You are on page 1of 18

Journal of Cleaner Production 356 (2022) 131763

Contents lists available at ScienceDirect

Journal of Cleaner Production


journal homepage: www.elsevier.com/locate/jclepro

Assessment of forest cover loss and impacts on ecosystem services: Coupling


of remote sensing data and people’s perception in the dry deciduous forest
of West Bengal, India
Dipankar Bera *, Nilanjana Das Chatterjee , Subrata Ghosh , Santanu Dinda , Sudip Bera ,
Mrinmay Mandal
Department of Geography, Vidyasagar University, Midnapore, 721102, West Bengal, India

A R T I C L E I N F O A B S T R A C T

Handling Editor: Bin Chen Forest loss is one of the major environmental issues and is threatening livelihoods across the world. Under­
standing the drivers of forest cover loss in the different forest cover densities and its impacts on ecosystem
Keywords: services are relevant for sustainable management of the forest ecosystem. Extensive studies exploring topics
Forest loss related to spatial-temporal deforestation, little is known regarding quantitative forest loss by individual distur­
Forest disturbance drivers
bance driver and people’s perceptions of forest loss’s impacts on provisioning ecosystem services. The present
Ecosystem services
study focused on trends and patterns of forest loss by forest disturbance drivers in the different forest cover
Google Earth Engine
Landsat densities and its impacts on provisioning ecosystem services integrated with remote sensing data and local
Dry deciduous forest people’s perceptions in the dry deciduous forest of West Bengal. This study employs stratified probability
Collect Earth Online sampling approach using time series Landsat data on Google Earth Engine platform to quantify forest loss be­
tween 2006 and 2020 in the different pre-disturbance forest cover densities by forest disturbance drivers.
Structured household interviews were used to examine local people’s perceptions about forest cover loss and its
impacts on ecosystem services. We found that 68.463% of forest area is cleared by disturbance drivers between
2006 and 2020. Annual rate of forest clearing is increasing trend and dominating in dense forests, followed by
medium forests and open forests. Logging and agricultural clearing was the most dominant drivers over space
and time. Overall 73.92% of households perceived decline in the availability of provisioning ecosystem services.
Wild foods and medicinal plants were the most affected ecosystem services followed by livestock feed, con­
struction materials and fuel wood. This study may help to the regional assessment of gross forest loss by indi­
vidual disturbance driver, and its impacts on provisioning ecosystem services for supporting decision making to
sustainable forest management.

1. Introduction pasture land, the harvest of industrial wood, the harvest of fuel wood,
expansion of agriculture, draining and burning of peatland, tree plan­
The loss of forest cover is one of the major anthropogenic factors tation and fires (Houghton, 2012). Agricultural expansion is the main
responsible for the deterioration in ecosystem services (Balthazar et al., cause of forest cover loss in tropical countries (Kanninen et al., 2007)
2015; Fang et al., 2021) and carbon sequestration capacity (Le Quéré followed by infrastructure development and wood extraction (Geist and
et al., 2009; Pan et al., 2011). Though deforestation rates at the global Lambin, 2002). But drivers of forest cover loss vary from global to local
level have decreased in the past decade, there are numerous localities scale and all drivers do not have equal impacts (Curtis et al., 2018).
and nations where deforestation rates continue to increase. (FAO, 2015). Associated drivers interact in critical and complex ways and making our
Reduction of forest cover is larger mainly in tropical developing coun­ understanding incomplete at the local level (Brown and Schreckenberg,
tries due to heavy dependence on the forest recourses for livelihoods 1998). Therefore, understanding the drivers requires not only broader
(Hosonuma et al., 2012). In developing countries, tropical forests are scale (Lambin et al., 2003), also local scale quantitative assessment of
exploited for different purposes such as shifting cultivation, expansion of forest cover loss by disturbance drivers provide complete things within a

* Corresponding author.
E-mail addresses: rsge_dipberageo@mail.vidyasagar.ac.in (D. Bera), nilanjana_vu@mail.vidyasagar.ac.in (N.D. Chatterjee).

https://doi.org/10.1016/j.jclepro.2022.131763
Received 14 August 2021; Received in revised form 4 January 2022; Accepted 12 April 2022
Available online 20 April 2022
0959-6526/© 2022 Elsevier Ltd. All rights reserved.
D. Bera et al. Journal of Cleaner Production 356 (2022) 131763

landscape which help to take a decision for sustainable management forest loss by forest disturbance drivers in different forest cover densities
(Shriar, 2014). and its impacts on provisioning ecosystem services integrated with
Forest cover loss impacts on the ecosystem services, namely provi­ remote sensing data and local people’s perceptions in the dry deciduous
sioning, cultural, regulating and supporting services (Gouwakinnou forests. Historically forest has an important role in the livelihoods in this
et al., 2019). Among these services, provisioning ecosystem services region (Dinda et al., 2020), but enormous population pressure and un­
have direct impacts on the local scale in the livelihoods of rural people informed management increase deforestation in the recent period.
who directly depend on the forest for various services including, food, Limited availability of data on forest cover loss by associated drivers and
fodder, fuelwood, and non-timber forest products (Ellison et al., 2017; its impacts on ecosystem services in this region poses further challenges
Gray et al., 2015). The availability of provisioning ecosystem services to sustainable management. There have been no evidence-based studies
depends on the location, utility and exert capability of the societies in a undertaken to understand forest loss by associated drivers in the
given biogeographical and socio-economic context (Ngom et al., 2014). different forest cover densities, and how forest cover loss affects
Change in the availability of provisioning ecosystem services may create ecosystem services at the local level. Therefore, in this study we focused
vulnerable condition for the well-being (Millennium Ecosystem Assess­ on some specific questions: a) What are the major drivers of forest loss
ment, 2005). Information about ecosystem services are very significant across the southwestern region of West Bengal; b) How does the relative
in term of cleaner production and maintain the services in future for magnitude of these drivers vary between forest densities (dense, mod­
livelihoods (Liu et al., 2018). So, it is important to know how local erate, and open forests)?; c) what is the temporal trend of forest cover
people perceived the availability of provisioning ecosystem services loss in each pre-disturbance forest cover density and forest disturbance
with forest cover loss who are the main actors in forest cover loss and driver; d) what are the local people’s perceptions about forest cover and
direct beneficiaries of these services (Gouwakinnou et al., 2019). availability of provisioning ecosystem services? To achieve these ques­
Until recent studies, researchers have applied many different algo­ tions this study was designed to analyse forest cover loss using proba­
rithms to assess the forest cover change, such as Bayesian Updating of bility sampling approach, identify associated driving factors from
Land Cover (BULC) algorithm (Fortin et al., 2020), Classification And reference data and impacts on provisional ecosystem services using
Regression Trees (CART) algorithm (Potapov et al., 2012), Least Square people’s perception in the dry deciduous forest of West Bengal. The
Moving Average (LSMA) analysis (Mayes et al., 2015), Support Vector present research emphasized on the integrated implication of remote
Machine (SVM) algorithm, and Random forest (RF) algorithm (Senf sensing data to analyse forest cover loss using Google Earth Engine and
et al., 2020) of time series satellite data. Limited studies are based on local people’s perception to analyse impacts on ecosystem service of
probability sampling of reference data, which provide better accuracy forest cover loss for sustainable management.
rather than counting map pixels (Olofsson et al., 2014). Also, limited
studies based on quantitative assessment of forest cover loss by associ­ 2. Materials and methods
ated drivers (Tyukavina et al., 2017, 2018) at the national level, but
have not assessment at the local level in different forest cover densities 2.1. Description of the study area
with impacts of forest cover loss on the ecosystem services. Most of the
previous studies differentiate forests as primary forests, secondary for­ The study area is located south-western part of West Bengal.
ests, primary woodland and dry forests, and plantation, then measure Geographically, the locational extent of this area is 22◦ 10′ 15′′ N to 24◦
the forest loss in each forest type at the national level (Potapov et al., 09′ 54′′ N and 85◦ 49′ 35′′ E to 88◦ 14′ 07′′ E. This area covers the full part
2012; Tyukavina et al., 2018). But, distinguished forest loss by forest of Purulia, Bankura, and the maximum part of Birbhum, Bardhaman,
cover density (dense, medium, and open forests) is very important Jhargram, Paschim Medinipur and the few parts of Murshidabad, Nadia,
because carbon storage, biodiversity value and driving factors differ Hugli, Purba Medinipur and Haora district of West Bengal (Fig. 1). The
significantly from low canopy cover density to high canopy cover den­ study area covers approximately 3,201,000 ha and 36% of total
sity. Although, few studies have shown impacts of forest cover change geographical area of the state of West Bengal.
on livelihoods based on local people experience (Ehara et al., 2016; The forest type is typical dry deciduous forest, and principal species
Thanichanon et al., 2013). However, these studies lack explaining forest of this area are Shorea robusta, Anogeissus latifolia, Terminalia tomentosa,
cover loss by disturbance drivers and impacts of forest loss on provi­ Pterocarpus marsupium, Madhuka latifolia, Schleichera trijuga, Boswellia
sioning ecosystem services over the landscape with the integration of serrata, Terminalia belerica (WBFD, 2018). Topographically, this tropical
remote sensing and people’s perceptions. People’s perceptions can dry deciduous forest is a plateau fringe zone also the dominance of
minimize the gaps between satellite data and ground truth as well as scrubs and thorny bushes. The climate is tropical wet and dry (AW)
reflect community view toward the past forest cover and benefits of according to the Köppen-Geiger climate classification (Beck et al.,
availability of ecosystem services (Ahammad et al., 2019). 2018). The middle and western parts are received less rainfall (annual
In India, National Forest Inventory (NFI) provides essential data 1236–1288 mm) than north eastern part (annual 1287–1353 mm) and
about forest cover for formulating national forest policy, monitoring and south eastern part (annual 1436–1749 mm) (Climate and Weather Av­
forecasting (Tewari, 2016). In 2008, Forest Survey of India (FSI) erages in West Bengal, India, 1998). Dominated soil is lateritic soil with
launched a new study to measure the components of forest biomass that humus content in the surface layer but, somewhere old alluvial soil
are not measured by NFI, as required under REDD+. These studies dominating in the eastern part of the study area especially the track
provide spatial-temporal information of forest cover and biomass stock, bounded by the river.
but have not addressed the forest cover loss by individual disturbance
drivers and impacts on provisioning ecosystem services. Recently, the 2.2. Field observation
Joint Forest Management (JFM) policy has been promoting forest pro­
tection and small-scale afforestation. But, JFM has achieved limited Field observation is a pre-required thing to determine the direct
success in identifying the proper cause of forest cover loss, and its direct drivers of forest loss, and nature of forest cover as well as to understand
impacts on human beings. The global forest loss data are now available how people are interrelated with forests. A ground survey was con­
(Hansen et al., 2013; Kim et al., 2014), but lack of explaining local level ducted in the month of February 2021 in the different parts (most
forest cover loss and also lack of integrated assessment of forest cover affected forest loss areas from global forest loss map of Hansen et al.
loss by associated drivers and impacts of forest cover loss. Also, global (2013) between 2006 and 2020) of the study region to identify the forest
level all datasets may overestimate or underestimate the area of forest disturbance drivers. We visited different forest loss patches, and also
cover change at the local level (Tyukavina et al., 2018). interviews were conducted with neighbouring households, villagers,
Within this context, there is no study available for understanding the (who are knowledgeable about forest) and forest rangers in the study

2
D. Bera et al. Journal of Cleaner Production 356 (2022) 131763

Fig. 1. Location of the study area.

region. We asked to the respondents two question: 1) what is the cause of (Broich et al., 2009). Study area was stratified into three strata using
forest loss to your neighbourhood patches?; and 2) what are the major GEE platform (Fig. 2): (i) “Loss” stratum, any pixel within the study area
drivers of forest loss around your village and surroundings over the time where forest loss was determined from global forest loss map (Hansen
period? After that, disturbance drivers are categories into six types such et al., 2013) between 2006 and 2020; (ii) “Loss buffer” stratum, 60 m
as, logging, agriculture, roads, settlements, fires, and other (flooding, buffer around the loss pixel; (iii) “No loss” stratum, all other remaining
windfalls, droughts, and river meandering) which are distinguishable pixels within the study area without out “Loss” stratum and “Loss buffer”
and measurable using remote sensing. Livestock grazing, artisanal wood stratum. The total 2000 pixel were selected using “randomPoints”
marketing, timber and fuel collection are also present in the study re­ function in GEE, with 20% (400) of the sample randomly allocated to the
gion. But it is almost impossible to quantify the area affected by these “Loss” stratum, 30% (600) to the “Loss buffer” stratum, and 50% (1000)
drivers, and these drivers not resulting in significant canopy loss in an to the “No loss” stratum. The sampled pixel size was one Landsat pixel
area. (30 m by 30 m), and the mean pixel area was 823.668 sq. m. Distribution
of sampled pixels among the pre-disturbance forest cover densities is
2.3. Sampling design shown in Table S1 and Fig. 3A. Total number of pixels by
pre-disturbance forest cover density reported in Table S1, was estimated
The current study uses stratified probability sampling approach to using following equation
estimate gross forest loss by associated drivers, with Hansen et al. (2013)

K
global forest loss map use to stratify the region to improve the accuracy Nh =
Ni
nhi (1)
of the estimation. A stratified random sampling method provides better i=1
ni
accuracy compared to other sampling designs for forest loss estimation

Fig. 2. Sampling strata of the study area. “Loss” stratum: pixel within the study area where forest loss was determined from the global forest loss map (Hansen et al.,
2013) between 2005 and 2020, “Loss buffer” stratum: 60 m buffer around the loss pixel: “No loss” stratum: remaining pixel within the study area without out “Loss”
stratum and “Loss buffer” stratum.

