Persistent loss of biologically-rich tropical forests in the Indian Eastern Himalaya

Deforestation is a major cause of biodiversity loss in Asia. India’s biologically-diverse state of Arunachal Pradesh has been undergoing forest loss due to multiple drivers. We assessed the change in forest cover in a state-managed Reserved Forest adjoining an important Protected Area (PA), i.e. the Pakke Tiger Reserve using satellite imagery at a fine spatial resolution. A conservation program to protect three species of endangered hornbills and their nesting habitat outside the PA had been set up in 2011-12. We assessed the effectiveness of the conservation programme in protecting forests. We report a loss of 32 km2 of forest cover between 2013 and 2017 with a 5% decline in total forest area in four years. In the habitat around the 29 hornbill nest trees we estimated a loss of 35% of forest cover. This loss occurred despite varied efforts through the conservation program and by individuals in the community/government. We identify illegal logging (despite a ban by the Supreme Court of India) as the main driver that is depleting forest cover within this important area. Our results highlight the ongoing threats to biologically-rich forests and the need for urgent measures to halt this loss. We suggest that this study has general practical implications for the governance of non-PA state-managed forests in Arunachal Pradesh. The ongoing deforestation appears to be due to organized crime, institutional inadequacy from a combination of limited resources, bureaucratic apathy, and/or ambiguity in use and ownership of forest land compared to other community forests which appear to have robust governance systems.


3 8
In this paper, we aimed to assess the extent of forest loss and the effectiveness of the Hornbill Nest Adoption Program 1 3 9 in protecting hornbill habitat. Our specific purpose is to 1) estimate forest loss in the Papum RF using satellite data at a  Act. In Reserved Forests, all extractive activities are prohibited unless legally permitted (Indian Forest Act 1927).

5 0
Part of Papum RF (346.25 km 2 ) is included in the buffer area of Pakke TR as per the National Tiger Conservation

5 6
With an elevational range from 200 to 1500 m above sea level, Papum RF receives an average total annual rainfall of 1 5 7 2500 mm. Mean (± standard deviation) maximum temperature is 29.3°C (± 4.2) and the minimum temperature is species, several species are quite rare and natural regeneration is low.

6 7
The total area of Papum RF is 1064 km 2 , however for this study, we marked out an area of 737 km 2 for classifying the 1 6 8 forest and analysis of change in forest cover (Fig. 1). We restrict our analyses to 70% of the total area for two reasons:

7 3
RapidEye and PlanetScope satellite data processing and classification 1 7 4 We obtained ortho-rectified surface/top-of-atmosphere (TOA) reflectance data imaged by either the RapidEye (5 m 1 7 5 spatial resolution) or PlanetScope (3 m spatial resolution) constellations, to ensure a complete cloud-free coverage of 1 7 6 the area (for a list of images acquired refer to Supplementary

8 3
Ortho-rectification of satellite images is a process of terrain correction in a region with irregular topography. Ortho-1 8 4 rectification is applied to ensure the same geographical region is analyzed year-to-year within a region of interest (ROI) 1 8 5 (Tucker et al. 2004). We used images that were corrected to surface reflectance or TOA reflectance since a year's image

9 3
Using a combination of field sampling (using a global positioning unit) and Google Earth imagery, ROIs were 1 9 4 identified for three land-cover classes. Each scene (or partial scene) was independently classified as stable forest, stable 1 9 5 non-forest and logged-forest using the randomForest library 4.6-14 in the R (R version 3.3., R Core Development Team 1 9 6 2016). Stable forest regions comprised ROIs of uncut closed canopy forests with little or no detectable anthropogenic 1 9 7 disturbance. Stable non-forest regions comprised water bodies, grasslands, permanent settlements, sand bars and 1 9 8 landslides. Logged-forest ROIs were defined using ground reports of active/past logging, studying satellite images at 1 9 9 GFW deforestation hot-spots, and for roads, new clearings, plantations and fire scars. Logged-forest ROIs generally 2 0 0 comprise areas previously under forest but currently with higher albedo than forest. The shape of the clearings is often 2 0 1 geometrical and close to older forest clearings. Roads are linear in shape with the lower slope scarred with discarded 2 0 2 debris. The training datasets of the above three classes consisted of at least 40 ROIs and ~29 million pixels, per year.

0 3
Land cover change around hornbill nest trees 2 0 4 The HNAP is confined to the lower and south-western parts of Papum RF (

0 7
To investigate if the habitat around the monitored and protected hornbill nest trees were affected by forest loss, scenes 2 0 8 that covered >90% of the hornbill nest sites were chosen. In the latter case, cloud-free, single day scenes were available 2 0 9 and could be analysed from 2011 to 2019. This allowed us to make comprehensive fine-scale forest loss estimations for 2 1 0 9 years. Cloud-free satellite images for all years were from November-December, except for 2018 and 2019 which 2 1 1 were from April-May) (dry season). During the dry season, secondary vegetation in clear felled areas is visibly 2 1 2 dissimilar from primary forest. While we do not test for this difference, we think the visible difference may be 2 1 3 attributed to the drying and browning of vegetation in the summer season when soil moisture and rainfall are low.

1 4
Secondary vegetation in winter months (post-monsoon) are visibly greener as the soil moisture is still high. An 2 1 5 identical approach (to that used for classifying forest loss in Papum RF) was implemented to classify the area around 2 1 6 29 hornbill nests. A 1-km buffer was created and the satellite scenes were clipped to the buffered extent (48 km 2 ).

1 9
The spatial accuracy of the land-cover classification was assessed by manual checking of the scenes coupled with a  Table 4 and Table 5).

