Scholarly article on topic 'Hotspot Pattern Distribution in Peat Land Area in Sumatera Based on Spatio Temporal Clustering'

Hotspot Pattern Distribution in Peat Land Area in Sumatera Based on Spatio Temporal Clustering Academic research paper on "Earth and related environmental sciences"

CC BY-NC-ND
0
0
Share paper
Keywords
{Cluster / hotspot / "peat land" / "scan statistics" / "spatio temporal"}

Abstract of research paper on Earth and related environmental sciences, author of scientific article — Annisa Puspa Kirana, Imas Sukaesih Sitanggang, Lailan Syaufina

Abstract This research applied statistical approach to recognize the distribution pattern of hotspot clusters in spatial and temporal dimension using Kulldorff's Scan Statistic (KSS) method. We analyzed hotspot data in peat land area in Sumatera from 2001 to 2014. The result shows that provinces with the highest hotspot occurrence cluster are Riau and South Sumatera province. Clusters distribution of hotspot in the period of 2001-2006 are mostly found in ‘Hemic (100) moderate’ and ‘Hemic/Sapric (60/40) deep’ peat land maturity of level. During the period of 2007-2014, the distribution of clusters are mostly found in ‘Hemic/Sapric (60/40), deep’ and Hemic/Sapric (60/40), very deep’. Whereas, in term of the peat land thickness, there was a shift in the distribution of hotspots and the use of peat land area from the ‘moderate’ and ‘very deep’ depth in 2001 until 2006 into ‘deep’ and ‘very deep’ peat land types in 2007 until 2014. Based on the decomposition level of peat, hotspot clusters are mostly found in ‘hemic’ peat land maturity level and land use type of ‘swamp forest’.

Academic research paper on topic "Hotspot Pattern Distribution in Peat Land Area in Sumatera Based on Spatio Temporal Clustering"

CrossMark

Available online at www.sciencedirect.com

ScienceDirect

Procedia Environmental Sciences 33 (2016) 635 - 645

The 2nd International Symposium on LAPAN-IPB Satellite for Food Security and Environmental

Monitoring 2015, LISAT-FSEM 2015

Hotspot pattern distribution in peat land area in Sumatera based on

spatio temporal clustering

Annisa Puspa Kirana^*, Imas Sukaesih Sitangganga, Lailan Syaufinab

a Faculty of Mathematics and Natural Science, Bogor Agricultural University, Jl. Meranti Kampus, Dramaga, Bogor 16680, Indonesia b Faculty of Forestry, Bogor Agricultural University, Jl. Lingkar Akademik Kampus, Dramaga, Bogor 16680, Indonesia

Abstract

This research applied statistical approach to recognize the distribution pattern of hotspot clusters in spatial and temporal dimension using Kulldorffs Scan Statistic (KSS) method. We analyzed hotspot data in peat land area in Sumatera from 2001 to 2014. The result shows that provinces with the highest hotspot occurrence cluster are Riau and South Sumatera province. Clusters distribution of hotspot in the period of 2001-2006 are mostly found in 'Hemic (100) moderate' and 'Hemic/Sapric (60/40) deep' peat land maturity of level. During the period of 2007-2014, the distribution of clusters are mostly found in 'Hemic/Sapric (60/40), deep' and Hemic/Sapric (60/40), very deep'. Whereas, in term of the peat land thickness, there was a shift in the distribution of hotspots and the use of peat land area from the 'moderate' and 'very deep' depth in 2001 until 2006 into 'deep' and 'very deep' peat land types in 2007 until 2014. Based on the decomposition level of peat, hotspot clusters are mostly found in 'hemic' peat land maturity level and land use type of 'swamp forest'.

© 2016 The Authors.PublishedbyElsevierB.V. Thisis an open access article under the CC BY-NC-ND license (http://creativecommons.Org/licenses/by-nc-nd/4.0/).

Peer-review under responsibility of the organizing committee of LISAT-FSEM2015 Keywords: Cluster; hotspot; peat land; scan statistics; spatio temporal

1. Introduction

Forest and land fire is a serious problem and having a huge impact on the ecosystem environment. There are several impacts of forest and land fire, including smog pollution, decreased level of health, damaged ecosystem, high

* Corresponding author. Tel.: +62-812-8588-5114. E-mail address: annisapuspakirana@gmail.com.