3
D. Bera et al. Journal of Cleaner Production 356 (2022) 131763

Fig. 3. Pre-disturbance forest cover densities and sampled pixels. (A) Distribution of pre-disturbance forest cover densities for each sampled pixel, (B) percentage of
pixels in pre-disturbance forest cover densities by the year 2005, (C) percentage of sampled pixel identified as forest loss of pre-disturbance forest cover densities
between 2006 and 2020.

Fig. 4. Conceptual diagram of forest pixels and pre-disturbance forest cover densities. If >75% pixel area lies within the forest lands and having canopy cover >25%
defined as a complete forest pixel, but if 25%–75% pixel area lies within forest lands and canopy cover having >25% defined as an edge forest pixel. The canopy
cover of each grid cell was determined using the count of small squares coincided with tree crowns divided by the total number of squares within the forest lands. (A)
Approximate 100% of pixel area occupy forest lands, (B) approximate 67% of pixel area occupy forest lands, (C) approximate 33% of pixel area occupy forest lands.

4
D. Bera et al. Journal of Cleaner Production 356 (2022) 131763

Where Nh is total number of pixels in pre-disturbance forest cover and canopy cover of 2003 is considered as a percentage of forest land
density h, K is the number of sampling strata, Ni is the total number of and canopy cover of 2005 and recorded as a low confidence. Sampled
pixels in stratum i, ni is the total number of sampled pixels in stratum i, pixels 761 and 1012 identify as forest loss by the year 2004. The loss
nhi is the number of sampled pixels in pre-disturbance forest cover happened by the year 2005 consider as non-forests, and pixels loss by the
density h under stratum i. Only 7.369%, 7.047% and 6.749% of pixels year 2004 also consider as non-forests because within one year loss pixel
were dense forests, medium forests and open forests, remaining would not be converted to the forest pixel. After that, categories into
78.835% pixels were non-forests or 21.161% of pixels were forest pixels following pre-disturbance forest cover densities (Fig. S2 and Fig. 4): (i)
within the study area (Fig. 3B). dense forests (>75% canopy cover), (ii) medium forests (>50% canopy
cover), (iii) open forests (>25% canopy cover), and (iv) remaining as
non-forests for analysis forest loss in different forest densities.
2.4. Definitions and measurement principles: forest land, forest loss, forest Forest loss is considered as complete or partial removal of tree
loss year, canopy cover, forest disturbance drivers and ecosystem services crowns within a sampled 30 m by 30 m pixel (Fig. S3). Complete
removal includes stand-replacement disturbance or complete loss >75%
We defined as forest pixels in the current study if >25% area of a 30 of pixel area of complete forest pixel and partial loss associated with
m by 30 m sampled pixels lies within the forest lands by the year 2005 edge pixels. Stand-replacement disturbance or complete loss of 25–75%
and > 25% canopy cover, within the sampled pixel (Fig. 4). If >75% of pixel area of complete forest pixel also considers as a partial loss
pixel area lies within the forest lands and having canopy cover >25% (Figs. S3C and S3D), this category includes pixels located on the edges of
defined as a complete forest pixel. But if 25%–75% pixel area lies within 2006–2020 forest loss patches. Edge pixel considers as a loss that has
forest lands and canopy cover having >25% defined as an edge forest completely cleared between 2006 and 2020 (Figs. S3E and S3F), this
pixel. category includes pixel located on the boundaries of forest patches in the
Collect Earth Online (CEO) tool was used to determine the percent­ year 2005. Thereafter, forest loss was measured in three gradations:
age of forest land and canopy cover of sampled pixels from Google Earth 75–100% loss (counted as 100% of pixel area lost), 25–75% loss (50% of
image by the year 2005, and recorded confidence as high/low. CEO is a pixel area lost), and <25% loss (0% of pixel area lost). All partial loss
web-based free and open-source tool for land monitoring, a product of pixels are treated as 50% of pixel area loss (Figs. S3C–S3F).
collaborative effort between NASA and FAO. Recently, CEO tools have Year of the maximum percentage canopy cover loss of the sampled
increased attention in various purposes as data collection for land use pixel (between 2006 and 2020) is defined as forest loss year. For
land cover (Bey et al., 2016; Saah et al., 2019) and canopy cover example, if multiple partial loss occurred within the study period, the
(Anchang et al., 2020; Bera et al., 2021). To determine the forest land maximum percent canopy cover loss year was recorded as a forest loss
and canopy cover, each sampled pixel was extracted from GEE platform year. If multiple loss occurred within the study period and was later
and arranged consisting of 36 squares of 2 m × 2 m in size using CEO completely cleared, the complete clear year was recorded as a forest loss
tools (Fig. S1). Then arranged pixels were overlapped on Google earth year. But if multiple complete loss and regrowth occurred within the
images using CEO tools. The arranged pixels were used in the visual 2005 to 2020, we only recorded the first complete loss event.
interpretation to the observe presence/absence of tree crowns. Forest Forest disturbances drivers are distinguished using remote sensing
land is simply defined as the approximate percentage of pixel area include clearing for logging, agriculture, roads, settlements, fires, and
occupied by tree crowns and then imaginary boundary was demarcated other (Fig. 5 and Fig. S4). Among drivers, some drivers are easily
between forest and non-forest land (Fig. 4). The canopy cover of each distinguishable using remote sensing (e.g., roads and settlement), but
grid cell was determined using the count of 2 m × 2 m small squares few drivers are difficult to separate from others drivers (e.g., logging vs
coincided with crowns of tree divided by total number of squares within agriculture, and small scale agriculture vs large scale agriculture). In
the forest lands (Fig. 4). From all sampled pixels, 65% had very high these cases, we use set of criteria to distinguish the drivers from each
resolution (<1 m) image on Google Earth, and 18% had image from other using satellite imagery (Table S2). Clearing for logging and agri­
SPOT satellite (2.5 m resolution), and 17% had only Landsat image by culture includes large and small scale clearing categories. Logging is
the year 2005. To get an accurate percentage of forest land and canopy industrial mechanized (large scale) and smallholder non-mechanized
cover of sampled pixels we only assessed the Google Earth images, (small scale) clearing of an area for plantation and wood marketing.
which reached a ≤2.5 m resolution within the sampled pixel. Therefore, Agriculture clearing is characterized by mechanized (large scale) and
to get high resolution images of these 17% sampled pixels, we used smallholder non-mechanized (small scale) clearing for non woody crops.
images of the previous year (2004) and after year (2006) of 2005 from Large and small scale clearing is distinguished based on the size of a
Google Earth image (image having ≤2.5 m resolution). If the percentage clearing. Annual clearing exceeding 10 ha was marked as large scale and
of forest land and canopy cover of the previous and after year is the same below 10 ha was marked as small-scale clearing. Area of the annual
or increase, then the average percentage of forest land and canopy cover clearing patches were estimated on the basis of manually digitized
of this two year considers as a percentage of forest land and canopy annual clearing patches from Landsat composites. Average size of the
cover of 2005. If percentage of forest land and canopy cover is decreased small-scale clearing for logging and agriculture was 2.952 ha and 2.381
(forest loss) from 2004 to 2006, then looking at 16 days cloud-free ha. Large and small scale clearing is further subdivided into clearing for
compositions of 2005 Landsat images with spectral signature (NDVI rotational and semipermanent. Rotational clearing was identified by
and SWIR1) to identify the forest loss. Also, looking at 2005 Google forest regrowth starting 4 years after the clearing. If clearing occurs
Earth low resolution (Landsat) images to identify forest loss. If the loss towards the end of the study period (after 2016), clearing was identified
events not happened throughout the year 2005, the percentage of forest on the basis of the regrowth dynamics of the neighbouring clearing.
land and canopy cover of 2004 is considered as a percentage of forest Clearing for agriculture separated from logging by the post clearing land
land and canopy cover of 2005 because within one year the percentage use: if clearing land converted into cropland within 4 years after clearing
of forest land and canopy cover would not be changed very much. If the marked as clearing for agriculture. Forest clearing for roads includes
forest loss events happened within the year 2005, then sampled pixels construction of roads through the forests (Fig. S4G). Forest clearing for
consider as non-forests. Only for 9 sampled pixels, high resolution roads is distinguished from all other drivers based on the shape and
Google Earth images (≤2.5 m resolution) was not available for the years spatial pattern of the clearing. Clearing for settlement includes resi­
2004, 2005 and 2006. Then looking at 16 days cloud free Landsat dential (rural to urban settlers) as well as commercial (governmental to
compositions with spectral signature (NDVI and SWIR1) and Google non-governmental) construction. We mostly depended on high resolu­
Earth images of 2004 and 2005 to identify forest loss. If the loss events tion Google Earth imagery to identify residential clearing, because using
not happened throughout 2004 and 2005, the percentage of forest land Landsat data it was a very difficult task. Forest loss from fire includes

5
D. Bera et al. Journal of Cleaner Production 356 (2022) 131763

Fig. 5. Example of forest disturbance drivers. Images are subset of pre and post disturbance from Landsat annual cloud-free composite. (A) and (B) represent pre and
post disturbance by roads, (C) and (D) represent pre and post disturbance by settlements, (E) and (F) represent pre and post disturbance by large scale logging, (G)
and (H) represent pre and post disturbance by large scale agriculture, (I) and (J) represent pre and post disturbance by small scale logging, (K) and (L) represent pre
and post disturbance by small scale agriculture, (M) and (N) represent pre and post disturbance by fires, (O) and (P) represent pre and post disturbance by other
(drought). Small red rectangles represent sampled pixels (30 m by 30 m).

areas affected by burning (Fig. S4D) that was not followed by agricul­ indicator functions (Stehman, 2014), this method was also used by
tural activities (clearing for slash and burn agriculture practice). Other Tyukavina et al. (2018) for forest loss estimation. Pixels counted as
forests loss disturbance includes flooding, windfalls, droughts, and river 100%, 50% and 0% of pixel area lost recorded as 1, 0.5 and 0. Forest loss
meandering. area by pre-disturbance forest cover density, by disturbance driver, and
Ecosystem services simply defined as benefits people obtain from by year was calculated using the following equation
ecosystems (Millennium Ecosystem Assessment, 2005). There are vari­

