Forest ecosystem plays a key role in climate change reduction through their exceptional nature of carbon sequestration which regulates the global temperature (Pan et al.
, 2011). Forest covers about 30% of Earth’s land surface and stored 45% of the carbon stored on land (Saatchi et al., 2011).
The role of forest to store a large amount of carbon than any other terrestrial ecosystem serves as the best carbon bank and natural brake of climate change (Gibbs et al., 2007). According to Saatchi et al., (2011) researchers found that around 247 billion tons of carbon sequestered in tropical forests.
Among the sequestered carbon 78.14% stored in aboveground biomass (trunks, branches, and leaves) and 21.86% stored below ground in the roots. Forest ecosystem destruction releases the vast amount of carbon back to the atmosphere as carbon dioxide (CO2) (Gibbs et al., 2007; Mohren et al., 2012). The tropical rainforest is one of the forest ecosystems which can store and release a vast amount of carbon dioxide (Soares Filho et al., 2010).
Nowadays, tropical rainforest is declining due to deforestation and forest degradation (Jungle Boy, 2013). Tropical deforestation and forest degradation contribute around 20% of the highest global carbon dioxide emission next to fossil fuels (Hirata et al., 2011). Carbon dioxide is a type of Green House Gas (GHG) emitted to the atmosphere which causes climate change (Paustian et al., 2000). Additionally, the intergovernmental panel on climate change (IPCC) third assessment report (2001) mentioned that Carbon dioxide emission due to tropical forest destruction has serious impact in climate change (Houghton et al., 2001) The United Nation Framework Convention on Climate Change (UNFCCC) was established in 1992 for the purpose of greenhouse gas (GHG) emission reduction and countries agreed on the Convention to stabilize the GHG emission.
Monitoring and reporting the forest carbon emission status is needed by the Convention (Peltoniemi et al., 2006). Following the convention, Reducing Emission from Deforestation and forest Degradation (REDD+) including sustainable forest management for carbon stock enhancement has been initiated to make a follow-up on emission reduction activities (Graham et al., 2017). The REDD+ needs an accurate measurement, reporting and verification (MRV) mechanism of forest carbon stock balance monitoring to offer result based reimbursement for REDD+ adopted countries (Goetz et al., 2012).
Accurate forest biomass estimation is crucial for accurate carbon stock assessment since 47-50% of the forest biomass is carbon (Drake et al., 2002; IPCC, 2007). The assessment of aboveground biomass is a major concern to REDD+ MRV mechanism (Phua et al., 2016). The most accurate method of aboveground biomass estimation is cutting, drying and weighing all parts of the tree but, this method is not environmentally friendly (Basuki et al.
, 2009). This destructive method of biomass estimation can give an opportunity to corroborate other nondestructive methods to estimate carbon stock using allometric equation (Clark et al., 2001). The allometric equation needs forest parameters such as tree height, diameter at breast height, Crown projection area (CPA) and wood density as an input (Basuki et al., 2009; Ketterings et al., 2001; Sampaio et al.
, 2010). The use of remote sensing technology to extract forest inventory parameters is a nondestructive method to estimate aboveground biomass (Ahamed et al., 2011). Numerous studies have been carried out in estimating and measuring these forest inventory parameters using different types of remote sensing tools and techniques (Brovkina et al., 2017; Ene et al.
, 2016; Kankare et al., 2013; Liang et al., 2014; Nelson et al., 2017). Diameter at breast height and tree height have a link to biomass and allow indirect biomass estimation (Yang et al.
, 2013). In simple forest modelling diameter at breast height from LiDAR data by making a relationship with other forest parameters is feasible (Andersen et al., 2005; Drake, Dubayah, Clark, et al., 2002; M. A. Lefsky et al., 2005; Naesset et al.
, 2005). Terrestrial LiDAR is the only remote sensing technique which allows diameter at breast height measurement easily (Hopkinson et al., 2004; Pfeifer et al., 2004). Scanning forests using Terrestrial LiDAR for diameter at breast height measurement needs multiple scans in order to extract 3-D point clouds of the maximum number of individual trees and of sufficient quality. A solitary scan in the tropical dense forest using Terrestrial LiDAR leads to occlusion effect (Pfeifer et al., 2004; Yao et al.
, 2011). Various studies have been carried out in retrieving tree height using different LiDAR types at different point density (D. J. Harding et al., 2001; David J. Harding & Carabajal, 2005; M. A. Lefsky et al.