1878-0296 © 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Peer-review under responsibility of the organizing committee of LISAT-FSEM2015 doi:10.1016/j.proenv.2016.03.118

release of carbon in the air, and other negative impact on various sectors. The impact of peat land fire is harmful to human life and the environment and it occurs almost every year in Indonesia. A peat land fire in 1997/1998 contributed 13-40% emissions of global carbon emissions it held when the El Nino occurred in Indonesia [1-3]. There are two kinds of factors which trigger the occurrence of that fire: human factors and natural factors. Based on the data [4], in most of the fires cases are caused by human factors, whereas a small fraction of the cases is caused by the natural factors. Very important to develop an early warning system to prevent forest and land fire considering the bad impact on forest and land fire and its various triggering factors. In order to develop an early warning system we need to recognize the distribution pattern of hotspots as one of forest and land fire indicators. By recognizing the distribution pattern of hotspot, we able detect the area that has high fires density so any early prevention steps can be performed in that area. Several studies have been conducted on developing an early warning system for fire prevention by analyzing hotspot occurrences. Sitanggang et al. [5] applied a spatial decision tree algorithm on spatial data of forest fires for classification model for hotspot occurrences. Based on the same algorithm Nurpratami and Sitanggang [6] developed classification rules for hotspot occurrences using spatial entropy. Predictive models for hotspots occurrence are essential in developing early warning systems so that damages caused by forest fires can be minimized. Sitanggang et al. [7] applied decision tree based on spatial relationships for predicting hotspots occurrences. Hotspot occurrence as an indicator of forest and land fires is essential in developing an early warning system for fire prevention. Sitanggang et al. [8] developed hotspot prediction models using decision tree algorithms to generate false alarm data for hotspot prediction models.

This work used clustering method to recognize the distribution pattern of the hotspot occurrence. In this research, we apply statistical approach to recognize the distribution pattern of hotspot in both spatial and temporal domain using Kulldorff's Scan Statistic (KSS) method [9]. The basic idea is that there is scanning windows that moves across space and/or time. For each location and size of the window, the number of observed and expected cases is noted. Every scanning a window will find the likelihood ratio using Poisson model. Then evaluation of the statistical significance of the cluster calculated using Monte Carlo simulation. The region with the highest value of likelihood ratio is the area that has the most potential clusters. The benefits of this research are as early warning system and early detection of forest and land fire, especially in the peat land area in Sumatera by providing illustrations of clustering pattern of hotspot based on the spatial and temporal aspect.

The purpose of this research is to recognize the distribution pattern of hotspot clusters in peat land areas in Sumatera between 2001 until 2014 using KSS method. We analyzed the recognized distribution pattern of hotspot clusters based on physical characteristics of peat land. The physical characteristic of peat land that we use are the maturity level of peat, thickness of peat, and land use of peat. There are several related researches about detecting clusters and distribution pattern of hotspot, especially in the forestry field. For example Usman et al. [10] applied hotspot clustering in Sumatera in the period of 2002 and 2013 based on the density approach using Density-Based spatial clustering. Ekadinata et al. [11] analyzed hotspot distribution in Riau and once of the conclusion found that hotspots are concentrated on the deepest peat soil. Wahyunto et al. [12] about distribution, properties and carbon stock in Indonesia indicated that higher degree of peat maturity will be followed by the high degree of bulk density and C organic content. The high carbon content of peat means peat fires release vast amounts of smoke and carbon in the atmosphere. The massive carbon emissions from peat fires makes major contributor to the global increase in atmospheric CO2 [13].

2. Study Area and Data

2.1. Study area

This study is located in Sumatera Island, Indonesia. Sumatera Island has an area of 443,065.8 km2, which is divided into 10 provinces that is Aceh, Sumatera Barat, Sumatera Utara, Riau, Riau Islands, Bengkulu, Jambi, Sumatera Selatan, Bangka Belitung Islands, Lampung. According to the census in 2010 the total population of Sumatra Island about 52,210,926 [14]. Total peat land in Indonesia is about 20.6 million hectares which 35% of peat land are located in Sumatera [15]. The map of study area shown in Fig. 1.