8
Post land-cover classification, we calculated the annual rate of forest area loss using a modified compound-interest-rate 2 2 9 formula for its mathematical clarity and biological relevance (Puyravaud 2003): where A 1 and A 2 is the forest area in time periods t 1 and t 2 , respectively. P is the annual percentage of area lost. There was very high forest loss in Papum RF as determined from analysis at a fine-scale resolution. Table 1  km 2 ). Out of a total area of 737 km 2 classified, 156 km 2 was logged-forest by 2017.

3 8
Our analyses recorded forest loss to be lower in 2017 than in 2014, for two reasons: (1) an area (~5 km 2 ) in the eastern

4 6
However, by 2019, only 45% of the 48 km 2 of the 1-km buffer area around 29 hornbill nests was forested as compared 2 4 7 to 80% in 2011 (Table 2). Forest loss is also evident from the construction of roads, burn scars and clear-cut felling of  (Table 2). In the last 9 years, there has been a total loss of 16.61 km 2 2 5 1 in a 1 km buffer around the 29 nest sites (Fig. 2b). Annual rate of forest area loss around the nest trees was 7% year -1 , 2 5 2 corresponding to 2.07 km 2 year -1 .

5 4
The forest loss has serious consequences for tropical biodiversity, as the destruction of suitable habitat threatens the  the Seijosa area was mainly for household needs and subsistence use by people.

7 0
On-ground observations/media reports show that tree felling increased after 2015 and coincided with the use of  habitat to act as a deterrent to felling.

9 0
However, the forest cover change analysis shows that the loss and degradation of the surrounding habitat and hornbill 2 9 1 food trees continued despite these protection efforts. This will likely have negative consequences for hornbill nesting 2 9 2 and persistence in the Papum RF. Tree density/basal area and food and nest tree density is considerably lower in the RF 'selective' logging after some years since logging or when the logging was officially permitted before 1996, this study 2 9 8 notes the alarming loss of forest despite the 1996 Supreme Court ban and the lack of any working plan under which the 2 9 9 current logging is occurring within Papum RF.

0 0
Hornbills are highly mobile species with large home ranges, and nesting males move from the RF to the Pakke TR to 3 0 1 forage for fruits. Our telemetry data of tagged Great and Wreathed hornbills show that some individual hornbills move 3 0 2 between the Pakke TR and the RF (Naniwadekar et al. 2019). However, despite their ability to move between these 3 0 3 areas, a continuing loss of forest cover will result in nest trees in the RF becoming inactive. As the forest is becoming 3 0 4 more degraded and is being logged it has also become more common to find only nests of the more adaptable Oriental

0 6
The tree felling occurs mainly in the drier months starting from September to March-April, but in some years, illegal 3 0 7 logging activity has continued in the wetter period. March is the beginning of the breeding season for the larger-sized 3 0 8 Great Hornbill and Wreathed hornbill when the females start entering the nest cavities, sealing them and laying eggs.

0 9
Apart from the direct loss of forest habitat and individual trees, the sound of mechanized chainsaws, movement and disturbance from illegal logging and loss of habitat, may also affect the use of roost sites by hornbills in the future.

1 5
The loss of 32 km 2 of forest over 4 years within Papum RF is a cause for concern also because the area receives heavy 3 1 6 rainfall often resulting in floods and landslides. The depletion of tropical forests in Papum RF severely threatens the 3 1 7 future subsistence needs of the local and regional population. Although we do not explicitly test for these effects of  ecosystem services will be a pragmatic goal for all privileged and underprivileged stake-holders as per several 3 3 0 sustainable development goals laid out by the United Nations.

1
Some amount of timber extraction for local house construction and subsistence needs is legitimate. However, the spurt 3 3 2 in illegal commercial logging activities on a large-scale, with timber being sold and transported out of the state, using 3 3 3 mechanized chainsaws and hired labourers from a neighbouring community, is driving an alarming loss of forest cover 3 3 4 in this area. In addition, with the construction of new roads, the continuation of these illegal activities to newer areas in 3 3 5 the higher northern parts of the RF deeper inside Arunachal Pradesh is also being facilitated and is a threat to the long-3 3 6 term status of this important forest area for both people and wildlife.

7
One of the challenges in our study was the strict classification of land-cover as non-forest and logged-forest. Our ROI 3 3 8 includes areas that often flood in the monsoon changing the percentages of these areas every year. New road 3 3 9 construction or mining in recently logged forests can be classified as non-forest, while previously cleared primary forest 3 4 0 can show regrowth as secondary vegetation. The difficult terrain in the region makes robust collection of ground-control 3 4 1 points challenging. Hence, we make the following suggestions: 1) dry summer season images are best to distinguish 3 4 2 secondary and non-woody vegetation from primary forest, 2) a binary classification system of forest and non-forest, and 3 4 3 3) forest loss estimations within a completely forested region such that loss in later years can be detected using year-to-

4 4
year image subtraction techniques. However, we hope our work is a step towards achieving accurate forest loss 3 4 5 estimates for an under-explored, mountainous region with exceptional forests and biodiversity.

4 6
The key management measures to stop the illegal logging are 1) a complete ban on the use/sale and possession of 3 4 7 mechanized chainsaws in the area. While prohibitory orders have been issued in the past by the district administration, 3 4 8 these have not been enforced, 2) stopping the unregulated movement of hired labour from the neighbouring state into external agency to ensure that illegal logging has been stopped. In the long-term, for better governance, clarity in the 3 5 5 use and ownership of forest land also needs to be addressed under the law given that some of the designated forest area 3 5 6 is under settlements and multiple use areas by people.