Investigating forest fragmentation through earth observation datasets and metric analysis in the tropical rainforest area

Extensive mining operations, deforestation, jhumming, and soil erosion coupled with population stress in the study area have put an adverse effect on its forest resources. This study investigates the transition in forest cover classes and its fragmentation in the Jaiñtia Hills District of Meghalaya (India). Satellite data (multispectral images from Landsat 5 and 8) for 1995, 2001, 2007, and 2015 were classified using the supervised classification method. Landscape metrics from the classified images were calculated using FRAGSTATS. The overall accuracy of classification was found to be 87.50% (1995), 87.50% (2001), 85.00% (2007) and 91.67% (2015), respectively. The results revealed an increase in dense forest with an increase in the patch number from 1995 to 2007. Additionally, a decrease in non-forest cover with an increase in the number of patches from 2001 to 2015 was observed which further suggests fragmentation. It has been reported that 8.13% of the dense forest increased and 19.47% of non-forested areas decreased during the study period. Overall, this study highlights the changes in the distribution of forest area which could aid policy makers to adopt appropriate forest conservation strategies.


Introduction
Fragmentation can be defined as a division of the landscape into smaller isolated patches which decrease the natural habitat in a landscape. Loggers, commercial cultivators, settlement planners, infrastructure developers, and expansion in the population are some of the destructive trends maximizing forest fragmentation at an alarming rate [1]. These aforementioned factors severely expose the forested area leading to vulnerability of wildlife species as well as disturbing the entire forest ecosystem.
Mining operations can also contribute to forest fragmentation which is apparent in many regions of India including the state of Meghalaya. Meghalaya possesses huge reserves of various minerals including coal, limestone, kaolin, clay, granite, glass-sand, uranium, etc. Overexploitation of such resources, e.g. extensive coal mining [2] has led to a drastic change in the land use/land cover (LULC) of the state.
The district of Jaiñtia Hills is among the most adversely affected area of the state of Meghalaya in regard to forest fragmentation resulting from coal mining. Extraction of coal is carried out by the primitive method known as rat hole mining. Major coal-bearing areas in this district are Sutnga, Shkentalang, Sohkymphor, Lakadong, Ladrymbai, Bapung, Khliehriat, Musiang Lamare, Ïooksi, and Jaraiñ [3].
Use of technology along with appropriate policies is need of the hour for assessing, inventorying, and restricting over-exploitation of minerals towards conserving the forest ecosystem. In the last two decades, integration of remote sensing (RS) data with geographic information systems (GIS) made noteworthy contributions towards the assessment of spatial-temporal patterns and processes of forest ecosystems in criteria-based decision-making and selection of the optimal alternative [4,5]. Numerous researchers have developed metrics to measure multiple aspects of landscape patterns [6][7][8][9][10][11][12]. FRAGSTATS 4.2.1 has emerged as one of the most promising tools for performing spatial analysis to compute the disturbance index [12].
Considering the aforementioned issues with regard to forest fragmentation particularly in the district of Jaiñtia Hills, this study aims to analyse the spatial and temporal pattern of forest fragmentation during the past two decades using FRAGSTATS model along with RS and GIS techniques.

Study area
The study area (district of Jaiñtia Hills) consists of the eastern part of Meghalaya with a geographic area of 3819 km 2 . It is situated between east longitude 91.59° and 92.45° and between north latitude 25.3° and 25.45° (Fig. 1). It covers around 17% of the total area of Meghalaya State. The elevation of the district ranges between 1050 m and 1350 m. Jaiñtia Hills has a comparably flatter topography with a mild gradient. The district has a forest area of 1540.6 km 2 , i.e. about 40% of the total geographic area [13].
Study area is enclosed on the north and east by the state of Assam, on the south by Bangladesh, and on the west by the East Khasi Hills District of Meghalaya. Jaiñtia Hills District is divided into five blocks consisting of Amlarem, Thadlaskeiñ, Laskeiñ, Khliehriat, and Saipung [14]. Amlarem is the smallest block in the district with a population of 43,844, whereas Thadlaskeiñ is the biggest block with a population of 137,939 [15]. Myntdu is one of the major water bodies of the study area (Appendix 1). The drainage pattern of the study area is sub-parallel to parallel. It is being controlled by joints and faults as indicated by the straight courses of the rivers and streams with deep gorges.