Fig. 1. The map of study area.

2.2. Data and Tools

Datasets that we use in this research is daily hotspot data with spatial and temporal dimension, especially in the peat land area in Sumatera Island between 2001 until 2014. Hotspot data were obtained from FIRMS MODIS Fire/Hotspot, NASA/University of Maryland that provide an important multi-year record of fire data [16]. FIRMS MODIS Fire/Hotspot data monitor seasonal, inter-annual variability, and long-term trends in fire activity [17, 18]. Peat land distribution data were obtained from Wetland International in 2002 [15]. Spatial domain of hotspot data refers to the location of hotspot fire from longitude and latitude fields. While, temporal refers to the occurrence time of fire from date fields. Peat land data consisting of the maturity level of peat, the depth of peat land, and land use of peat. We analysed the recognized distribution pattern of hotspot clusters based on physical characteristics of peat land. The physical characteristic of peat land that we use are the maturity level of peat, thickness of peat, and land use of peat.

Peat land in Sumatera categorised according to the level of maturity consist of three kinds there are fibric, hemic, and sapric. Fibric is the early stage of peat decomposition where recognizable plant fires dominate. Hemic is intermediate stage of peat decomposition, between fibric and sapric. Then, sapric is the advanced stages of peat decomposition into organic-matter rich 'earth' without visible fires [15]. The maturity level of peat and thickness indicated in the category of peat. Suppose, 'Hemic/Sapric (60/40), very deep' means that Hemic/Sapric is the maturity level of peat. Value (60/40) shows the area covering 60 percent of hemic and 40 percent of sapric. 'Very deep' means that in the category of the thickness of peat with a depth 400-800 cm. The thickness of peat can be grouped to: 1) Very shallow thickness (D0) (having a thickness <50 cm) with maturity level Hemic/Sapric, Hemic/Minerals and Sapric/Hemic, 2) Shallow thickness (D1) (50-100 cm with maturity level Fibric/Sapric, Hemic/Sapric, Hemic/mineral, Sapric/Hemic and Sapric/Mineral, 3) Moderate thickness (D2) (100-200 cm) with maturity level of peat are Hemic/Sapric, Hemic/Mineral, Sapric, Sapric/Hemic and Sapric/Mineral, 4) Deep thickness (D3) (200-400 cm with maturity level of peat are Hemic/Sapric, Sapric and Sapric/Hemic, 5) Very Deep thickness (D4) (400-800 cm) with maturity level of peat are Hemic/Sapric and Sapric/Hemic [15]. The kind of land use detected in peat land area in Sumatera consist of 25 types [15].

3. Methods

3.1. Data preprocessing

We performed pre-processing data which consists of two kind data there are hotspot and peat land data. Hotspot data pre-processing is decomposed into four steps. The first step of pre-processing hotspot data is selecting important attributes for the clustering process, there are latitude, longitude coordinate of the hotspot and date of hotspot occurrence. The second step of pre-processing data is selecting of hotspots in the non-peat land area and hotspot in the peat land area. The third step of pre-processing is selecting the distribution location of the hotspot in each district or city. The final step is loading information into the database. Pre-processing of peat land data is calculating the areal per kilometer square based on the physical characteristic information of peat, including the maturity level of peat, the thickness of peat land, and land use of peat.

3.2. Kulldorf's scan statistic method

Scan statistic can detect increases of cases in spatial and temporal dimensions, detect the existence of cluster in a certain region and detect the precise position the cluster [19]. KSS method is the extension of scan statistic were used to detect and evaluate the clusters of cases in spatial and temporal dimension. Spatial and temporal cluster detection is a more accurate technique compared to purely spatial scan, as it assesses the two dimensions simultaneously [20]. KSS uses large collections of overlapping scan windows to detect clusters, both the location and the size, and evaluate their significance [9]. KSS use cylindrical windows with a circle indicating a geographic base and with height representing to time was used in the space-time scan statistic [20]. For each location and size of the scanning window, the alternative hypothesis was that there was an elevated risk within the window compared to outside. Multiple different window sizes are used and the variable window size scan statistic described by [9]. The expected count is still based on the full size of the widow, when the scanning window extends outside the study area. The p-values are automatically adjusted for these boundary effects because KSS use analysis based on Monte Carlo randomizations [20]. Initial analysis was conducted including maximum spatial cluster size of 50% of the total number of cases in the temporal window [20].