, 2005; Næsset & Økland, 2002).Terrestrial LiDAR cannot detect upper canopy of tropical rainforest as it cannot perceive long distance vertically. Mostly airborne and spaceborne LiDAR detect the forests upper canopy fruitfully but it needs canopy openings to record data from the forest floor (Magnussen & Boudewyn, 1998; Naesset, 1997). Even though researchers acknowledge the potential of LiDAR tools in forest inventory parameter estimation, in dense forest like tropical rainforest are faced with challenges due to occlusion (Coops et al.
, 2007; Lovell et al., 2003). In addition to the diameter at breast height and tree height, crown dimension provides an estimation of aboveground biomass and carbon stock since they serve as input for the allometric equation (Brown et al., 2005; Popescu et al.
, 2003). Unmanned Aerial Vehicle(UAV) also called Drone, is an exciting tool which is able to supply imagery at high spatial and temporal resolution (Turner et al., 2012). According to Turner et al., (2012) there are two kinds of UAV; fixed wing and multi-rotors UAV design. The fixed-wing UAV fly faster and can cover a large area with large overlap while multi rotors can fly slowly and can capture images with any required overlap. The multi rotors UAV needs a larger area for takeoff compared to the single rotor (Nonami et al.
, 2010). Even though different UAVs have a different performance based on payload, they cannot potentially carry any payload of the other remote sensing tools. This technology can be used in sustainable forest management issues as it assists observation, assessment and mapping forests (Remondino et al., 2011).
UAV application in sustainable forest management program is crucial because of its applicability in harsh inaccessible places. According to Grenzdörffer et al., (2008) the application of this new technology in forest management can facilitate estimation of forest inventory parameters and the achievement of aims of REDD+ strategy. There are possibilities and advantages in assessing tropical rainforest aboveground biomass and carbon stock using UAV since it has a potential to provide a high-resolution image at any moment in time if the weather allows (Messinger et al., 2016).
UAV image users have the chance to plan the flight mission time and minimize the impact of weather on image quality. The advantages of UAV in the forest carbon stock assessment are like light-weight, low-cost and cost diminution of image atmospheric correction which are essential to achieving REDD+ MRV goals (Getzin et al., 2012). UAV platform imagery can supplement expensive and labour intensive forest inventory methods (Messinger et al.
, 2016b). In spite of open space requirement for GCPs, UAV imagery has a potential to capture multi-view images and can reduce occlusion effect. Structure from motion (SfM) is a photogrammetric range imaging technique for estimating three-dimensional structures from two-dimensional image sequences. SFM method UAV image processing solves the camera pose and scene geometry automatically by matching the series 2D images (Westoby et al., 2012).
Tree height estimation using UAV image through photogrammetric image processing is potential and the estimated tree height can be used as an input to the allometric equation for biomass estimation (Magar, 2014; Reuben, 2017). The extraction of forest parameters using remote sensing can be achieved through object-based image analysis (OBIA). Object-based image analysis is a type of image analysis which has been successfully applied to crown projection area delineation in simple forest canopies using high-resolution images as an input (Hay et al.
, 2005; Kim et al., 2009). In object-based image analysis, the segments are generated using image segmentation algorithm which portions a large image into the non-overlapping unit (Chubey et al., 2006).
Object-based image segmentation uses spectral and other information such as shape, texture and contextual relationships(T. Blaschke, 2010).Therefore, this study is aiming at estimating and mapping of aboveground biomass and carbon using UAV stand-alone imagery with REDD+ MRV context in Berkelah tropical rainforest, Malaysia. 1.2. Statement of the problem and Justification The main element of the agreement reached in Paris at the UNFCCC COP21 in December 2015 for climate change mitigation was about the implementation of REDD+ (Pasgaard et al., 2016). The aim of the initiative is to encourage sustainable forest management activities and for Greenhouse gas emission reduction.
REDD+ program increase degraded primary rainforest rehabilitation(Ansell et al., 2011). Thus, sustainable management of forests in the logged-over and non-logged-over tropical rainforest can reverse the change direction of primary forest to secondary forest (Ansell et al., 2011). As a result, some logged-over forests of Malaysia have been rehabilitated. moreover, Measurement, Reporting and Verification (MRV) of carbon stock is a REDD+ mechanism to measure the status of the forest and to monitor the emission balance of REDD+ adopted countries such as Malaysia (Köhl et al., 2009). In addition, regular carbon stock assessment in tropical rainforest needs a low-cost and simple method of carbon estimation.