H
ety of benefits mainly, provisioning, cultural, regulating and supporting ̂ = A*
A
Nh
yh (2)
services (Gouwakinnou et al., 2019). We have selected five important h=1
N
provisioning ecosystem services for this study (wild foods, fuel wood,
livestock feed, construction materials, and medicinal plants) by studying Where A is the total study area, H is the number of pre-disturbance forest
various research articles and reports, which are highly affected by forest cover density, N is the total number of pixels in the study region, Nh is
cover loss and also highly impacted to well-being (Gouwakinnou et al., the total number of pixels in pre-disturbance forest cover density h, yh =
2019; Millennium Ecosystem Assessment, 2005). Definition of each ∑
y
selected ecosystem services and their assessment criteria are shown in
uεh u
nh is the mean of the yu values in pre-disturbance forest cover density
Table S3. h, yu is the recorded values of pixels area loss (1, 0.5, and 0) of sampled
pixels u, nh is the sampled pixels in pre-disturbance forest cover density
h. The standard error (SE) of the forest cover loss area is estimated using
2.5. Forest loss area estimation equation

Area estimating for stratified random sampling was performed using

6
D. Bera et al. Journal of Cleaner Production 356 (2022) 131763

√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
√ H ( ) consider as forest loss. Loss pixels were recorded with forest loss year,
√∑ nh s2yh
̂ = A*√
SE( A) N 2h 1 − (3) and forest disturbance drivers. Experts also recorded their confidence as
h=1
Nh nh high/low independently for presence/absence of forest loss, forest loss
∑ year, and forest disturbance drivers. Some major procedures were fol­
Where s2yh =
(yu − yn )2
uεh
is the sample variance for pre-disturbance forest lowed to overcome the uncertainty during sampled interpretation are
nh − 1
shown in Table S2. After initial interpretation, 79% (828) of pixels
cover density h.
screening results were matched between two experts. Another 21%
(220) of pixel results were not matched between two experts in terms of
2.6. Reference data and their availability forest loss, forest loss year, and forest disturbance drivers. Among 21%
disagreement pixels, only 4% (9) had disagreement for all three cate­
The main source of reference data for visual interpretation of gories: forest loss, forest disturbance drivers, and forest loss year. After
sampled pixels was Landsat surface reflectance product (Landsat 7 that, disagreement pixels were iteratively rechecked by two experts
ETM+ and Landsat 8 OLI) that obtained from USGS available on Google together using Table S2 until the consensus was reached.
Earth Engine (GEE) (Gorelick et al., 2017). Available datasets are During the study period 85.587%, 81.286%, and 73.725% of
atmospherically corrected using Land Surface Reflectance Code (LaRSC) sampled pixels were identified as loss of dense forests, medium forests,
for Landsat 8 and the Landsat Ecosystem Disturbance Adaptive Pro­ and open forests (Fig. 3C). Distributions of forest loss pixels by the forest
cessing System (LEDAPS) for Landsat 7, includes cloud, shadow, water, disturbance drivers are shown in Fig. 6A. Highest percentage of loss
and snow mask produced using CFMASK (Vermote et al., 2016). Cloud pixels coming from rotational large scale logging (43.684%), followed
and shadow masking was applied based on the pixel quality band of by rotational small scale logging (21.959%), settlements (7.319%),
Landsat data, and marked as cloud-free Landsat data. 16 days and semipermanent small scale agriculture (5.078%), rotational small scale
annual compositions were applied to each band of cloud-free Landsat agriculture (4.723%), rotational large scale agriculture (4.251%),
data. These cloud-free data was used for the graphs of NDVI (Rouse semipermanent large scale agriculture (3.423%), semipermanent large
et al., 1974), and sort wave infrared reflectance of each sampled pixel scale logging (3.306%), semipermanent small scale logging (2.361%),
independently (Fig. S5) and identify the exact date of forest loss. To fires (2.007%), other (1.417%), and roads (0.472%) (Fig. 6B). Among
provide the landscape context to the visual interpretation of sampled loss pixels (852), highest percentage of loss pixels coming from dense
pixels annual composites of cloud-free images include subset of 40 by 40 forests (45.305%), followed by medium forests (32.629%), and open
Landsat pixels (144 ha) around the sampled pixels (Fig. 5). forests (22.066%) (Fig. 6C).
Availability of the average number of cloud-free observations per As we have used Landsat time series data and high resolution Google
year of each sampled pixel is shown in Fig. S6A. Number of sampled Earth image to get the best information of each sampled pixel, and also
pixels by the average number of cloud-free observations per year are the absence of historical (2005–2020) ground surveys, it is practically
shown in Fig. S6C. The mean average number of observations was impossible to directly assess the accuracy of our results. Therefore, we
10.906, with a minimum of 7.764 and a maximum of 14.647. But the have indirectly assessed the potential errors of our results by analysing
mean average number of observations was highest in non-forests forests confidence level as high/low of each sampled pixel. A total of 91
(12.527), followed by dense forests (11.125), medium forests (10.759), sampled pixels (approximate 5% of the total sampled pixels of 2000)
and open forests (9.213). Minimum number of cloud-free observations were identified as low confidence during canopy cover measurement
per year of each sampled pixel is shown in Fig. S6B. Only 4% (86) of and categorization as pre-disturbance forest cover density (Table S4A).
2000 sampled pixels had ≤3 cloud-free observations per year Fig. S6D, Highest percentage of low confidence (potential commission error)
but among 4% highest was in medium forests (29 pixels), followed by coming from medium forests (11%), followed by open forests (9%),
open forests (22), dense forests (18), and non-forests (17). This bias of dense forests (4%), and non-forests (1%). Medium forests and open
availability of cloud-free observations among the pre-disturbance forest forests had the highest potential commission error due to their likely
cover density could affect the visual interpretation results. confusion (medium forests: confusion with dense forests and open for­
Google Earth images were used to determine the forest land and ests, open forests: confusion with medium forests and non-forests) with
canopy cover as well as to identify the forest cover density (Fig. S2). other categories of pre-disturbance forest cover density because of the
Forest land and canopy cover was measured by the year 2005 from unavailability of high resolution Google Earth image. Regarding, pres­
Google earth imagery using the CEO tools (see Materials and methods: ence/absence of forest loss, total of 37 sampled pixels (approximate 4%
“Definitions and measurement principles” section). High resolution of the total forest pixels of 1048) were classified as low confidence
Google Earth images also help to detect the presence/absence of forest (potential commission error) during sample interpretation (Table S4B).
loss and forest disturbance drivers (Fig. S4). Sampled pixels with Highest percentage of potential commission error coming from open
detected forest disturbance drivers in the disturbance years had higher forests (6%; loss 11, no loss 4), followed by medium forests (4%; loss 8,
availability of Google Earth images (82% VHR, 11% SPOT, and 7% no loss 4), and dense forests (2%; loss 7, no loss 3). Years with highest
Landsat only). These availabilities of very high resolution imagery on percentage of forest loss area coming from low confidence sampled
Google Earth (<1 m resolution) facilitated the visual interpretation of pixels were 2006 (29%), 2010 (28%), 2007 (22%), and 2012 (17%);
sampled pixels. years with lowest percentage were 2019 (1%), 2014 (2%), 2009 (3%),
and 2015 (4%). An average, lowest potential commission error was
2.7. Sampled pixels analysis and accuracy estimation between 2014 and 2020 (6%) compared with 2006–2013 (16%). These
results might be explained by the increased availability of cloud-free
At first, canopy covers were measured of all 2000 sampled pixels by Landsat data (Fig. S7) and the very high resolution Google Earth im­
two experts together and grouped as dense forests, medium forests, open agery in the later year.
forests, and non-forests with high/low confidence. From sampled pixels, Regarding forest disturbance drivers, total of 22 (3% of total loss
52.4% (1048) of pixels were determined as forests pixels and the pixels of 852) loss pixels were classified as low confidence level
remaining 48.6% (952) of pixels were non-forests pixels (Table S1). (Table S4C). Highest contributed by open forests (10 pixels), followed by
Among 1048 forest pixels 451, 342, and 255 pixels were dense forests dense forests (7 pixels), and medium forests (5 pixels). Highest per­
pixel, medium forests pixel, and open forests pixel. centage of area with low confidence sampled pixels was coming from
Visual interpretation of forests sample pixels was performed by two fires (11%), followed by semipermanent small scale logging (10%),
experts independently, who determined forest loss (0%, 50% or 100%) semipermanent large scale agriculture (7%), semipermanent large scale
of each forest pixel. Pixels counted as 50% and 100% of pixel area lost logging (7%), other (7%), rotational large scale agriculture (3%),

7
D. Bera et al. Journal of Cleaner Production 356 (2022) 131763

Fig. 6. Forest disturbance drivers and sampled pixels. (A) Distribution of forest disturbance drivers for each sampled pixel, (B) percentage of loss sampled pixels by
forest disturbance drivers between 2006 and 2020, (C) percentage of loss sampled pixels shared by pre-disturbance forest cover densities between 2006 and 2020.
rota-rotational, semi-semipermanent.

rotational small scale agriculture (3%), semipermanent small scale asked the respondents two questions: 1) what have you observed on
agriculture (2%), rotational small scale logging (2%), rotational large forest cover around your village and surroundings over the last 15 year?;
scale logging (2%), settlement (2%), and roads (0%). These differences 2) and what have you observed the availability of provisioning
are due to their likely confusion among the few disturbance drivers. For ecosystem services over time? During the interviews, we explained the
example, roads are totally distinct drivers, but consequently, fires have each ecosystem services to the respondents using local language, and
vicinity with other disturbances like rotation small scale agriculture asked how the availability of these ecosystem services are affected by
(slash and burn agriculture practice). Semipermanent small scale log­ forest cover change. The perceptions of households about forest cover
ging is likely to be confused with semipermanent small scale agriculture and the availability of ecosystem services were noted as “increased”,
and rotational small scale logging with rotational small scale agricul­ “decreased”, or “stayed the same” toward the forest cover area.
ture. Most importantly years of unavailability of high resolution Google Chi-square goodness of fit test was performed to reveal significant uni­
Earth imagery of sampled pixels was created difficulty to identify proper formity or difference of people’s perception for forest cover decreased,
forest disturbance drivers. increased and stable among the villages.

3. Results and analysis


2.8. Household survey using structured questionnaires
3.1. Temporal analysis of forest cover loss
A household survey with questionnaires was conducted at 50 villages
(forest loss affected villages during 2006–2020) in the study region. For Temporal forest cover loss by pre-disturbance forest cover densities
the selection of villages, we quantified forest loss of each village of the are shown in Fig. 7 and Table 1. Total forest cover loss was increasing
study region from global forest loss map (Hansen et al., 2013) over the trend from 2006 to 2020 with the rate of 1.30 thousand ha per year. The
last 15 years (2006–2020). After that, 50 villages were selected increasing rate of forest loss was higher in dense forests (0.64 thousand
randomly which are affected by forest loss. In each village, 10 house­ ha per year), followed by medium forests (0.50 thousand ha per year)
holds were selected which are totally dependent on agriculture and and open forests (0.20 thousand ha per year) (Fig. 7). Although forest
forest over the 15 years by the consulted local government respondents cover loss was accelerated, but not all pre-disturbance forest cover
and villagers. The survey was conducted at each selected household with densities display the same increasing trend during the study period
most senior and knowledgeable persons. The household survey was (Table 1). In open forests, forest cover loss first accelerated (2006–2010:
conducted on the people’s perception about forest cover change and 34.764 ± 3.338 thousand ha to 2011–2015: 60.514 ± 3.137 thousand
availability of ecosystem services, around village and surrounding over ha), then slowed down again (2016–2020: 42.059 ± 4.058 thousand
the last 15 years. In recent years, many researchers have used people’s ha), but consequently, the loss was accelerated in dense forests
perceptions to evaluate forest cover changes and availability of (Table 1).
ecosystem services over the years (Ahammad et al., 2019; Gouwakinnou Temporal forest cover loss by forest disturbance drivers is shown in
et al., 2019). Local people’s experience provided historical trend of Table 2 and Table S5. In pre-disturbance forest cover densities, forest
availability of ecosystem services (Alfonso et al., 2016), which is useful cover loss by forest disturbance drivers also varied over the time period.
for forest management and improvement of rural livelihoods (Fisher and Among forest disturbance drivers, only rotational large-scale logging is
Hirsch, 2008; Yang et al., 2015). Hence, this present study has used increasing over the period in dense forests, and the remaining drivers do
people’s perceptions to understand the trend of natural forest cover and not have the clear pattern of increasing or decreasing over time in any
the impact of forest cover loss on provisioning ecosystem services. We

8
D. Bera et al. Journal of Cleaner Production 356 (2022) 131763

Fig. 7. Temporal trends of forest loss. (A) Total forest loss, (B) forest loss in dense forests, (C) forest loss in medium forests, (D) forest loss in open forests. Error bands
represent ± SE.