Data used and methodology
Satellite images from Landsat 5 and 8 for the period 1995-2015 were used towards spatio-temporal analysis of Jaiñtia Hills District, Meghalaya. The study area is covered by path 136 and row 42/43 of Landsat images. Details of the satellite images utilized in the present study are provided in Appendix 2. The Landsat images utilized in the present study were acquired for the month of March to minimize the significant exposure of the sun angle on the northern slope. Top of atmospheric (ToA) calibration was performed to obtain the reflectance values on a scale of 0 to 1. Many studies have used remote sensing data based on classification algorithms for the extraction of forest land cover information [16,17]. Satellite images were classified using the supervised classification method with a maximum likelihood algorithm [18][19][20][21]. Major land cover classes of the study area were categorized into dense forest, open forest, scrub forest, and non-forest (the wastelands, built-up, croplands, and water bodies were clubbed in the non-forest category). After that, post-classification techniques were used which help in removing noises and improve the quality of the classified image. It was achieved by sieve, clump, elimination, and majority filter tools which has enhanced the output image quality before comparing two different time period images. A description of various land cover classes is provided in Appendix 3.
Change analysis was performed to assess changes in the past decade of 1995-2015. A colour coded scheme was adopted to delineate positive change with green colour (non-forest to forest change) as well as negative change with red colour (forest to non-forest change). A change matrix for different land cover classes was calculated. The positive and negative changes were also identified at block-level in the study area. Lastly, the data were processed using FRAGSTATS 4.2.1 software to analyse the spatial metrics and comprehend the fragmentation of various land covers. The methodology followed in the present study is summarized in Fig. 2.

Computation of landscape metrics
Landscape ecology is the interaction between various elements of a landscape, and how these patterns/interactions change over time [22][23][24][25]. Landscape metrics explores site variability and the effects of fragmentation. Several studies have shown that landscape metrics have the potential for analysing the spatial arrangement of LULC and monitoring the spatio-temporal changes [26][27][28][29][30][31][32]. However, a selection of appropriate key approaches must be done in  35,46]. NP indicates the rate of loss denoting the information about the area, shape, or distribution of the fragments. The density of patches is computed using PD metrics, ultimately indicating the level of fragmentation. IJI is used for identifying the intermixing of different patch types irregularly.

Accuracy assessment
Accuracy assessment of land cover maps (1995, 2001, 2007, and 2015) was performed to obtain user accuracy, producer accuracy, overall classification accuracy, and kappa statistics using equalized random sampling strategy [47]. Additionally, 50 ground control points were utilized for comparing the accuracy of land cover maps. The overall classification accuracy was found to be 87. 50

Spatial extent of forest cover classes (1995-2015)
Forest cover change throughout the study period is presented in Table 2 and Fig. 3, respectively. Based on   19.45% of the area lost its forest cover (Fig. 3).  (Table 2). Table 2 summarizes the changes in forest cover. Increase in dense forest apparent between 1995 and 2007 can be attributed to the intensification of commercial plantations such as Areca nut (Areca catechu), bamboo (Bambusa sp.), banana (Musa paradisiaca), black pepper (Piper nigrum), and canes (Calamus sp.) cultivated with Acquilaria [48]. Betel leaf is an important cash crop in Meghalaya with high demands across local, national and international markets which are planted in lower altitude, i.e. low-lying area of Khliehriat block near Bangladesh border. Secondly, it may be due to plantation via afforestation programmes under the social forestry division. Three forest divisions of Khasi, Jaiñtia, and Garo Hills were under programmes such as social forestry where the forest department carried out plantation in more than 1645 km 2 area [49]. Although [50] reported a decrease in a dense forest in the study area, the result obtained for this area is same as statistics

Matrix of land cover change
The square transition matrix gives insights into the degree of change of classes when the two images acquired in different years are directly compared to assess increase and decrease in classes. In this study, a change matrix table was generated by the intersection of two datasets viz., 1995-2001, 2001-2007, 2007-2015, and 1995-2015, respectively. The results were summarized using tables and figures that aided in the comparison and interpretation of the results for the area and its block.

LULC changes (1995-2001)
Land cover change matrix using the classified data for the years 1995 and 2001 is presented in  Fig. 4.