The test statistic is determined as the maximum likelihood ratio over all windows. Evaluating the statistical significance is calculated by generating a large number of random datasets under the null hypothesis of no clustering. Calculating the value of the test statistic for each of those datasets. The window with the maximum likelihood was the most likely cluster. Secondary clusters were also identified and ranked based on their likelihood ratio test statistic. KSS assigned p-values on the basis of 999 Monte Carlo replications [9].

3.3. Research tools

This work employs several GIS based tools. The tools are Quantum GIS Dufour 2.0.1, PostGIS 2.1.8, PostgreSQL 9.2 (64-bit), Leaflet Web Maps, Apache 2.4, and R Studio 0.99.484. Quantum GIS 2.0.1 Dufour for spatial data visualization, PostGIS for spatial analysis, PostgreSQL 9.2 for database management system, Leaflet Web Maps for converting QGIS packages into interactive web maps, Apache 2.2.14 as the web server and R Studio 0.99.484 for clustering process. In the implementation of this research, we utilized clustering packages of data in R.We utilized 'spatialepi' R package developed by [21] to cluster dataset.

4. Result and Discussion

4.1. Distribution of hotspot in the peat land area in Sumatra between 2001 until 2014

The number of hotspots in the peat land area in Sumatera between 2001 until 2014 increase from year by year. The highest frequency held in 2006 and 2014. In 2006, the frequency of hotspot amounted 18,852 and amounted 26,193 hotspots in 2014. In 2006, hotspot were highly occur with the number of hotspots increased by the beginning of July 2006 [22]. In 2014, reach the peak number of hotpot between period 2001 until 2014. The frequency number

of hotspot in 2014 especially in the peat land area increase up to 71.97% compared the total frequency of hotspot in 2006. Fires occur on Indonesian peat lands in the dry season and, unsurprisingly, are worse in drier years. The Indonesian climate is strongly influenced by the El Niño Southern Oscillation (ENSO): in El Niño years, dry-season rainfall can be less than half of normal and severe El Niño events have long been associated with fire (e.g. 1972-73, 1982-83, 1987, 1991-92, 1994, 1997-98, 2002, 2006) [13]. El Niño years, of which 2006 and 2014 has a high probability of becoming, are characterised by periods of drought which increase the extent and longevity of fires. El Niño weather phenomenon also threatened to make 2014 a drier-than-normal year throughout much of Indonesia and Southeast Asia, raising the risk of fires across the country [23]. El Niño stirs up more fires in Indonesia especially in 2006 and 2014. Besides of El Niño effect, forest fires in Indonesia are mostly caused by humans (community people and companies), following slash and burn methods to clear land for plantations [22].

In the span of 14 years i.e., 2001-2014 the average frequency of hotspots in non- peat land area higher than the frequency of the hotspots in the peat land area. The average number of hotspots for 14 years in non- peat land area are 11,337 hotspots while in the peat land area are 10 325 hotspot. The number of distribution hotspots in peat land and non- peat land area from 2001 until 2014 in Sumatra can be shown in Fig. 2. The highest density of hotspot based on the maturity level of peat in Sumatera between 2001 until 2014 located at the maturity level of 'Hemic/Sapric (60/40), very deep' with average density value 4,001/km2. The highest density based on the thickness of peat land in Sumatera between 2001 until 2014 dominated by very deep thickness which reached 2.34/km2. Research conducted by Ekadinata et al. [11] indicated similar result. There was odds ratio indicates that peat soils of 4-8 m deep are twice as likely to be a hot spot [11], while shallow peat has virtually no hotspots. There are 25 land use types in peat land area. The highest density based on the land use of peat land in Sumatera between 2001 until 2014 dominated by shrub on former rice fields which reached 2.34/km2.

2001 2002 2003 20W 200 S 2006 2007 200S 2009 2010 2011 2012 2013 2014

■ Nou-peathnd ^Peailaad

Fig. 2. Number of hotspots in the area of peat land and non- peat land in Sumatera between 2001 until 2014.