9
D. Bera et al. Journal of Cleaner Production 356 (2022) 131763

Table 1 pre-disturbance forest cover density (Table S5). But the gross forest
Temporal forest loss (5 year epochs) by pre-disturbance forest cover densities cover loss by small scale agriculture (rotational and semipermanent) is
(thousand hectares ± SE). decreasing, consequently rotational large-scale logging, fires, and other
2006–2010 2011–2015 2016–2020 Total drivers is increasing over the time period (Table 2). Rotational large-
Dense 46.369 ± 53.523 ± 80.020 ± 179.913 ±
scale logging and rotational small scale logging was the top two
forests 3.631 3.799 3.570 4.435 drivers, which also was the most dominant drivers over the time period
Medium 45.774 ± 41.096 ± 65.821 ± 152.691 ± in the pre-disturbance forest cover densities (Table S5), except for
forests 3.106 2.828 3.855 4.983 2011–2015 in medium forests, where rotational large scale agriculture
Open forests 34.764 ± 60.514 ± 42.059 ± 137.338 ±
prevails over rotational small scale logging.
3.338 3.137 4.058 5.008
Total 126.907 ± 155.134 ± 187.901 ± 469.942 ±
5.264 4.591 4.358 7.886 3.2. Comparison of forest loss among pre-disturbance forest cover
densities

Table 2 Comparative results of forest cover loss among pre-disturbance forest


Temporal forest loss (5 year epochs) by forest disturbance drivers (thousand cover densities from 2006 to 2020 are shown in Fig. 8 and Table 1. The
hectares ± SE). rota-rotational, semi-semipermanent. estimated total forest cover loss from 2006 to 2020 in the study region
2006–2010 2011–2015 2016–2020 Total loss was 469.942 ± 7.886 thousand ha (thousand hectares ± SE). Most of the
Small scale 8.652 ± 5.916 ± 4.969 ± 19.537 ±
year dense forests shared the highest forest loss area than the other pre-
agriculture 3.210 3.001 3.979 4.152 disturbance forest densities. Also, among total forest cover loss, forest
rota. loss in dense forests during 2006–2020 was highest (179.913 ± 4.435
Small scale 13.339 ± 5.680 ± 2.661 ± 21.679 ± thousand ha), followed by medium forests (152.691 ± 4.983 thousand
agriculture 4.690 2.226 2.661 5.515
ha), and open forests (137.338 ± 5.008) (Table 1). During 2006–2010
semi.
Large scale 4.678 ± 10.248 ± 9.384 ± 24.309 ± and 2016–2020, the forest cover loss was largest in the dense forests,
agriculture 2.335 3.266 2.069 3.593 followed by medium forests, and open forests, but during 2011–2015
rota. largest was in the open forests, compared to dense forests, and medium
Large scale – 10.000 ± 10.000 ± 20.001 ± forests (Table 1).
agriculture 3.331 3.332 4.329
semi.
Almost 75.279% area of dense forests marked a loss by disturbance
Small scale 32.556 ± 26.300 ± 28.909 ± 87.765 ± drivers during 2006–2020, followed by medium forests (66.813%), and
logging rota. 4.364 3.100 3.803 5.356 open forests (62.745%). The increasing trend of forest loss is also higher
Small scale – – 4.239 ± 4.239 ± in dense forest than other forests cover densities, thereafter, dense for­
logging semi. 2.297 2.938
ests are mostly under threat in near future.
Large scale 52.284 ± 75.597 ± 99.359 ± 227.241 ±
logging rota. 4.739 3.946 3.386 5.139
Large scale 5.150 ± 12.869 ± 5.184 ± 23.204 ± 3.3. Associated drivers of forest cover loss
logging semi. 2.797 3.259 2.242 3.889
Fires – 3.445 ± 6.757 ± 10.202 ±
2.126 3.112 4.938 Forest disturbance drivers and their relative contributions to forest
Roads 0.265 ± – 0.795 ± 1.060 ± cover loss from 2006 to 2020 are shown in Fig. 9. Among all drivers,
0.265 0.795 0.892 rotational large scale logging was the largest driver from 2006 to 2020
Settlements 8.626 ± 2.730 ± 12.080 ± 23.437 ± (48.355%), followed by rotational small scale logging (18.676%),
3.118 2.730 3.311 4.782
Other 1.356841 2.3479921 3.5633945 7.268 ±
rotational large scale agriculture (5.173%), settlements (4.987%),
3.562 semipermanent large scale logging (4.983%), semipermanent small
Total loss 126.907 ± 155.134 ± 187.901 ± 469.942 ± scale agriculture (4.613%), semipermanent large scale agriculture
5.264 4.591 4.358 7.886 (4.256%), rotational small scale agriculture (4.156%), fires (2.171%),
other (1.547%), semipermanent small scale logging (0.902%), and roads
(0.226%) (Fig. 9A).
Rotational large and small scale logging was the top two direct

Fig. 8. Comparison of forest loss among pre-disturbance forest cover densities from 2006 to 2020. See Fig. 7 for SE of estimates of each year.

10
D. Bera et al. Journal of Cleaner Production 356 (2022) 131763

Fig. 9. Forest loss estimates of 2006–2020 by forest disturbance drivers. (A) Total forest loss, (B) forest loss in dense forests, (C) forest loss in medium forests, (D)
forest loss in open forests. Error bands represent ± SE. rota-rotational, semi-semipermanent.

drivers in the pre-disturbance cover densities (Fig. 9 and Table 3), and thousand ha) of total forest loss area, which is characterized by clearing
both rotational logging constituted 67.031% of the total forest loss area for the planting of non-woody raw crops (Fig. 9A and Table 3). Among
from 2006 to 2020. Consequently, forest cover loss by semipermanent agricultural clearing, rotational agriculture and semipermanent agri­
logging (large and small) is much less common than rotational logging culture is shared 51.266% and 48.734% of forest cover loss area. Small
(5.885% vs 67.031%). Logging activity (72.916% of total loss area) scale agriculture contributed 48.191% (41.216 ± 5.875) of agricultural
basically rotational logging (67.031% of total loss area) is the domi­ clearing, and among small scale agriculture, 52.598% was semi­
nating driver in the study region for plantation and wood marketing. permanent clearing. Because of less size of clearing (average size 2.986
Among rotational logging (325.006 ± 6.324), large scale logging ha) and lack of access roads visible in very high-resolution imagery,
constituted 69.912% (227.241 ± 5.139 thousand ha) of forest loss area. most of the small scale clearing for agriculture is likely non-mechanized
The largest contributor for rotational large scale logging was dense and clearing by local farmers (Fig. S4F). Although all four types of
forests (46.423%), followed by open forests (28.434%), and medium clearing for agriculture is dominating in medium forests (Table 3), but
forests (25.143%) (Table 3). Semipermanent large-scale logging semipermanent large-scale agriculture was most dominant (32.5928%)
constituted 4.983% (23.204 ± 3.889: 84.553% of semipermanent log­ than other types of disturbance. Also, medium forests contributing
ging) of the total loss area, and the highest loss area is coming from open 73.501% loss area of semipermanent large-scale agriculture, followed
forests (44.389%), followed by dense forests (29.689%), and medium by dense forests (26.499%), and open forests (0%). From the total
forests (25.922%) (Table 3). The rotational small scale is the highly agricultural clearing, medium forests were contributed 50.399% of
contributed to the forest cover loss in open forests (36.596%) and me­ agricultural clearing, followed by dense forests (28.199%), and open
dium forests (35.024%) and followed by dense forests (28.380%) forests (21.402%).
(Table 3). Forest clearing for settlement is the fourth largest contributor to total
Forest clearing for agriculture constituted 18.199% (85.526 ± 6.229 forest area loss, which is shared 4.987% (23.437 ± 4.782 thousand ha)

11
D. Bera et al. Journal of Cleaner Production 356 (2022) 131763

Table 3 has been decreasing over time around the village and surroundings
Forest loss in pre-disturbance forest cover densities by forest disturbance drivers (Table S6). Only 4.4% of respondents perceived a slight increase in
between 2006 and 2020 (thousand hectares ± SE). rota-rotational, semi- forest cover due to the plantation across the region and 11.6% perceived
semipermanent. forest is stable over the time period. Chi-square goodness of fit test
Drivers Dense Medium Open forests Total loss showed no significant difference of people’s perception about forest
forests forests cover loss (X2 = 42.857, p < 0.05) among the villages, but significant
Small scale 4.504 ± 12.028 ± 3.004 ± 19.537 ± difference in people’s perception about stable or dynamic (X2 =
agriculture rota. 2.989 2.563 2.390 4.152 195.862, p < 0.05) and increase of forest cover among the villages (X2 =
Small scale 6.889 ± 8.353 ± 6.438 ± 21.679 ± 370.909, p < 0.05). That means people perceived natural forest cover
agriculture semi. 2.938 3.169 2.426 5.515
Large scale 7.419 ± 10.023 ± 6.867 ± 24.309 ±
has decreased in similar trend or uniformly across the villages, and
agriculture rota. 2.504 2.595 2.690 3.593 stable and increase of forest cover followed the diverse pattern across
Large scale 5.299 ± 14.701 ± – 20.001 ± the villages.
agriculture semi. 2.875 2.381 4.329 Availability of ecosystem services also decreased due to forest cover
Small scale logging 24.907 ± 30.739 ± 32.119 ± 87.765 ±
loss. Perceptions of households about the availability of specific
rota. 3.296 3.077 3.372 5.356
Small scale logging 4.239 ± – – 4.239 ± ecosystem services in the study region are shown in Fig. 10. It was
semi. 2.938 2.938 evident that overall, 73.92% of households perceived the decrease in the
Large scale logging 105.493 ± 57.134 ± 64.613 ± 227.241 ± availability of provisioning ecosystem services over time.
rota. 3.515 3.893 4.1995 5.139 Our result also showed wild foods were the most affected ecosystem
Large scale logging 6.889 ± 6.014 ± 10.300 ± 23.204 ±
services due to forest cover loss. Gross 81.6% of households perceived
semi. 2.561 2.582 2.797 3.889
Fires 4.239 ± 1.671 ± 4.292 ± 10.202 ± that the availability of wild foods decreased, while 18.4% perceived
2.905 1.671 2.754 4.938 stability in the availability of wild foods. Medicinal plants were the
Roads 1.060 ± – – 1.060 ± second most affected provisioning ecosystem. Total 81% of households
0.892 0.892
perceived decreases in the availability of medicinal plants. But 17.8%
Settlements 5.829 ± 12.028 ± 5.579 ± 23.437 ±
2.732 2.984 3.310 4.782 reported availability of medicinal plants remained stable and only 1.2%
Other 3.143 ± – 4.125 ± 7.268 ± of household’s perceived increases in their villages and surroundings.
2.356 2.551 3.562 Total 75.4% of household’s perceived decreases in the availability of
Total loss 179.913 ± 152.691 ± 137.338 ± 469.942 ± livestock feed. But only 19.4% perceived stable and 5.2% perceived
4.435 4.983 5.008 7.886
increased availability of livestock. Availability of construction materials
and fuel wood also decrease, gross 67.2% and 64.4% of households
of total forest area loss (Fig. 9A). The largest contributor of settlements perceived decreases the availability of construction materials and fuel
to forest loss is observed in medium forests (51.322%), followed by wood. But more households (33.2%) reported availability of fuel wood
dense forests (24.872%), and open forests (23.805%). Clearing for set­ remained stable compared to other ecosystem services (Fig. 10). Only
tlements is likely the semipermanent type of clearing, which is 2.4% and 3.2% of households perceived increases in fuel wood and
contributing to forest loss in the edge of the forests (Fig. S4C). construction materials.
Clearing for roads is a very rare phenomena, contributing only
0.226% (1.060 ± 0.892 thousand ha) of total forest loss area during 4. Discussion
2006–2020 in the study region (Fig. 9A and Table 3), which is only
observed in the dense forests. The estimated forest loss due to road has 4.1. The loss of forest cover varies with forest cover densities
high standard error because it is only based on 3 sampled pixels (1.060
± 0.892: SE expressed 84.151% of driver area). Based on the result, it was evident that forest cover loss varied among
Fire constitutes only 2.171% (10.202 ± 4.938 thousand ha) of the forest cover densities over time. The diverse pattern of forest cover loss
total forest loss area in the study region (Fig. 9A and Table 3). The over spatial context due to the variation of human activity and its in­
largest contributors are open forests (42.070%), followed by dense tensity over time and space (Sulieman, 2018). Gross forest cover loss was
forests (41.551%), and medium forests (16.379%). Almost 83.333% of 469.942 ± 7.886 thousand hectares with an increasing trend (average
fire sampled pixels were identified as fire during the dry period (De­ 1.30 thousand ha per year) from 2006 to 2020. But the loss was higher
cembers – April). Also, 66.667% of fire sampled pixels are found in the spatial-temporally in dense forests (179.913 ± 4.435 thousand ha with
edge of the forests, but all fire sampled pixels were adjacent to roads, an average increasing rate of 0.64 thousand ha per year) compared to
settlements or agricultural fields. We can expect most of the fire is likely medium forests (152.691 ± 4.983 thousand ha with an average
related to the human induced activities. increasing rate of 0.50 thousand ha per year), and open forests (137.338
Other forest disturbance drivers including windfalls, river ± 5.008 thousand ha with an average increasing rate of 0.20 thousand
meandering, drought and inundation, contributed only 1.547% (7.268 ha per year). This result might be explained by the dominance and
± 3.562) of the total forest loss area (Fig. 9A and Table 3). Dense forests increasing trend of rotational large scale logging for wood marketing
contributed 43.244% as inundation and the remaining 56.756% of the and plantation in the study region. Rotational large scale logging
loss area coming from open forests as windfalls and drought. contributed 58.636% of dense forest loss area between 2006 and 2020
(Fig. 9B) and is continuously increasing over the time period (Table 2).
In open forests, forest cover loss first accelerated (2006–2010:
3.4. Households perceptions of forest cover loss and availability of 34.764 ± 3.338 thousand ha to 2011–2015: 60.514 ± 3.137 thousand
ecosystem services ha), after that loss was declined 30.497% during 2016–2020 compared
to the previous period 2011 to 2015 (Table 1). The declining trend of
Compared to the findings of remote sensing imagery, the household clearing in open forests is related to the way the open forests are defined
survey about natural forest cover loss revealed the similar type of re­ and the limited pool of open forests later the year of the study period.
sults. Sampled pixels analysis result shows that 68.463% of forest area is Consider as open forests pixel if it had canopy cover >25–50% in the
marked as a loss by disturbance drivers, and loss is increasing over time year of 2005. Non-forests that reached canopy cover >25% after 2005
(Fig. 7). Respondents in the household survey perceived loss of forest was not considered as forest pixels and if this were later cleared were not
cover over the time period across the study region. It was evident that an considered as forest loss. And 43.529% area of defined open forests were
average 84% of household respondents perceived natural forest cover cleared by the end of 2015 (Table 1), which was the highest percentage