Accurate measurement of forest inventory parameters has a relationship with accurate aboveground biomass estimation which is a major concern of REDD+MRV program as it assists computing accurate carbon estimation (Phua et al., 2016). But, accurate measurement of these parameters in the multilayered tropical rainforest is a challenging task due to various uncertainties (D. Lu, 2005). Most of the allometric equations are developed based on forest parameters such as wood density, tree height and DBH (Bragg, 2001; Chave et al., 2014). Among those parameters, DBH explains about 95% of the variation in aboveground biomass (Brown, 2002). Tree height is a supplementary input into the biomass estimation equations (Chave et al.
, 2014). But, measuring DBH and tree height in the field is a labour-intensive and time-consuming (Brown, 2002; Kwak et al., 2007) as it is tree based and not applicable to large areas. In order to quantify the aboveground biomass timely and efficiently, remote sensing technology is vital as it reaches inaccessible and widespread areas of distance (Calders et al., 2011). Moreover, application of this technology in measuring forest parameters can be considered as an excellent technique to achieve the aim of REDD+ strategy.
But, measuring DBH from a remotely sensed data is impossible (Sium, 2015) except near distance through TLS. A plot based single scan faces the occlusion effect as the outlying stem is covered by the near stem and making multiple scans is also time-consuming (Pfeifer et al., 2004; Yao et al., 2011).
Crown Projection Area is the region of the vertical projection of the furthest boundary of the crown which can be detected using remote sensing tools and it is vital for segmentation (Gartner et al., 2010; Song et al., 2010). CPA assists to separate individual tree crowns and to make a relationship with other forest inventory parameters. It has a relationship with DBH (Hirata et al., 2009; Shimano, 1997). Numerous researchers showed a significant relationship between DBH and tree CPA (Hemery et al., 2005; Michael A Lefsky et al.
, 2002; Sium, 2015; Song et al., 2010). Therefore, making a relationship between CPA and DBH is a promising mechanism to minimize the challenge of DBH detection from a far distance and can make aboveground biomass estimation straightforward to achieve the aim of REDD+ program.
Forests canopy height extraction and mapping are mainly achieved using LiDAR and Digital Photogrammetry (Lim & Treitz, 2004). The LiDAR can provide accurate aboveground forest biomass estimation as it directly measures the canopy structure in less dense forests (Drake, Dubayah, Knox, et al., 2002). Even though, LiDAR can penetrate through canopy openings in less complex secondary forests, still cannot see the DBH which is crucial in calculating the aboveground biomass.
The crown projection area can be used as a proxy for DBH in more simple forests while in multilayer tropical rainforest; the CPA cannot be successfully extracted from LiDAR data due to occlusion.Unmanned aerial vehicle(UAV) is an elating technology which has a potential to supply high spatial and temporal resolution image (Turner et al., 2012). Photogrammetric UAV image processing is a promising cost-effective technique since it yields 3-D point cloud comparable to LiDAR data for estimating aboveground biomass (Magar, 2014). The success of LiDAR-based forest inventory parameter estimation increases with increasing the point cloud density of LiDAR data but, increasing LiDAR data point cloud density is expensive (Gibbs et al., 2007).
A 3-D point cloud obtained through a photogrammetric processing of UAV image can substitute the perspective of LiDAR data through reducing costs which incurred during data acquisition (Leberl et al., 2010). in addition, it has a potential to collect data over areas of a few hundred or few thousand hectares where the use of a LiDAR is not feasible due to employment costs (Messinger et al., 2016b). UAV low flight altitude allows operating under a cloud which avoids uncertainty due to atmospheric effects during forest inventory parameters estimation.
However, UAV flight in windy weather and dense forest has the challenge to keep image position properly. In addition the GPS in UAV image has a positional inaccuracy and needs ground control points recording and geo-referencing at time of generating the dense point cloud. Object-based image segmentation uses spectral and contextual information and has a significant contribution to individual tree crown projection area delineation on a sparse forest canopy (Baral, 2011; Karna, 2012; Tsendbazar, 2011). However, dense forest image segmentation is influenced by multilayer forest structure and canopy intermingles. Multilayer forest structure image segmentation cannot successfully delineate the incomplete tree crowns of the lower canopy. The effect of canopy structure complexity on segmentation for biomass estimation is not much studied and has not a solution yet. UAV image segmentation of temperate forest has been successfully