4.2. Cluster distribution of hotspot in Sumatera

Hotspot spread in 53 districts in Sumatera with total areas of peat land distribution reached 78,673.53 km2. The period of time that we use in this research is the period 2001 to 2006 and 2007 to 2014. In the period 2001 to 2006 there are 64,776 peat land hotspots that detected in Sumatera. Locations which have the most distribution of hotspot during the period 2001 to 2006 are Riau province (Rokan Hilir, Bengkalis, Siak, and Indragiri Hilir district), South Sumatra province (Ogan Komering Ilir district), and Jambi province (Muaro Jambi district). While in the period 2007 to 2014 there are 79,779 peat land hotspots that detected. Location which have the most distribution of hotspot during the period 2007 to 2014 are Riau province (Rokan Hilir, Bengkalis, Siak, Indragiri Hilir Dumai, and Pelalawan district), South Sumatra province (Ogan Komering Ilir district), and Jambi province (Muaro Jambi district). Clustering hotspots with KSS method in peat land area in Sumatera can detect where the occurrence of cluster hotspots, when cluster hotspot held, and the geographical size of cluster hotspots. KSS clusters and measure

their significance via Monte Carlo replication. Results showed that the method is reliable to detect the clusters of hotspots which have the accuracy of 95%. Table 1 shows cluster distribution of hotspot between 2001 until 2006 that consist of 10 clusters. As shows in Table 1 that most likely cluster (P) in Rokan Hilir district in Riau province. The cluster is resulted from scanning windows that centred on the latitude and longitude coordinates (100.736996, 1.54702). The maximum cluster is defined 94.13 km with the scope of the cluster area covering Rokan Hulu and Dumai district.

Table 1. Cluster distribution between 2001 until 2006.

Cluster District Province Longitude Latitude Radius LLR P-value

P Rokan Hilir Riau 100.736996 1.54702 94.13 3475.3 0.00000001

Rokan Hulu Riau 100.789814 1.021444 - -

Dumai Riau 101.583911 1.575051 - -

S1 Rokan Hilir Riau 100.736996 1.54702 58.7 2187 0.00000001

Rokan Hulu Riau 100.789814 1.021444 - -

S2 Ogan Komering Ilir South Sumatera 105.591389 -3.115678 0 1318.5 0.00000001

S3 Dumai Riau 101.583911 1.575051 0 1202.6 0.00000001

S4 Pesisir Selatan West Sumatera 100.998983 -2.373818 0 846.29 0.00000001

S5 Indragiri Hulu Riau 102.338812 -0.597247 0 435.55 0.00000001

S6 Asahan North Sumatera 99.756999 2.732256 0 185.38 0.00000001

S7 Bengkalis Riau 102.295207 1.178453 0 152.23 0.00000001

S8 Mandailing Natal North Sumatera 99.067844 0.904823 0 83.406 0.00000001

S9 Aceh Singkil Nad 97.843686 2.590035 0 11.425 0.00021

*Note: P = Most likely clusters (primarily cluster) and S = Secondary clusters

Comparison research about clustering hotspot in Sumatera's peat land area are conducted by Usman [10] using DBSCAN method in periode 2002 and 2013. The number of hotspot clusters in 2002 amounted 53 cluster. The highest frequency of hotspot cluster are on Riau province and South Sumatera province. In Riau province are in Dumai, Bengkalis, Siak, Rokan Hilir, Pelalawan and South Sumatera province are in Ogan Komering Ilir, Banyuasin, and Tulang Bawang (Lampung) district. The results of research using KSS methods (period 2001-2006) and the results of comparison research using DBSCAN (period 2002) [10] having almost the same result.

Fig. 3 shows cluster distribution between 2001 until 2006 that consist of 10 cluster where 1 cluster as most likely cluster and the others is secondary cluster. Each cluster sorted based on the value of the ratio of probabilities where most likely clusters are clusters with the highest possibility of p-value. The location that has most potential cluster is in Riau province, especially in the district of Bengkalis and Rokan Hilir. The location of secondary cluster that has highest density of cluster hotspot is in Ogan Komering Ilir district, South Sumatra province. Other areas including Rokan Hulu, Asahan, and Mandailing Natal district have small case, but detected in the area of the cluster. This is because those areas are affected by the highest neighbourhood area that has the numerous number of cases.