12
D. Bera et al. Journal of Cleaner Production 356 (2022) 131763

Fig. 10. Proportions of households (%) perceived changes in the most important provisioning ecosystem services.

of forest area loss compared to other forest cover densities within this quietly consistent with study local perspectives in interpreting the
period (2006–2015). Therefore toward the end of the study period, most decline of forest area in Indonesia (Fisher et al., 2013). These strong
of the open forest has been cleared, resulting in decreasing of clearing in agreement between remote sensing and local people also indicated
open forests. That indicates the dependency on open forests has been forest cover loss highly concentrated and widespread in the village
decreased which might be explained by the decreasing trend of forest surroundings and neighbouring areas (Ahammad et al., 2019). Hansen
loss area by most of the drivers during 2016–2020 compared to the et al. (2013) global forest loss map indicates the increasing trend of
previous period 2011–2015 (Table S5). And consequently is increasing forest loss, but very much underestimates the forest loss area over the
dependency on medium and dense forests towards the end of the study study period (Fig. 11). The large difference in the results might be
period, which is explained by the increase of forest loss 60.164% and explained by the fact that global-level studies are causes of over­
49.506% during 2016–2020 compared to previous period 2011–2015 in estimation or underestimation at the local level (Tyukavina et al., 2018).
medium forests and dense forests. Therefore, the intensity of forest loss The national forest report (FSI, 2019) and state report (WBFD, 2018)
by most of the drivers was increased during 2016–2020 compared to the showed that overall area of forests has increased in the recent decade’s
previous period 2011–2015 in medium forests and dense forests national and state level. But our study on gross forest cover loss is
(Table S5).The dominance and increasing trend of rotational logging in evident that forest cover loss was significantly increasing trend from
medium and dense forests for wood marketing due to the large amount 2006 to 2020 in this area. And trends of forest cover loss were higher in
of availability of big and mature trees, which is replaced by the plan­ dense forests compared to the other forest cover densities, which in­
tation of single tree species. dicates dense forests are under threat in near future. This bias is due to
There is no previous study on forest cover loss at different forest Forest Survey estimates forest cover using automated algorithms by
cover densities to compare with this study. Only few previous studies are analysing satellite imagery that fails to separate native natural forests
available (Dutta et al., 2020; Sudhakar Reddy et al., 2016) on overall from plantations, which are often monocultures of exotic species that
forest cover change in this areas, but these results are not directly have limited value for endangered biodiversity (Puyravaud et al., 2010).
comparable with current results due to different definitions, study This finding evident that forest loss is increasing as a result of clearing
period, and study boundaries. But local people perceived continuous native natural forests and also increasing plantation of exotic species at
forest cover loss and natural forest cover has been decreasing trend the same time. In this context, a forest preservation and management
which was similar to remote sensing data analysis. These findings were strategy is critical to halting forest loss and ensuring the existence of

Fig. 11. Comparison of annual forest loss estimates between 2006 and 2020. Current study estimates are compared with the global forest loss map (Hansen
et al., 2013).

13
D. Bera et al. Journal of Cleaner Production 356 (2022) 131763

natural forests in the future. findings quietly agree with the several studies of forest cover loss at
national and global levels in poor countries (Curtis et al., 2018; Sloan
4.2. Role of drivers of forest cover loss and Sayer, 2015). Under the assumption, the forest cover loss is related
to the demographic growth and uninformed management in this region.
The current study emphasis on the quantitative assessment of forest Because higher population density and increasing rate over time with
disturbance drivers of forest cover loss in different pre-disturbance for­ low level development are the main criteria of this region (State statis­
est cover densities. Details information on forest cover loss by distur­ tical handbook, 2015). Overall indicates that under developing stage
bance drives does not exist in this region. Our analysis fills the existing annual forest loss area is increasing, when industrial development will
knowledge gap in this research by quantifying forest loss of each be implemented, might bring new challenge to forest resource man­
disturbance driver (Fig. 9). Intensity and contribution of each driver agement. Decrease in wood production is the most important mitigation
varied over pre-disturbance forest cover densities over time (Table S5). strategy to prevent logging (Liu et al., 2020) in the forest ecosystem.
The variation of driving factors in a spatial context is primarily deter­ Need to focus on reducing deforestation for industrial-scale, residential
mined by the variety of activities over space. (Hosonuma et al., 2012). urban growth and agricultural activity to address the deforestation in
Rotational large and small scale logging which was the dominant the future (DeFries et al., 2010). This quantitative assessment of forest
driver in all forest cover densities (Fig. 9), responsible for gross 67.031% loss by each disturbance drivers provide complete information about
of total forest loss between 2006 and 2020. Large scale logging which is forest cover loss over space and time. This information can be used to
likely mechanized was the most dominant and is increasing native take mitigation strategy to tackle individual disturbance drivers for
natural forest clearing by large scale logging in medium and dense for­ forest management via control over forest loss.
ests over time due to the availability of large amount of big and mature But it is not easy to control the drivers of forest cover loss without
trees, and replaced by the plantation of single tree species. This result secure tenure rights, management practice, and consideration of appli­
might be explained by the demand for industrial wood increased in the cable laws and policies. Various previous studies showed the secure
tropical poorer countries since 1990, and demand will accelerate into tenure rights encourage people to plantations (Dolisca et al., 2007;
the future (Sloan and Sayer, 2015). Demographic and economic growths Walters, 2012), and laws and policies with suitable strategies contribute
are the main accelerator of deforestation and forest exploration in lower to protecting the forests (Ongugo et al., 2014; Saraiva et al., 2020;
income countries (Nikolakis and Innes, 2014) and increased industrial Tacconi et al., 2019).
removal rapidly between 1990 and 2015 (Köhl et al., 2015). Increasing
demand for agricultural products threatens to forest loss (Laurance 4.3. Impacts of forest cover loss on provisioning ecosystem services
et al., 2014) and will continue rapidly in the developing tropical coun­
tries. Sample based estimates Curtis et al. (2018) also found that global Our results highlight the overall availability of provisioning
forest loss is mainly driven by forestry for plantation and agriculture. ecosystem services decreased by the forest cover loss. This result is
Due to demographic growth, increase demand for forest products and quietly similar to the findings of Ehara et al. (2016) which reported the
services as well as agricultural land for economic growth (Sloan and forest cover loss has negative impacts on provisioning ecosystem ser­
Sayer, 2015). Almost 18.199% of total forest loss area is cleared by vices. Over 80% of households perceived that declined availability of
agriculture. Rotational agricultural clearing was responsible for wild foods and medicinal plants, which pose the greatest threat to their
51.288% of the agricultural clearing. This indicates shifting type culti­ livelihoods due to health risks associated with loss of availability of
vation is still dominating driver to forest loss in this region, which re­ nutrient rich foods and medicinal plants. The reduced availability of
flects the lack of agricultural strengthening. construction materials, livestock feed, and fuel wood always increased
Although the government implemented few forest related acts (In­ threat to the livelihoods of local people due to increased time spent
dian Forest Act 1927, Wildlife Protection Act 1972, Forest Conservation collecting woods and livestock feed (Ahammad et al., 2019). If there is
Act 1980, Biological Diversity Act 2002, Scheduled Tribes and Other adequate access to provisioning ecosystem services also means that the
Traditional Forest Dwellers Act 2006) to protect the forest resource and considerable amount of time people spends collecting services can be
plantations, but small scale agriculture (responsible for 19.578% of total spent on other productive tasks that improve livelihoods by economic
forest loss area) and logging (responsible for 19.578% of total forest loss productivity, which is a key component of poverty alleviation (Van
area) which is likely non-mechanized clearing by local people, is also Jaarsveld et al., 2005). This declined trend of provisioning ecosystem
dominating over the time period in this region. This indicates that the services in their village and surroundings further forces to decline of
lack of intensification of the forest related policy at the regional level natural forests in dense and core areas to meet their essential needs. This
and failure to fulfil the essential needs of local people by the government persistent decline of ecosystem services suggests more and more decline
over time. of natural forests.
Clearing for constructions or infrastructures development such as Decline in ecosystem services is associated with loss of forest cover,
settlements (responsible for 4.987% of total forest loss) and roads but the perceptions about the availability of ecosystem services also
(responsible for 0.226% of total forest loss) is the sign of development to varied (Fig. 10). Variation in perceived availability of ecosystem ser­
fulfil the demand of demographic growth. Settlements and roads are vices is directly dependent on variation in education level, poverty
important forces influencing others drivers to forest loss. Roads stimu­ index, household size, proximity to the forest and nature of the forest
late deforestation by facilitating the construction of smaller side roads (Gouwakinnou et al., 2019). But we selected villages for household
and human settlements in remote areas and play many important roles survey based on the dominance of forest cover loss over time, and most
in facilitating deforestation (Fearnside, 2015). For infrastructure of the respondents perceived highly declining trend of forest cover and
development needs better decisions by the government rather than po­ availability of ecosystem services over time. These integrated findings
litical considerations to avoid deforestation over time and space. Envi­ with combinations of remote sensing data and people’s perceptions
ronmental Impact Assessment (EIA) system must be practised prior to indicate that the decline of ecosystem services is closely related to forest
decision making for implementing the new development project in the cover loss in the study region. This declining trend of provisioning
developing countries for future sustainable development (Shah et al., ecosystem services may be the greatest threat to their well-being in the
2010). future. Although in this study region most of the forests are cleared by
Any generalization about some specific drivers of forest loss may not rotational drivers (Table 2) and has been replaced by plantation of single
be valid compare to other national or global level studies. But demog­ tree species. But Gamfeldt et al. (2013) found that mature and higher
raphy and management practices are always related to forest clearing, biodiversity provides multiple services, and no single tree species are
lead to environmental degradation (Dasgupta, 1995). However, these able to promote all services. Clearing native natural forests and replaced