LULC changes (2001-2007)
The land cover change matrix was analysed using the classified thematic output for the years 2001 and 2007. It is apparent that out of the total dense forest area of 1269. 64

LULC changes (2007-2015)
The land cover change matrix was analysed using the classified thematic output for the period 2007-2015.
It is apparent that out of the total dense forest area of 1327.68, 779.88 km 2 (73.68%) of the dense forest was undisturbed (

Change area-block-wise
The block-wise changes in area are summarized in Table 7.
Change from forest to non-forest area was considered as a negative change and non-forest to forest area as a positive change (Fig. 8) [58]. Landscape shape index (LSI) is also one of the indices that quantify the complexity of the landscape which measures the total edge that adjusts for the landscape [59]. LSI also indicated the same trend . Additionally, a similar trend of positive change during the study period was also apparent in Laskeiñ block which can be attributed to horticultural plantations such as oranges around the Raliang area. It is widely adopted by the farmers in the Jaiñtia Hills of Meghalaya, where hill slopes are quite steep with low soil depth. Overall, the highest negative change for the period 1995-2015 was apparent in Khliehriat block can be attributed to the large scale of limestone mining and production activity for cement factories in Jaiñtia Hills.

Quantification of the spatial pattern of forest fragmentation
It has been observed that from 1995 to 2007 the number of patches (NP) and patch density (PD) has increased in the dense forest category. The number of patches was observed to be 12600, 17032, and 17950, and patch density was observed to be 1.87, 2.53, and 2.67, respectively, for the      The dense forest and non-forest category are reported to have experienced major disturbances which led to fragmentation. Assessment of forest canopy density and forest fragmentation has been adopted by various studies for analysing the status of forest conditions [65,66]. Despite of apparent fragmentation, the increase in forest cover can be attributed to regeneration, afforestation, or secondary forest succession [67,68]. The study area is rich in mineral resources, high population of rural and tribal communities which has a high dependence on forest resources for their livelihood, enterprise, and subsistence [69]. Such dependency on forest resources could ultimately lead to more fragmentation in the study area.
It was reported from 1994 to 2014 that the community forests and the reserved forest were more exploited as compared to state forests [70]. India State Forest Report 2015, outline that the forest cover of India has increased by 5081 km 2 with 21.34% for moderate dense (21.34%) whereas 2510 km 2 of the very dense forest has been lost [52]. An increase in forest fragmentation is due to anthropogenic activities [71,72], it is due to the cultivation of paddy, minor fuelwood, and shifting cultivation for turmeric cultivation [73]. The area reported 40% of rice is cultivated in jhum sites [74] which is detrimental to forests, soil, and biodiversity [75][76]. Moreover, the mining activities are high in the area which ultimately affects the common herbal remedies used by the Jaiñtia tribe community namely Litsea khasiana, Aegle marmelos, Averrhoa carambola, Gaultheria

Conclusion
The study reported that dense forest and open forest increased by 8.13% and 11.14%, respectively, however, 19.47% of non-forests has decreased during the study period. Specifically, dense forest areas have become more degraded due to anthropogenic activities in this area. Landscape fragmentation has a great impact on the ecological system, e.g. when natural forests areas are confined into smaller size this ultimately disconnects the ecological corridors and results in loss of biodiversity. An increase in the dense forest is apparent from the temporal land cover change analysis; however, a significant amount of fragmentation has also been observed. Similarly, over the years fragmentation was also observed in the non-forest category accounting for 19.45% of its spatial extent. This region of interest implies that it has been experiencing land cover changes with more activity resulting in fragmentation. The landscape of Jaiñtia Hills is under pressure due to the local population which is highly dependent on agriculture, coal, and cement industry. The study suggests to curb the intensification of forest cover lost in the study area. The conclusion drawn from this study is that the pattern of forest cover change and forest fragmentation is unpredictable which may be due to the complexity of the landscape structure. Hence, it is crucial to understand the pattern of land cover dynamics and its fragmentation to sustain biological diversity. The outcome will help in redefining ecological zones to maintain the overall spatial composition and configuration of the habitat.

Conflict of interest
The authors declare that they have no conflict of interest.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.

Appendix 3
See Table 11.  Tree species with canopy over more than 40% Open forest Tree species with canopy over more than 10% and less than 40% Scrub forest Degraded forest land with canopy less than 10% Non-forest Wasteland, built-up, cropland, water bodies

Metrics level Description
Patch level metrics Size, shape, and distance to neighbouring patches of an individual patch area quantified at the patch level Class level metrics Use these values for all the patches in the same LULC type to give a value for the entire class in the landscape.

Landscape metrics
Provide unique values without reference to individual patches or classes as they aggregate the properties for all the patches in the landscape

No unit
Interspersion juxtaposition index IJI is based on patch adjacency. It is the arrangement of patches on the basis of composition and configuration in the landscape.

No unit
Effective mesh size MESH measures the proportional area of each patch based on total landscape including background. ha