Table 2 shows cluster distribution of hotspot between 2007 until 2014. As shows in Table 2 that most likely cluster in Rokan Hilir district in Riau province. The cluster is resulted from scanning a window centred on the latitude and longitude coordinates (100.736996, 1.54702). The maximum cluster is defined 94.13 km with the scope of the cluster area covering Rokan Hulu and Dumai district. The p-value for this cluster is 0.00000001 that was obtained using the 999 Monte Carlo simulations.

Fig. 3. Cluster distribution of hotspots in peat land area in Sumatra between 2001 until 2006.

Table 2. Cluster distribution between 2007 until 2014.

Cluster District Province Longitude Latitude Radius LLR P-value

Rokan Hilir Riau 100.736996 1.54702 94.13 10401.159 0.001

P Rokan Hulu Riau 100.789814 1.021444 - -

Dumai Riau 101.583911 1.575051 - -

S1 Aceh Barat Aceh 96.111492 4.29327 0 1445.132 0.00000001

S2 Ogan Komering Ilir South Sumatera 105.591389 -3.11568 0 516.162 0.00000001

S3 Bengkalis Riau 102.295207 1.178453 0 389.528 0.00000001

S4 Indragiri Hulu Riau 102.338812 -0.59725 0 247.103 0.00000001

S5 Tapanuli Selatan North Sumatera 98.919429 1.404741 0 108.895 0.00000001

S6 Pesisir Selatan West Sumatera 100.998983 -2.37382 0 47.469 0.00000001

S7 Muara Enim South Sumatera 104.192212 -3.19618 0 17.619 0.00000062

S8 Aceh Singkil Aceh 97.843686 2.590035 0 16.988 0.0000011

*Note: P = Most likely clusters (primarily cluster) and S = Secondary clusters

Fig. 4 shows a cluster distribution on period 2007 to 2014. There are consist of 9 cluster where 1 cluster as most likely cluster and 9 cluster as secondary cluster. The location that has most potential cluster is in Riau province, especially in the district of Rokan Hilir. Ogan Komering Ilir district, South Sumatra province is the location that has the highest density in secondary cluster. Comparison research by Usman [10] obtained the number of clusters

amounted 108 cluster. The highest frequency of hotspot cluster are on Rokan Hilir, Bengkalis, Dumai, Siak, Pelalawan, Rokan Hulu, Kampar, Indragiri Hilir, Indragiri Hulu, and Pekanbaru. The results of research using KSS methods (period 2007-2014) and the results of comparison research using DBSCAN (period 2013) [10] having almost the same result.

Fig. 4. Cluster distribution of hotspots in peat land area in Sumatra between 2007 until 2014.

Fig. 5 shows about the percentage comparison of hotspot cluster based on the type of peat land in Sumatera. The distribution clusters of hotspot between of 2001 until 2006 are dominated by 'Hemic/Sapric (60/40), very deep' and 'Hemic/mineral (90/10), moderate'. During 2007 until 2014, there is the same type of peat land as in 2001 until 20106 which dominated by 'Hemic/Sapric (60/40), very deep' and 'Hemic/mineral (90/10), moderate'. But, in the period 2007-2014 the intensity of hotspot increased on the type of 'Hemic/Sapric (60/40), very deep' and 'Sapric/Hemic (60/40), deep'.

Fig. 6 shows the percentage comparison of hotspot cluster based on the thickness of peat land in Sumatera. The distribution clusters of hotspot in the period of 2001-2006 are dominated by 'very deep thickness (D4) (400-800 cm)' and 'moderate thickness (D2) (100-200 cm)'. While period 2007-2014 dominated by 'very deep thickness (D4) (400-800 cm)' and 'deep thickness (D3) (200-400 cm)'. Many hotspots are found in peat land with the maturity level of 'hemic' and the thickness level of 'moderate'. When 'hemic' maturity levels of peat dominated on the outskirts of peat domes indicated the damage of the peat dome very high. Combined with the drainage to overcome dryness of peat it makes excretion of water that is in the dome of peat. If the water at the dome of peat is not exist, hence peat experienced leakage so function hydrology peat being broken [2].