14
D. Bera et al. Journal of Cleaner Production 356 (2022) 131763

by plantation of single tree species would not provide all essential and increase availability of high resolution imagery on GEE platform in
products and services for human beings in the study region. This is the future satellite image oriented research would be more feasible.
main issue of this region, which is lead oppose to sustainable
management. 6. Conclusions
Various studies have reported effective ecological projects and pol­
icies have achieved some success in promoting environmental condi­ Our study examined the forest cover loss by forest disturbance
tions by producing ecosystem services by the transformation of arable drivers in the different forest cover densities and the impacts on provi­
land to revegetated land (Deng et al., 2019; Lu et al., 2020). But some sioning ecosystem services by using remote sensing data and local
management efforts also have failed to promote environmental condi­ people’s perception in the dry deciduous forest of West Bengal from
tions with maintain ecosystem services due to lack of national policies 2006 to 2020. After the study, it was observed that forest cover loss
(Dou et al., 2020), inappropriate vegetation species (Feng et al., 2017), varied over pre-disturbance forest cover densities with dominating in
and mismatches of production and consumption pattern of ecosystem dense forests, and has been increasing trends over time. Local people
services in an area (Raudsepp-Hearne and Peterson, 2016). Our study also perceived continuous forest cover loss to their villages and sur­
found that decline of provisioning ecosystem implying that production is roundings over the time period. Intensity of associated driving factors
falling short of local demand. In such circumstances, governments must was varied over pre-disturbance forest cover densities through time, but
implement a better match of vegetation species with proper planting logging and agricultural clearing were dominant over the time. Local
density to the natural environments for ecological restoration with the people perceived that ecosystem services are negatively affected by the
management of ecosystem services. But for cleaner production and forest cover loss over time. Wild foods and medicinal plants were most
maintain the services in future cross-scale and long term monitoring effected services to well-being due to forest cover loss. There is scare
policy is constantly needed (Dou et al., 2020). information on provisioning ecosystem services which is warning to
local livelihoods in future. To avoid an ever-increasing rate of forest loss
5. Limitations and future research advance in dense forests, forest protection should be prioritised in order to ach­
ieve long-term sustainability. Need urgent control over logging and
Direct assessment of forest loss by forest disturbance drivers from agricultural clearing to control forest loss from open to dense forests.
remote sensing data on Google Earth Engine is providing detailed in­ Rotational agriculture by clearing of forest is still dominating in this
formation with measurement feasibility. This is easily adaptable to region, it is time to strengthening the agriculture activity over the re­
determine region-specific drivers in the different geographical domains. gions. We also suggest rather than planting single tree species planting
But we have faced difficulty to distinguished a few drivers from others multiple tree species would be more beneficial to attained sustainability,
drivers. For example, logging versus agriculture, clearing for residential because no single tree species are able to promote optimum biodiversity
was difficult to identify from the 30 m by 30 m Landsat images, fire have and all ecosystem services (Gamfeldt et al., 2013). The present study
vicinity with rotational slash and burn agricultural practice. High res­ also revealed that ineffectiveness of the present management strategies
olution and high frequency reference data would be more effective to in the forest as well as maintaining ecosystem services due to lack of data
improve the estimation in the different environmental situations. Future and lapse in government law and policies. Due to variation in forest
research should have more focused on the distinguished process of each cover loss and intensity of driving factors over pre-disturbance forest
forest disturbance driver, which will provide global certainty to forest cover densities, and the difference in the perceptions of ecosystem ser­
loss estimation. Indirect forest disturbance drivers such as livestock vices across the region, community based local adaptive management
grazing, artisanal wood marketing, timber and fuel collection are also (Williams, 2011) strategy requires for increase forest cover, and benefits
impacts on forest loss in the tropical developing countries (FAO, 2015). to local communities in term of access to forest, land ownership and
But it is nearly impossible to quantify the area affected by these drivers economic opportunities (Lu et al., 2020). In these areas, Joint Forest
using remote sensing data. Incorporation of these drivers may provide Management (JFM) policy contribute to the protection, plantation and
further details information about forest loss in an area. Regarding forest provide economic benefits to rural people in some areas (Pattnaik and
pixel classification we only used canopy cover, but forest cover defini­ Dutta, 1997; Guha et al., 2000). Strong implementation of JFM policy
tion is always matter of debate, which creates uncertainty in forest loss can result in positive social interaction, plantation, protection and share
area measurement. In future, improvement of LIDAR data shall help to economic benefits among the local people.
direct mapping of canopy height at the local level and other structural Implications of these results in the broadest context can be very
metrics such as tree density, biomass may reduce such uncertainty. effective to take management incentives in forest ecosystem for main­
Although sample based area estimation using probability sampling taining biodiversity and ecosystem services. Forest cover changes from
approach reporting higher accuracy, but for few drivers such as road has satellite images widely explored by many researchers, but this is first
less accuracy (Table 3) due to less availability in the sample population. study in the dry deciduous forests using stratified probability sampling
Sampling for each individual driver may help to quantify accurately approach to estimate gross forest loss by forest disturbance drivers,
which drivers are very rare in the study region. Also, future research is which is provide critical information to produce unbiased area estima­
required on the spatial quantification and mapping of forest loss by tion. It is evident that different methods can provide contrasting results,
disturbance drivers and spatial quantification of ecosystem services but integrating local people’s experiences can minimize the gaps be­
including regulating, cultural and supporting services to raising tween satellite data and ground truth, and provide truthful impacts of
awareness among local people. Spatial information help policy maker forest cover loss on provisioning ecosystem services. So, this integrated
and governments to make efficient decision immediately to protection method can be effective in a particular data poor region, where limited
and reforestation and afforestation use of limited funds (Babbar et al., historical information on forest management exists. It is very important
2021) for sustainable development (Peng et al., 2021). to study forest loss by disturbance drivers in the different forest cover
Online tools such as Google Earth Engine recently increase attention densities to forest management via understanding the trend and pattern
for land use and land cover mapping, change detection analysis, and of forest loss. One can take action to tackle factors that drive forest loss
natural resource management (Gorelick et al., 2017; Sidhu et al., 2018; over time and space. Local people’s perceptions about impacts on pro­
Tsai et al., 2018). This API also provides faster and efficient cloud visioning ecosystem services can be beneficial to manage biodiversity
computing environment using open sources satellite imagery (Gorelick and ecosystem services by knowing demand side behaviour about
et al., 2017; Kumar and Mutanga, 2018) and provide an opportunity to ecosystem services. Because understanding the scale of demand is key to
use geometrically and radiometrically corrected large volume satellite managing ecosystem services effectively, for example by developing
images (Praveen et al., 2019). Further improvement of GEE platform scale-appropriate management incentives (Raudsepp-Hearne and

