…The 1997/1998 ENSO-driven fire in Sumatra and Kalimantan was the most studied. In general, climatic factors significantly affect the occurrence of fires (Brasika, 2022;Hidayati et al, 2020). ENSO-related fires have encouraged more scientific inquiries in Indonesia and other tropical countries (Alencar et al, 2006;Juárez-Orozco et al, 2017).…”

Implementation of Plant Selection Based-On Plant Growth on Revegetation of Peatland in South Kalimantan

“…study by [6] shows that the relative humidity contributes most to the burned area in Kalimantan. Meanwhile, revegetation is able to maintain a vegetation cover on peat and increasing humidity in the soil and air, slowing peat decomposition and also decreasing fire risks, thus becoming the second important tenet of peatland restoration after restoration of the hydrology [7].…”