Fig. 5. Clusters distribution of hotspots based on the peat land maturity level of peat period 2001-2006 and 2007-2014.

Fig. 6. Clusters distribution of hotspots based on the thickness of peat period 2001-2006 and 2007-2014.

There has been a shift distribution hot spots and the use of peat from 'moderate depth' and 'very deep' peat in the 2001-2006, toward to 'very deep' and 'deep' peat in 2007-2014. A shift in the distribution of hotspots and the use of peat from the 'moderate' depth to 'very deep' and 'deep' for this 14 years need special attention. As required by an Indonesian government regulation number 32 years 1990, peat land area that have 'deep' and 'very deep' thickness (>300 cm) should not be opened for agriculture development. Based on Indonesian government regulation number 32 years 1990, peat land area that have 'deep' and 'very deep' thickness (>300 cm) should not be opened for agriculture development. Based on RTRWN that in line with constitution number 21 years 1992 regarding areas planning (UUTR), the protection against the peat should be done to control hydrological areas, serves as a fastening water, flood prevention, and protecting the ecosystem that is typical of the area [24]. Fig. 7 shows the cluster distribution of hotspot based on the land use of peat land in Sumatera. In times of dry season, the swamp forests area prone to wildfires and the damage caused by forest fire in the swamp forest aims serious negative impact. The result equivalent with the comparison research conducted by Usman [10] that hotspot distribution in 2002 and 2013 dominated in swamp forest.

- Bog shrub Bush, Lawn on former rice fields

* Coconut on a former swamp forests > 5 year

• Intensive rice fields (rice crops, fallow), orange îe Logging concessions land

s Open land for farm = Paddy fields and coconut 11 Palm oil on a former swamp forests > 5 year l Palm oil on a forms swamp forests < 5 yeai

♦ Planting lathi an industrial plant

* Rainfed rice (rice crops, fallow) Ricefidd. ebbtide

.Rubber

fir Shrubs, marsh grasses arid former fire Swamp forests

Fig. 7. Clusters distribution of hotspots based on the land use of peat period 2001-2006 and 2007-2014.

5. Conclusion

Clustering hotspot in the peat land areas in Sumatera between 2001 until 2014 with the KSS method discovers patterns of hotspot clusters distribution. The study conclude that the spatial and temporal pattern of hotspot is clustered. Experiments of clustering on the dataset hotspot in periods 2001-2006 resulted in 10 clusters hotspots and clustering on the dataset hotspot in periods 2007-2014 resulted in 9 clusters hotspot. The study revealed that in periods 2001-2014, the highest density of hotspot found in Riau province and South Sumatera province, especially in Dumai, Bengkalis, Rokan HIlir and Ogan Komering Ilir district. There were changes in the pattern of distribution of hotspot and land cover of peat land from 2001 to 2014. In period 2001-2006 the distribution of hotspot are mostly found in moderate (100-200 cm) and very deep (> 400 cm) thickness, whereas in period 2007-2014 the distribution of hotspot were mostly found in deep (200-400 cm) and very deep (> 400 cm) thickness of peat land area. Land covers of peat land are dominated by 'swamp forest'. Based on the physical characteristics of peat, hotspot clusters are found in peat land level of maturity 'hemic' and land use type of 'swamp forest'.

References

1. Harrison ME, Page SE, Limin SH. The global impact of Indonesian forest fires. Biologist 2009;6(3):156-63.

2. Langner A, Siegert F. Spatiotemporal fire occurrence in Borneo over a period of 10 years. Global Change Biology 2009;15:48-62.

3. Page SE, Siegert F, Rieley JO, Boehm HV, Jayak A, Limin SH. The amount of carbon released from peat and forest fires in Indonesia during 1997. Nature 2002;420:61-5.

4. Syaufina L. Kebakaran hutan dan lahan di Indonesia. Malang: Bayumedia Publishing; 2008. In Bahasa.

5. Sitanggang IS, Yaakob R, Mustapha N, Ainuddin AN. Classification model for hotspot occurrences using spatial decision tree algorithm. Journal of Computer Science. Science Publications 2013;9(2):244-51.