15
D. Bera et al. Journal of Cleaner Production 356 (2022) 131763

Peterson, 2016). The management of any forest ecosystem in terms of References


forest cover with maintaining biodiversity and ecosystem services, re­
quires some general information such as (i) what is the trend and pattern Ahammad, R., Stacey, N., Eddy, I.M.S., Tomscha, S.A., Sunderland, T.C.H., 2019. Recent
trends of forest cover change and ecosystem services in eastern upland region of
of forest cover loss (ii) what are the main drivers of forest loss and their Bangladesh. Sci. Total Environ. 647, 379–389. https://doi.org/10.1016/j.
relative contribution in different forest cover densities over time (iii) scitotenv.2018.07.406.
how forest loss and provisioning ecosystem services are related. This Alfonso, A., Zorondo-Rodríguez, F., Simonetti, J.A., 2016. Perceived changes in
environmental degradation and loss of ecosystem services, and their implications in
study depicts the trend and pattern of forest loss in different human well-being. Int. J. Sustain. Dev. World Ecol. 24 (6), 561–574. https://doi.
pre-disturbance forest densities by forest disturbance drivers from 2006 org/10.1080/13504509.2016.1255674.
to 2020. This study provides details quantitative information about large Anchang, J.Y., Prihodko, L., Ji, W., Kumar, S.S., Ross, C.W., Yu, Q., Lind, B., Sarr, M.A.,
Diouf, A.A., Hanan, N.P., 2020. Toward operational mapping of woody canopy cover
and small scale clearing with the description of clearing type (rotational in tropical savannas using Google earth engine. Front. Environ. Sci. 8, 4. https://doi.
or semipermanent) and clearing process (mechanized or org/10.3389/fenvs.2020.00004.
non-mechanized). It also depicts the impacts of forest loss on provi­ Babbar, D., Areendran, G., Sahana, M., Sarma, K., Raj, K., Sivadas, A., 2021. Assessment
and prediction of carbon sequestration using Markov chain and InVEST model in
sioning ecosystem services vice versa. Therefore, the above results are
Sariska Tiger Reserve, India. J. Clean. Prod. 278, 123333 https://doi.org/10.1016/j.
very meaningful and convincing and this will helps us to take protection jclepro.2020.123333.
measure for forest ecosystem via providing guidance to governments, Balthazar, V., Vanacker, V., Molina, A., Lambin, E.F., 2015. Impacts of forest cover
NGOs, stakeholders. Such information is useful in supporting their de­ change on ecosystem services in high Andean mountains. Ecol. Indicat. 48, 63–75.
https://doi.org/10.1016/j.ecolind.2014.07.043.
cision for sustainable forest management. Various studies show that Beck, H.E., Zimmermann, N.E., McVicar, T.R., Vergopolan, N., Berg, A., Wood, E.F.,
spatial-temporal forest cover change, no study shows the contribution of 2018. Present and future Köppen-Geiger climate classification maps at 1-km
each driver to forest loss and what is the trend of disturbance drivers resolution. Sci. Data 5 (1), 180214. https://doi.org/10.1038/sdata.2018.214.
Bera, D., Chatterjee, N.D., Bera, S., 2021. Comparative performance of linear regression,
over time in different forest cover densities and how forest loss impacts polynomial regression and generalized additive model for canopy cover estimation
on provisioning ecosystem services. Our study fills this gap of studies in in the dry deciduous forest of West Bengal. Remote Sens. Appl.: Soc. Environ. 22,
dry deciduous forests. Although a few limitations are there, but these 100502 https://doi.org/10.1016/j.rsase.2021.100502.
Bey, A., Sánchez-Paus Díaz, A., Maniatis, D., Marchi, G., Mollicone, D., Ricci, S.,
results will help to decision makers to make decisions for long term Bastin, J.-F., Moore, R., Federici, S., Rezende, M., Patriarca, C., Turia, R.,
protection and management of forest ecosystem in this area. Gamoga, G., Abe, H., Kaidong, E., Miceli, G., 2016. Collect earth: land use and land
cover assessment through augmented visual interpretation. Rem. Sens. 8 (10), 807.
https://doi.org/10.3390/rs8100807.
CRediT authorship contribution statement Broich, M., Stehman, S.V., Hansen, M.C., Potapov, P., Shimabukuro, Y.E., 2009.
A comparison of sampling designs for estimating deforestation from Landsat
Dipankar Bera: Conceptualization, Methodology, Writing – original imagery: a case study of the Brazilian Legal Amazon. Remote. Sense. Environ. 113
(11), 2448–2454. https://doi.org/10.1016/j.rse.2009.07.011.
draft, Formal analysis, Software, Visualization. Nilanjana Das Chat­
Brown, D., Schreckenberg, K., 1998. Shifting Cultivators as Agents of Deforestation:
terjee: Supervision, Writing – review & editing, Investigation, Valida­ Assessing the Evidence. Overseas Development Institute, London.
tion. Subrata Ghosh: Interpretation, Manuscript organization, Climate and Weather Averages in West Bengal, India, 1998. District wise annual weather
Investigation, Writing – review & editing. Santanu Dinda: Resources, average for West Bengal, India. https://www.timeanddate.com/weather/?
query=west+bengal. (Accessed 16 June 2020).
Investigation, Writing – review & editing. Sudip Bera: Data curation, Curtis, P.G., Slay, C.M., Harris, N.L., Tyukavina, A., Hansen, M.C., 2018. Classifying
Resources, Writing – review & editing. Mrinmay Mandal: Writing – drivers of global forest loss. Science 361 (6407), 1108–1111. https://doi.org/
review & editing. 10.1126/science.aau3445.
Dasgupta, P.S., 1995. Population, poverty and the local environment. Sci. Am. 272 (2),
40–45.
Declaration of competing interest DeFries, R.S., Rudel, T., Uriarte, M., Hansen, M., 2010. Deforestation driven by urban
population growth and agricultural trade in the twenty-first century. Nat. Geosci. 3
(3), 178–181. https://doi.org/10.1038/ngeo756.
The authors declare that they have no known competing financial Deng, L., Kim, D.-G., Li, M., Huang, C., Liu, Q., Cheng, M., Shangguan, Z., Peng, C., 2019.
interests or personal relationships that could have appeared to influence Land-use changes driven by ‘Grain for Green’ program reduced carbon loss induced
the work reported in this paper. by soil erosion on the Loess Plateau of China. Global Planet. Change 177, 101–115.
https://doi.org/10.1016/j.gloplacha.2019.03.017.
Dinda, S., Ghosh, S., Chatterjee, N.D., 2020. Understanding the commercialization
Acknowledgements patterns of non-timber forest products and their contribution to the enhancement of
tribal livelihoods: an empirical study from Paschim Medinipur district, India. Small-
scale Forestry 19 (3), 371–397.
Dipankar Bera sincere gratitude to the Council of Scientific and In­
Dolisca, F., McDaniel, J.M., Teeter, L.D., Jolly, C.M., 2007. Land tenure, population
dustrial Research (CSIR), Government of India, for financial support in pressure, and deforestation in Haiti: the case of Forêt des Pins Reserve. J. For. Econ.
the form of Senior Research Fellowship (Sanctioned file no. is 09/599/ 13 (4), 277–289. https://doi.org/10.1016/j.jfe.2007.02.006.
(0083)/2019-EMR-I). Sudip Bera, Santanu Dinda and Subrata Ghosh Dou, H., Li, X., Li, S., Dang, D., Li, X., Lyu, X., Li, M., Liu, S., 2020. Mapping ecosystem
services bundles for analyzing spatial trade-offs in inner Mongolia, China. J. Clean.
thankfully acknowledges University Grant Commission (UGC), Gov­ Prod. 256, 120444 https://doi.org/10.1016/j.jclepro.2020.120444.
ernment of India, for providing Senior Research Fellowship (SRF). We Dutta, S., Dutta, I., Das, A., Guchhait, S.K., 2020. Quantification and mapping of
also thank to villagers and local government for incorporate with us fragmented forest landscape in dry deciduous forest of Burdwan Forest Division,
West Bengal, India. Trees, Forests and People 2, 100012. https://doi.org/10.1016/j.
during the field survey. The authors would like to thanks Mantu Paira, tfp.2020.100012.
Jayati Karmakar, Tuhina Maity and Swapan sau for valuable guidance Ehara, M., Hyakumura, K., Nomura, H., Matsuura, T., Sokh, H., Leng, C., 2016.
and support in the field survey. We deeply thank and appreciate all my Identifying characteristics of households affected by deforestation in their fuelwood
and non-timber forest product collections: case study in Kampong Thom Province,
colleagues in our lab for their continuous help, support and encour­ Cambodia. Land Use Pol. 52, 92–102. https://doi.org/10.1016/j.
agement from day one of the start of this research work. We would also landusepol.2015.12.006.
like to thank the anonymous reviewers for their helpful comments and Ellison, D., Morris, C.E., Locatelli, B., Sheil, D., Cohen, J., Murdiyarso, D., Gutierrez, V.,
Noordwijk, M. van, Creed, I.F., Pokorny, J., Gaveau, D., Spracklen, D.V., Tobella, A.
suggestions. B., Ilstedt, U., Teuling, A.J., Gebrehiwot, S.G., Sands, D.C., Muys, B., Verbist, B.,
Sullivan, C.A., 2017. Trees, forests and water: cool insights for a hot world. Global
Appendix A. Supplementary data Environ. Change 43, 51–61. https://doi.org/10.1016/j.gloenvcha.2017.01.002.
Fang, L., Wang, L., Chen, W., Sun, J., Cao, Q., Wang, S., Wang, L., 2021. Identifying the
impacts of natural and human factors on ecosystem service in the Yangtze and
Supplementary data to this article can be found online at https://doi. Yellow River Basins. J. Clean. Prod. 314, 127995 https://doi.org/10.1016/j.
org/10.1016/j.jclepro.2022.131763. jclepro.2021.127995.
FAO, 2015. Global Forest Resources Assessment (2015): How Are the World’s Forests
Changing? Food and Agriculture Organization of the United Nations, Rome.

16
D. Bera et al. Journal of Cleaner Production 356 (2022) 131763

Fearnside, P.M., 2015. Highway construction as a force in the destruction of the amazon Mayes, M.T., Mustard, J.F., Melillo, J.M., 2015. Forest cover change in Miombo
forest. In: Handbook of Road Ecology. John Wiley & Sons, Ltd, pp. 414–424. https:// Woodlands: modeling land cover of African dry tropical forests with linear spectral
doi.org/10.1002/9781118568170.ch51. mixture analysis. Remote. Sense. Environ. 165, 203–215. https://doi.org/10.1016/j.
Feng, Q., Zhao, W., Fu, B., Ding, J., Wang, S., 2017. Ecosystem service trade-offs and rse.2015.05.006.
their influencing factors: a case study in the Loess Plateau of China. Sci. Total Millennium Ecosystem Assessment, 2005. Ecosystems and Human Well-Being: Synthesis.
Environ. 607–608, 1250–1263. https://doi.org/10.1016/j.scitotenv.2017.07.079. Island Press, Washington, DC.
Fisher, J.A., Patenaude, G., Meir, P., Nightingale, A.J., Rounsevell, M.D.A., Williams, M., Ngom, D., Charahabil, M.M., Sarr, O., Bakhoum, A., Akpo, L.E., 2014. Perceptions
Woodhouse, I.H., 2013. Strengthening conceptual foundations: analysing communautaires sur les services écosystémiques d’approvisionnement fournis par le
frameworks for ecosystem services and poverty alleviation research. Global Environ. peuplement ligneux de la Réserve de Biosphère du Ferlo (Sénégal). VertigO - la revue
Change 23 (5), 1098–1111. https://doi.org/10.1016/j.gloenvcha.2013.04.002. électronique en sciences de l’environnement ume 14 (2), 14. https://doi.org/
Fisher, R., Hirsch, P., 2008. Poverty and agrarian-forest interactions in Thailand. Geogr. 10.4000/vertigo.15188. Numéro 2.
Res. 46 (1), 74–84. https://doi.org/10.1111/j.1745-5871.2007.00493.x. Nikolakis, William, Innes, J., 2014. Forests and Globalization: Challenges and
Fortin, J.A., Cardille, J.A., Perez, E., 2020. Multi-sensor detection of forest-cover change Opportunities for Sustainable Development. Routledge, London.
across 45 years in Mato Grosso, Brazil. Remote. Sense. Environ. 238, 111266 https:// Olofsson, P., Foody, G.M., Herold, M., Stehman, S.V., Woodcock, C.E., Wulder, M.A.,
doi.org/10.1016/j.rse.2019.111266. 2014. Good practices for estimating area and assessing accuracy of land change.
FSI, 2019. India State of Forest Report (2019): Forest Survey of India (Ministry of Remote. Sense. Environ. 148, 42–57. https://doi.org/10.1016/j.rse.2014.02.015.
Environment Forest and Climate Change), Dehradun, India. Ongugo, P.O., Langat, D., Oeba, V.O., Kimondo, J.M., Owuor, B., Njuguna, J., Russell, A.
Gamfeldt, L., Snäll, T., Bagchi, R., Jonsson, M., Gustafsson, L., Kjellander, P., Ruiz- J., 2014. A Review of Kenya’s National Policies Relevant to Climate Change
Jaen, M.C., Fröberg, M., Stendahl, J., Philipson, C.D., Mikusiński, G., Andersson, E., Adaptation and Mitigation: Insights from Mount Elgon. Center for International
Westerlund, B., Andrén, H., Moberg, F., Moen, J., Bengtsson, J., 2013. Higher levels Forestry Research (CIFOR). https://doi.org/10.17528/cifor/005332.
of multiple ecosystem services are found in forests with more tree species. Nat. Pan, Y., Birdsey, R.A., Fang, J., Houghton, R., Kauppi, P.E., Kurz, W.A., Phillips, O.L.,
Commun. 4 (1), 1340. https://doi.org/10.1038/ncomms2328. Shvidenko, A., Lewis, S.L., Canadell, J.G., Ciais, P., Jackson, R.B., Pacala, S.W.,
Geist, H.J., Lambin, E.F., 2002. Proximate causes and underlying driving forces of McGuire, A.D., Piao, S., Rautiainen, A., Sitch, S., Hayes, D., 2011. A large and
tropical deforestation. Bioscience 52 (2), 143. https://doi.org/10.1641/0006-3568 persistent carbon sink in the world’s forests. Science 333 (6045), 988–993. https://
(2002)052[0143: PCAUDF] 2.0.CO; 2. doi.org/10.1126/science.1201609.
Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S., Thau, D., Moore, R., 2017. Pattnaik, B.K., Dutta, S., 1997. JFM in south-West Bengal: a study in participatory
Google earth engine: planetary-scale geospatial analysis for everyone. Remote. development. Econ. Polit. Wkly. 32 (50), 3225–3232.
Sense. Environ. 202, 18–27. https://doi.org/10.1016/j.rse.2017.06.031. Peng, K., Jiang, W., Ling, Z., Hou, P., Deng, Y., 2021. Evaluating the potential impacts of
Gouwakinnou, G.N., Biaou, S., Vodouhe, F.G., Tovihessi, M.S., Awessou, B.K., Biaou, H.S. land use changes on ecosystem service value under multiple scenarios in support of
S., 2019. Local perceptions and factors determining ecosystem services identification SDG reporting: a case study of the Wuhan urban agglomeration. J. Clean. Prod. 307,
around two forest reserves in Northern Benin. J. Ethnobiol. Ethnomed. 15 (1), 61. 127321 https://doi.org/10.1016/j.jclepro.2021.127321.
https://doi.org/10.1186/s13002-019-0343-y. Potapov, P.V., Turubanova, S.A., Hansen, M.C., Adusei, B., Broich, M., Altstatt, A.,
Gray, C.L., Bozigar, M., Bilsborrow, R.E., 2015. Declining use of wild resources by Mane, L., Justice, C.O., 2012. Quantifying forest cover loss in Democratic Republic of
indigenous peoples of the Ecuadorian Amazon. Biol. Conserv. 182, 270–277. https:// the Congo, 2000–2010, with Landsat ETM+ data. Remote. Sense. Environ. 122,
doi.org/10.1016/j.biocon.2014.12.022. 106–116. https://doi.org/10.1016/j.rse.2011.08.027.
Guha, A., Pradhan, A., Mondal, K., 2000. Joint forest management in West Bengal: a long Praveen, B., Mustak, S., Sharma, P., 2019. Assessing the transferability OF machine
way to go. J. Hum. Ecol. 11 (6), 471–476. https://doi.org/10.1080/ learning algorithms using cloud computing and earth observation datasets for
09709274.2000.11910423. agricultural land use/cover mapping. ISPRS Int. Archieve. Photogrammet. Remote.
Hansen, M.C., Potapov, P.V., Moore, R., Hancher, M., Turubanova, S.A., Tyukavina, A., Sense. Spatial. Info. Sci. XLII-3/W6, 585–592. https://doi.org/10.5194/isprs-
Thau, D., Stehman, S.V., Goetz, S.J., Loveland, T.R., Kommareddy, A., Egorov, A., archives-XLII-3-W6-585-2019.
Chini, L., Justice, C.O., Townshend, J.R.G., 2013. High-resolution global maps of Puyravaud, J.-P., Davidar, P., Laurance, W.F., 2010. Cryptic destruction of India’s native
21st-century forest cover change. Science 342 (6160), 850–853. https://doi.org/ forests. Conserv. Lett. 3 (6), 390–394. https://doi.org/10.1111/j.1755-
10.1126/science.1244693. 263X.2010.00141.x.
Hosonuma, N., Herold, M., De Sy, V., De Fries, R.S., Brockhaus, M., Verchot, L., Raudsepp-Hearne, C., Peterson, G., 2016. Scale and ecosystem services: how do
Angelsen, A., Romijn, E., 2012. An assessment of deforestation and forest observation, management, and analysis shift with scale—lessons from Québec. Ecol.
degradation drivers in developing countries. Environ. Res. Lett. 7 (4), 044009 Soc. 21 (3) https://doi.org/10.5751/ES-08605-210316.
https://doi.org/10.1088/1748-9326/7/4/044009. Rouse, J.W., Haas, J., R, Deering, D., Schell, J., Harlan, J., 1974. Monitoring the Vernal
Houghton, R.A., 2012. Carbon emissions and the drivers of deforestation and forest Advancement and Retrogradation (Green Wave Effect) of Natural Vegetation.
degradation in the tropics. Curr. Opin. Environ. Sustain. 4 (6), 597–603. https://doi. Technical Report. NASA/GSFC Type III Final Report, Greenbelt, MD, USA.
org/10.1016/j.cosust.2012.06.006. Saah, D., Johnson, G., Ashmall, B., Tondapu, G., Tenneson, K., Patterson, M.,
Kanninen, M., Murdiyarso, D., Seymour, F., Angelsen, A., Wunder, S., 2007. Do Trees Poortinga, A., Markert, K., Quyen, N.H., San Aung, K., Schlichting, L., Matin, M.,
Grow on Money? the Implications of Deforestation Research for Policies to Promote. Uddin, K., Aryal, R.R., Dilger, J., Lee Ellenburg, W., Flores-Anderson, A.I., Wiell, D.,
Center for International Forestry Research, Cifor, Bogor. Lindquist, E., Chishtie, F., 2019. Collect Earth: an online tool for systematic reference
Kim, D.-H., Sexton, J.O., Noojipady, P., Huang, C., Anand, A., Channan, S., Feng, M., data collection in land cover and use applications. Environ. Model. Software 118,
Townshend, J.R., 2014. Global, Landsat-based forest-cover change from 1990 to 166–171. https://doi.org/10.1016/j.envsoft.2019.05.004.
2000. Remote. Sense. Environ. 155, 178–193. https://doi.org/10.1016/j. Saraiva, M.B., Ferreira, M.D.P., da Cunha, D.A., Daniel, L.P., Homma, A.K.O., Pires, G.F.,
rse.2014.08.017. 2020. Forest regeneration in the Brazilian Amazon: public policies and economic
Köhl, M., Lasco, R., Cifuentes, M., Jonsson, Ö., Korhonen, K.T., Mundhenk, P., de Jesus conditions. J. Clean. Prod. 269, 122424 https://doi.org/10.1016/j.
Navar, J., Stinson, G., 2015. Changes in forest production, biomass and carbon: jclepro.2020.122424.
results from the 2015 UN FAO global forest resource assessment. For. Ecol. Manag. Senf, C., Laštovička, J., Okujeni, A., Heurich, M., van der Linden, S., 2020. A generalized
352, 21–34. https://doi.org/10.1016/j.foreco.2015.05.036. regression-based unmixing model for mapping forest cover fractions throughout
Kumar, L., Mutanga, O., 2018. Google earth engine applications since inception: usage, three decades of Landsat data. Remote. Sense. Environ. 240, 111691 https://doi.
trends, and potential. Rem. Sens. 10 (10), 1509. https://doi.org/10.3390/ org/10.1016/j.rse.2020.111691.
rs10101509. Shah, A., Khan, S., Shah, M.H., Khan, R., Jan, I., 2010. Environmental impact assessment
Lambin, E.F., Geist, H.J., Lepers, E., 2003. Dynamics of land-use and land-cover change (EIA) of infrastructure development projects in developing countries (SSRN scholarly
in tropical regions. Annu. Rev. Environ. Resour. 28 (1), 205–241. https://doi.org/ paper ID 1666339). Soc. Sci. Res. Network. https://papers.ssrn.com/abstract
10.1146/annurev.energy.28.050302.105459. =1666339.
Laurance, W.F., Sayer, J., Cassman, K.G., 2014. Agricultural expansion and its impacts on Shriar, A.J., 2014. Theory and context in analyzing livelihoods, land use, and land cover:
tropical nature. Trends Ecol. Evol. 29 (2), 107–116. https://doi.org/10.1016/j. lessons from Petén, Guatemala. Geoforum 55, 152–163. https://doi.org/10.1016/j.
tree.2013.12.001. geoforum.2014.06.002.
Le Quéré, C., Raupach, M.R., Canadell, J.G., Marland, G., Bopp, L., Ciais, P., Conway, T. Sidhu, N., Pebesma, E., Câmara, G., 2018. Using Google Earth Engine to detect land
J., Doney, S.C., Feely, R.A., Foster, P., Friedlingstein, P., Gurney, K., Houghton, R.A., cover change: Singapore as a use case. Euro J. Remote. Sense 51 (1), 486–500.
House, J.I., Huntingford, C., Levy, P.E., Lomas, M.R., Majkut, J., Metzl, N., https://doi.org/10.1080/22797254.2018.1451782.
Woodward, F.I., 2009. Trends in the sources and sinks of carbon dioxide. Nat. Sloan, S., Sayer, J.A., 2015. Forest Resources Assessment of 2015 shows positive global
Geosci. 2 (12), 831–836. https://doi.org/10.1038/ngeo689. trends but forest loss and degradation persist in poor tropical countries. For. Ecol.
Liu, X., Ziv, G., Bakshi, B.R., 2018. Ecosystem services in life cycle assessment - Part 1: a Manag. 352, 134–145. https://doi.org/10.1016/j.foreco.2015.06.013.
computational framework. J. Clean. Prod. 197, 314–322. https://doi.org/10.1016/j. State statistical handbook, 2015. Department of Planning & Statistics of West Bengal,
jclepro.2018.06.164. India.
Liu, W., Guo, Z., Lu, F., Wang, X., Zhang, M., Liu, B., Wei, Y., Cui, L., Luo, Y., Zhang, L., Stehman, S.V., 2014. Estimating area and map accuracy for stratified random sampling
Ouyang, Z., Yuan, Y., 2020. The influence of disturbance and conservation when the strata are different from the map classes. Int. J. Rem. Sens. 35 (13),
management on the greenhouse gas budgets of China’s forests. J. Clean. Prod. 261, 4923–4939. https://doi.org/10.1080/01431161.2014.930207.
121000 https://doi.org/10.1016/j.jclepro.2020.121000. Sudhakar Reddy, C., Jha, C.S., Dadhwal, V.K., Hari Krishna, P., Vazeed Pasha, S.,
Lu, S., Sun, H., Zhou, Y., Qin, F., Guan, X., 2020. Examining the impact of forestry policy Satish, K.V., Dutta, K., Saranya, K.R.L., Rakesh, F., Rajashekar, G., Diwakar, P.G.,
on poor and non-poor farmers’ income and production input in collective forest 2016. Quantification and monitoring of deforestation in India over eight decades
areas in China. J. Clean. Prod. 276, 123784 https://doi.org/10.1016/j. (1930–2013). Biodivers. Conserv. 25 (1), 93–116. https://doi.org/10.1007/s10531-
jclepro.2020.123784. 015-1033-2.

17
D. Bera et al. Journal of Cleaner Production 356 (2022) 131763

Sulieman, H.M., 2018. Exploring drivers of forest degradation and fragmentation in increasing smallholder clearing. Sci. Adv. 4 (11), eaat2993 https://doi.org/10.1126/
Sudan: the case of erawashda forest and its surrounding community. Sci. Total sciadv.aat2993.
Environ. 621, 895–904. https://doi.org/10.1016/j.scitotenv.2017.11.210. Van Jaarsveld, A. s, Biggs, R., Scholes, R. j, Bohensky, E., Reyers, B., Lynam, T.,
Tacconi, L., Rodrigues, R.J., Maryudi, A., 2019. Law enforcement and deforestation: Musvoto, C., Fabricius, C., 2005. Measuring conditions and trends in ecosystem
lessons for Indonesia from Brazil. For. Pol. Econ. 108, 101943 https://doi.org/ services at multiple scales: the Southern African Millennium Ecosystem Assessment
10.1016/j.forpol.2019.05.029. (SAfMA) experience. Phil. Trans. Biol. Sci. 360 (1454), 425–441. https://doi.org/
Tewari, V.P., 2016. Forest inventory, assessment, and monitoring, and long-term forest 10.1098/rstb.2004.1594.
observational studies, with special reference to India. For. Sci. Technol. 12 (1), Vermote, E., Justice, C., Claverie, M., Franch, B., 2016. Preliminary analysis of the
24–32. https://doi.org/10.1080/21580103.2015.1018962. performance of the Landsat 8/OLI land surface reflectance product. Remote. Sense.
Thanichanon, P., Schmidt-Vogt, D., Messerli, P., Heinimann, A., Epprecht, M., 2013. Environ. 185, 46–56. https://doi.org/10.1016/j.rse.2016.04.008.
Secondary forests and local livelihoods along a gradient of accessibility: a case study Walters, B.B., 2012. Do property rights matter for conservation? Family land, forests and
in northern Laos. Soc. Nat. Resour. 26 (11), 1283–1299. https://doi.org/10.1080/ trees in Saint Lucia, west indies. Hum. Ecol. 40 (6), 863–878. https://doi.org/
08941920.2013.788429. 10.1007/s10745-012-9541-0.
Tsai, Y., Stow, D., Chen, H., Lewison, R., An, L., Shi, L., 2018. Mapping vegetation and WBFD, 2018. Annual Administrative Report 2016-2017-West Bengal. Forest Department
land use types in fanjingshan national nature reserve using Google earth engine. of the Government of West Bengal, India.
Rem. Sens. 10 (6), 927. https://doi.org/10.3390/rs10060927. Williams, B.K., 2011. Adaptive management of natural resources—framework and issues.
Tyukavina, A., Hansen, M.C., Potapov, P.V., Stehman, S.V., Smith-Rodriguez, K., J. Environ. Manag. 92 (5), 1346–1353. https://doi.org/10.1016/j.
Okpa, C., Aguilar, R., 2017. Types and rates of forest disturbance in Brazilian Legal jenvman.2010.10.041.
Amazon, 2000–2013. Sci. Adv. 3 (4), e1601047 https://doi.org/10.1126/ Yang, H., Harrison, R., Yi, Z.-F., Goodale, E., Zhao, M.-X., Xu, J.-C., 2015. Changing
sciadv.1601047. perceptions of forest value and attitudes toward management of a recently
Tyukavina, A., Hansen, M.C., Potapov, P., Parker, D., Okpa, C., Stehman, S.V., established nature reserve: a case study in southwest China. Forests 6 (12),
Kommareddy, I., Turubanova, S., 2018. Congo Basin forest loss dominated by 3136–3164. https://doi.org/10.3390/f6093136.

18

You might also like