6. Nurpratami ID, Sitanggang IS. Classification rules for hotspot occurrences using spatial entropy based decision tree algorithm. A paper published in the Proceedings of the 1st International Symposium on LAPAN-IPB Satellite for Food Security and Environmental Monitoring. Procedia Environmental Sciences 2015;24:120-6.

7. Sitanggang IS, Yaakob R, Mustapha N, Ainuddin AN. A decision tree based on spatial relationships for predicting hotspots in peat lands. TELKOMNIKA (Telecommunication Computing Electronics and Control) 2014;12(2):511-8.

8. Sitanggang IS, Yaakob R, Mustapha N, Ainuddin AN. Burn area processing to generate false alarm data for hotspot prediction models. TELKOMNIKA (Telecommunication Computing Electronics and Control) 2015;13(3):1037-46.

9. Kulldorff M. A spatial scan statistic. Communications in Statistics: Theory and Methods 1997;26:1481-96.

10. Usman M, Sitanggang IS, Syaufina L. Hotspot distribution analyses based on peat characteristics using density-based spatial clustering. A paper published in the Proceedings of the 1st International Symposium on LAPAN-IPB Satellite for Food Security and Environmental Monitoring. Procedia Environmental Sciences 2015;24:132-40.

11. Ekadinata S, Van NM, Budidarsono S, Dewi S. Hotspots in Riau, haze in Singapore: the June 2013 event analyzed. ASB Policybrief No. 33. Nairobi: ASB Partnership for the Tropical Forest Margins. 6p; 2013.

12. Wahyunto, Ai Dariah, ahmuddin Agus. Distribution, Properties, and Carbon Stock of Indonesian peat land. Proceeding of Int. Workshop on Evaluation and Sustainable Management of Soil Carbon Sequestration in Asian Countries. Bogor, Indonesia Sept. 28-29; 2010.

13. Mark EH, Susan E, Suwido HL. The global impact of Indonesian forest fires. Biologist 2009;56(3).

14. Indonesia Statistics. Hasil sensus penduduk 2010 dalam agregatperprovinsi. Jakarta: Indonesia Statistics; 2010. In Bahasa.

15. Wahyunto, Ritung S, Suparto, Subagjo H. Sebaran gambut dan kandungan karbon di Sumatera dan Kalimantan. Bogor: Wetlands International IP; 2005. In Bahasa.

16. NASA FIRMS. MODIS active fire detections data set. Available on-line http://earthdata.nasa.gov/data/nrt-data/firms; 2015.

17. Korontzi S, McCarty J, Loboda T, Kumar S, Justice C. Global distribution of agricultural fires in croplands from 3 years of moderate resolution imaging spectroradiometer (MODIS) data. GlobalBiogeochem Cycles 2015;20:GB2021.

18. Giglio L, Csiszar I, Justice CO. Global distribution and seasonality of active fires as observed with the terra and aqua MODIS sensors. J. Geophys Res 2006;111:G02016.

19. Cao CX, Xu M, Chen JQ, Wu YS, Xie X. Space-time scan statistic based early warning of H1N1 influenza a in Shenzhen, China. International Archives of the Photogrammetry, Remote Sensing and Spatial Innformation Science 2010;28(8). Kyoto: Japan.

20. Kulldorff M. SatScan version 8.02: software for the spatial and space-time scan statistics. Silver Spring, MD: Information Management Services; 2009.

21. Kim AY, Wakefield J. R data and method for spatial epidemiology: the SpatialEpi Package. Dept of Statistics, University of Washington; 2010.

22. World Heal Organization. Sumatra and Kalimantan fire and haze. Emergency Situation Report # 1 29 August; 2006.

23. Greenpeace Southeast Asia. Sumatra: going up in smoke. Jakarta: Indonesia; 2014.

24. Agus F, Hairiah K, Mulyani A. Measuring carbon stock in peat soil. World Agroforestry Centre and Indonesian Centre for Agricultural Land Resources Research and Development. Bogor: World Agroforestry Centre (ICRAF); 2011.